Remove retired benchmark material

README.md

@@ -105,36 +105,6 @@ Orukeet improves 25 of 27 complete LibriSpeech/FLEURS splits and 36 of 47 accent
 
 [All 74 paired WER/CER scores and edit counts](docs/current-checkpoint-benchmarks.md) · [Methods](docs/technical-report.md) · [Technical report](orukeet-technical-report.pdf)
 
-<!-- orukeet-open-asr-20260918:start -->
-## Open ASR public English evaluation, September 18, 2026
-
-**Submission withdrawn September 19, 2026 (UTC).** [Open ASR PR #221](https://github.com/huggingface/open_asr_leaderboard/pull/221) is closed. The leaderboard submission has been withdrawn; the checkpoint, code, research results, and exposure disclosures remain available as historical research artifacts.
-
-The released r3 checkpoint was evaluated on all eight current public English Open ASR datasets using the official NeMo H200 container. Mean WER is **4.49%**, with aggregate throughput **5545.68 RTFx**. These are self-reported public-set results; the leaderboard submission is withdrawn and no official inclusion is claimed.
-
-**Training exposure:** r3 trained on all 2,939 LibriSpeech test-other recordings for three passes and used that split for selection. Monsoon English also informed earlier training-partition selection. The aggregate includes these datasets and is not a held-out generalization estimate. [Training and benchmark exposure](https://huggingface.co/oruk/orukeet/blob/main/evaluation/open_asr_20260918/training-disclosure.md).
-
-| Dataset | WER (%) | H200 RTFx |
-|:--|--:|--:|
-| AMI-Cleaned | 8.68 | 3741.22 |
-| Earnings22-Cleaned-AA | 5.74 | 2982.17 |
-| GigaSpeech-Cleaned | 7.48 | 5346.49 |
-| LibriSpeech test-clean | 1.47 | 4180.11 |
-| LibriSpeech test-other (trained on) | 2.84 | 4159.42 |
-| SPGISpeech | 3.38 | 6419.21 |
-| Monsoon English (prior selection exposure) | 3.84 | 4835.62 |
-| VoxPopuli-Cleaned-AA | 2.46 | 2488.16 |
-| **Public-eight aggregate** | **4.49** | **5545.68** |
-
-The run uses H200, NeMo 2.7.2, Torch 2.8.0, BF16, batch size 128, and the same greedy-batch TDT decoder with `max_symbols=10` across all datasets. Timing covers the complete transcription pass after up to four warmup batches, with CUDA synchronization. RTFx is total evaluated audio divided by total measured transcription time; it excludes model/data acquisition and warmup. Batch size follows the existing Parakeet profile; exhaustive throughput optimization is not claimed.
-
-All 74,443 eligible inputs from 74,544 source records are retained. The official reference rule excludes 90 AMI and 11 GigaSpeech records; Earnings22's 341 chunks are merged into six full calls for scoring. The exact official normalizer and compound-aware `kaldialign==0.12.0` scorer are used. The mean averages the eight two-decimal dataset WERs. The earlier CPU FP32 comparison remains a separate run.
-
-[Complete metrics and execution hashes](https://huggingface.co/oruk/orukeet/blob/main/evaluation/open_asr_20260918/public-eight-h200-results.json) · [Pinned reproduction code](https://huggingface.co/spaces/oruk/open-asr-leaderboard-orukeet/tree/171ab5e79454cc3772a8700479cb150cfe4cc84f)
-
-Monsoon currently lacks a registered Hub metric task. Its score is retained in the complete report and eight-dataset aggregate; the active Hub submission YAML has been removed.
-<!-- orukeet-open-asr-20260918:end -->
-
 ## sherpa-onnx inference
 
 The [ONNX INT8 archive](https://huggingface.co/oruk/orukeet/resolve/55a984d46f68323301837194ce647c702f55facc/onnx/sherpa-onnx-orukeet-v0.1.0-int8.tar.bz2) uses the standard Parakeet TDT v3 layout: `encoder.int8.onnx`, `decoder.int8.onnx`, `joiner.int8.onnx` and `tokens.txt`. It also includes the BPE vocabulary, weight license and attribution. Gabor filters are ordinary convolution weights; the model uses sherpa-onnx's existing offline transducer loader.

evaluation/open_asr_20260918/official-scorer-output.txt

@@ -1,34 +0,0 @@
-Filtering models by id: oruk/orukeet
-********************************************************************************
-Results per dataset:
-********************************************************************************
-oruk/orukeet | ArtificialAnalysis-Earnings22-Cleaned-AA-chunked__test: WER = 5.74 %, RTFx = 2982.17
-oruk/orukeet | VoiceArena-Monsoon_en_IN_test__test: WER = 3.84 %, RTFx = 4835.62
-oruk/orukeet | hf-audio-open-asr-leaderboard_ami_cleaned_test: WER = 8.68 %, RTFx = 3741.22
-oruk/orukeet | hf-audio-open-asr-leaderboard_gigaspeech_cleaned_test: WER = 7.48 %, RTFx = 5346.49
-oruk/orukeet | hf-audio-open-asr-leaderboard_librispeech_test.clean: WER = 1.47 %, RTFx = 4180.11
-oruk/orukeet | hf-audio-open-asr-leaderboard_librispeech_test.other: WER = 2.84 %, RTFx = 4159.42
-oruk/orukeet | hf-audio-open-asr-leaderboard_spgispeech_test: WER = 3.38 %, RTFx = 6419.21
-oruk/orukeet | hf-audio-open-asr-leaderboard_voxpopuli_cleaned_aa_test: WER = 2.46 %, RTFx = 2488.16
-
-********************************************************************************
-Composite Results:
-********************************************************************************
-oruk/orukeet: WER = 4.49 %
-oruk/orukeet: RTFx = 5545.68
-********************************************************************************
-
-********************************************************************************
-CSV Summary (Public):
-********************************************************************************
-avg WER (oruk/orukeet) = 4.49
-model,RTFx,License,Size (B),# Languages,Encoder,Decoder,AMI-Cleaned WER,Earnings22-Cleaned-AA-chunked WER,Gigaspeech-Cleaned WER,LS Clean WER,LS Other WER,SPGISpeech WER,Voice-Arena-Moonsoon WER,Voxpopuli-Cleaned-AA WER,AMI-Cleaned RTFx,Earnings22-Cleaned-AA-chunked RTFx,Gigaspeech-Cleaned RTFx,LS Clean RTFx,LS Other RTFx,SPGISpeech RTFx,Voice-Arena-Moonsoon RTFx,Voxpopuli-Cleaned-AA RTFx
-oruk/orukeet,5545.68,,,,,,8.68,5.74,7.48,1.47,2.84,3.38,3.84,2.46,3741.2159,2982.1706,5346.4904,4180.1148,4159.4244,6419.2142,4835.625,2488.1643
-********************************************************************************
-
-********************************************************************************
-CSV Summary (Extra):
-********************************************************************************
-model,AMI WER,Earnings22 WER,Gigaspeech WER,Voxpopuli WER
-oruk/orukeet,,,,
-********************************************************************************

evaluation/open_asr_20260918/public-eight-h200-results.json

@@ -1,1530 +0,0 @@
-{
-  "schema_version": 1,
-  "complete_public_eight_metrics": true,
-  "complete_per_dataset_hub_task_coverage": false,
-  "missing_registered_tasks": [
-    "voicearena_monsoon_en_in"
-  ],
-  "model": {
-    "model_id": "oruk/orukeet",
-    "parameter_count": 627008134,
-    "size_billions": 0.627008134,
-    "supported_language_count": 25,
-    "evaluated_languages": [
-      "en"
-    ],
-    "license": "cc-by-sa-4.0",
-    "encoder": "FastConformer",
-    "decoder": "Token-and-Duration Transducer (TDT)",
-    "checkpoint_file": "orukeet-v0.1.0.nemo",
-    "checkpoint_sha256": "031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56",
-    "parameter_scope": "total parameters, not trainable parameters only",
-    "known_exposure": [
-      "All 2939 LibriSpeech test-other recordings used for r3 training and selection.",
-      "Monsoon English informed earlier training-partition selection.",
-      "Other inherited training exposure is incompletely verified."
-    ]
-  },
-  "evaluation_date": "2026-09-18",
-  "aggregate": {
-    "mean_wer_percent": 4.49,
-    "mean_of_two_decimal_dataset_wer": 4.48625,
-    "rtfx": 5545.68,
-    "rtfx_unrounded": 5545.67879726494,
-    "audio_seconds": 588203.26,
-    "transcription_seconds": 106.06515117501839
-  },
-  "datasets": {
-    "ami_cleaned": {
-      "wer": 8.68,
-      "audio_length": 28727.5903125,
-      "inference_time": 7.678677501156925,
-      "rtfx": 3741.2159,
-      "ins": 1873,
-      "del": 2187,
-      "sub": 3085
-    },
-    "earnings22_cleaned_aa_chunked": {
-      "wer": 5.74,
-      "audio_length": 6900.9376250000005,
-      "inference_time": 2.3140653409063816,
-      "rtfx": 2982.1706,
-      "ins": 101,
-      "del": 237,
-      "sub": 545
-    },
-    "gigaspeech_cleaned": {
-      "wer": 7.48,
-      "audio_length": 126515.196,
-      "inference_time": 23.663223169045523,
-      "rtfx": 5346.4904,
-      "ins": 8004,
-      "del": 4624,
-      "sub": 17079
-    },
-    "librispeech_test_clean": {
-      "wer": 1.47,
-      "audio_length": 19452.480625,
-      "inference_time": 4.653575702803209,
-      "rtfx": 4180.1148,
-      "ins": 72,
-      "del": 65,
-      "sub": 640
-    },
-    "librispeech_test_other": {
-      "wer": 2.84,
-      "audio_length": 19229.570125,
-      "inference_time": 4.6231324940454215,
-      "rtfx": 4159.4244,
-      "ins": 132,
-      "del": 102,
-      "sub": 1265
-    },
-    "spgispeech": {
-      "wer": 3.38,
-      "audio_length": 360007.4995,
-      "inference_time": 56.082798800896846,
-      "rtfx": 6419.2142,
-      "ins": 17251,
-      "del": 2741,
-      "sub": 12728
-    },
-    "voicearena_monsoon_en_in": {
-      "wer": 3.84,
-      "audio_length": 20247.6074375,
-      "inference_time": 4.187174894148484,
-      "rtfx": 4835.625,
-      "ins": 481,
-      "del": 299,
-      "sub": 1095
-    },
-    "voxpopuli_cleaned_aa": {
-      "wer": 2.46,
-      "audio_length": 7122.378375,
-      "inference_time": 2.8625032720156014,
-      "rtfx": 2488.1643,
-      "ins": 99,
-      "del": 147,
-      "sub": 191
-    }
-  },
-  "runs": {
-    "ami_cleaned": {
-      "completion_sha256": "867369d9353d5006fb9219b0cb870bd7591383d54a862a4fd843657301610b31",
-      "manifest_sha256": "79c7419a85acda2054114234299ff7574efcf3cb0693d40e1c7bd5a6e69d416b",
-      "run_sha256": "823d21dbdd15fa3cd20c3fd002f2ba42bef6b0a0e8ee1d31382fc54099d13246",
-      "inputs_sha256": "e50554e60ce197c9c6d96a4efb86768d58c48e1bb50a57c05bce7780dcd3f28b",
-      "model_revision": "555136b50265a132d4cea0d35560c26fc4f657ab",
-      "model_sha256": "031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56",
-      "dataset": {
-        "dataset_path": "hf-audio/open-asr-leaderboard",
-        "dataset": "ami_cleaned",
-        "split": "test",
-        "dataset_revision": "b6bdcd0beb34f8975dc659796176d88f43aff502",
-        "parent_dataset_revision": null
-      },
-      "counts": {
-        "source_count": 7805,
-        "eligible_count": 7715,
-        "prediction_count": 7715
-      },
-      "runtime": {
-        "gpu_name": "NVIDIA H200",
-        "device": "cuda:0",
-        "dtype": "torch.bfloat16",
-        "batch_size": 128,
-        "num_workers": 1,
-        "packages": {
-          "torch": "2.8.0+cu128",
-          "nemo_toolkit": "2.7.2",
-          "datasets": "5.0.0",
-          "torchcodec": "0.6.0+cu128",
-          "soundfile": "0.14.0",
-          "numpy": "2.4.6",
-          "huggingface_hub": "0.36.2"
-        },
-        "decoding": {
-          "model_type": "tdt",
-          "strategy": "greedy_batch",
-          "compute_hypothesis_token_set": false,
-          "preserve_alignments": null,
-          "tdt_include_token_duration": null,
-          "confidence_cfg": {
-            "preserve_frame_confidence": false,
-            "preserve_token_confidence": false,
-            "preserve_word_confidence": false,
-            "exclude_blank": true,
-            "aggregation": "min",
-            "tdt_include_duration": false,
-            "method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            }
-          },
-          "fused_batch_size": null,
-          "compute_timestamps": null,
-          "compute_langs": false,
-          "word_seperator": " ",
-          "segment_seperators": [
-            ".",
-            "!",
-            "?"
-          ],
-          "segment_gap_threshold": null,
-          "rnnt_timestamp_type": "all",
-          "greedy": {
-            "max_symbols_per_step": 10,
-            "preserve_alignments": false,
-            "preserve_frame_confidence": false,
-            "tdt_include_token_duration": false,
-            "tdt_include_duration_confidence": false,
-            "confidence_method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            },
-            "loop_labels": true,
-            "use_cuda_graph_decoder": true,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "enable_per_stream_biasing": false,
-            "max_symbols": 10
-          },
-          "beam": {
-            "beam_size": 2,
-            "search_type": "default",
-            "score_norm": true,
-            "return_best_hypothesis": false,
-            "tsd_max_sym_exp_per_step": 50,
-            "alsd_max_target_len": 2.0,
-            "nsc_max_timesteps_expansion": 1,
-            "nsc_prefix_alpha": 1,
-            "maes_num_steps": 2,
-            "maes_prefix_alpha": 1,
-            "maes_expansion_gamma": 2.3,
-            "maes_expansion_beta": 2,
-            "language_model": null,
-            "softmax_temperature": 1.0,
-            "preserve_alignments": false,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "hat_subtract_ilm": false,
-            "hat_ilm_weight": 0.0,
-            "max_symbols_per_step": 10,
-            "blank_lm_score_mode": "LM_WEIGHTED_FULL",
-            "pruning_mode": "LATE",
-            "allow_cuda_graphs": true,
-            "tsd_max_sym_exp": 50
-          },
-          "temperature": 1.0,
-          "durations": [
-            0,
-            1,
-            2,
-            3,
-            4
-          ],
-          "big_blank_durations": []
-        },
-        "parameter_dtypes": [
-          "torch.bfloat16"
-        ],
-        "parameter_devices": [
-          "cuda:0"
-        ],
-        "warmup_batches": 4,
-        "cuda_synchronized": true,
-        "clock": "time.perf_counter",
-        "segmentation": "unchanged source clips; no added chunking or truncation"
-      },
-      "timing": {
-        "audio_seconds": 28727.5903125,
-        "transcription_seconds": 7.678677501156926,
-        "rtfx_unrounded": 3741.2158940353584,
-        "timing_scope": "full transcribe pass including file loading and decoding; four-batch warmup excluded; CUDA synchronized"
-      }
-    },
-    "earnings22_cleaned_aa_chunked": {
-      "completion_sha256": "09fe168223cb28d0f57fc7d33d660e7a23769dbb2e6f00e20010fe25364c1125",
-      "manifest_sha256": "6d5361d4b3946158906882ff454050ed31b2ec7e3458044a7a06ed4acd938229",
-      "run_sha256": "d09c8781619a5a6d279bc5900a29e5825308298ec0e8b23c517558e969936a67",
-      "inputs_sha256": "2ef1afa2dab81292b15a4c2e43fcded0ae925009f795e50136cbf86fafbd657e",
-      "model_revision": "555136b50265a132d4cea0d35560c26fc4f657ab",
-      "model_sha256": "031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56",
-      "dataset": {
-        "dataset_path": "ArtificialAnalysis/Earnings22-Cleaned-AA-chunked",
-        "dataset": "",
-        "split": "test",
-        "dataset_revision": "27f7d0056d593231974a752ce03c1e965da69b3f",
-        "parent_dataset_revision": "5999936ee3d8d3cde160e3c65a8cc6ff5b7c5276"
-      },
-      "counts": {
-        "source_count": 341,
-        "eligible_count": 341,
-        "prediction_count": 341
-      },
-      "runtime": {
-        "gpu_name": "NVIDIA H200",
-        "device": "cuda:0",
-        "dtype": "torch.bfloat16",
-        "batch_size": 128,
-        "num_workers": 1,
-        "packages": {
-          "torch": "2.8.0+cu128",
-          "nemo_toolkit": "2.7.2",
-          "datasets": "5.0.0",
-          "torchcodec": "0.6.0+cu128",
-          "soundfile": "0.14.0",
-          "numpy": "2.4.6",
-          "huggingface_hub": "0.36.2"
-        },
-        "decoding": {
-          "model_type": "tdt",
-          "strategy": "greedy_batch",
-          "compute_hypothesis_token_set": false,
-          "preserve_alignments": null,
-          "tdt_include_token_duration": null,
-          "confidence_cfg": {
-            "preserve_frame_confidence": false,
-            "preserve_token_confidence": false,
-            "preserve_word_confidence": false,
-            "exclude_blank": true,
-            "aggregation": "min",
-            "tdt_include_duration": false,
-            "method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            }
-          },
-          "fused_batch_size": null,
-          "compute_timestamps": null,
-          "compute_langs": false,
-          "word_seperator": " ",
-          "segment_seperators": [
-            ".",
-            "!",
-            "?"
-          ],
-          "segment_gap_threshold": null,
-          "rnnt_timestamp_type": "all",
-          "greedy": {
-            "max_symbols_per_step": 10,
-            "preserve_alignments": false,
-            "preserve_frame_confidence": false,
-            "tdt_include_token_duration": false,
-            "tdt_include_duration_confidence": false,
-            "confidence_method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            },
-            "loop_labels": true,
-            "use_cuda_graph_decoder": true,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "enable_per_stream_biasing": false,
-            "max_symbols": 10
-          },
-          "beam": {
-            "beam_size": 2,
-            "search_type": "default",
-            "score_norm": true,
-            "return_best_hypothesis": false,
-            "tsd_max_sym_exp_per_step": 50,
-            "alsd_max_target_len": 2.0,
-            "nsc_max_timesteps_expansion": 1,
-            "nsc_prefix_alpha": 1,
-            "maes_num_steps": 2,
-            "maes_prefix_alpha": 1,
-            "maes_expansion_gamma": 2.3,
-            "maes_expansion_beta": 2,
-            "language_model": null,
-            "softmax_temperature": 1.0,
-            "preserve_alignments": false,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "hat_subtract_ilm": false,
-            "hat_ilm_weight": 0.0,
-            "max_symbols_per_step": 10,
-            "blank_lm_score_mode": "LM_WEIGHTED_FULL",
-            "pruning_mode": "LATE",
-            "allow_cuda_graphs": true,
-            "tsd_max_sym_exp": 50
-          },
-          "temperature": 1.0,
-          "durations": [
-            0,
-            1,
-            2,
-            3,
-            4
-          ],
-          "big_blank_durations": []
-        },
-        "parameter_dtypes": [
-          "torch.bfloat16"
-        ],
-        "parameter_devices": [
-          "cuda:0"
-        ],
-        "warmup_batches": 4,
-        "cuda_synchronized": true,
-        "clock": "time.perf_counter",
-        "segmentation": "unchanged source clips; no added chunking or truncation"
-      },
-      "timing": {
-        "audio_seconds": 6900.937625,
-        "transcription_seconds": 2.3140653409063816,
-        "rtfx_unrounded": 2982.1705995116868,
-        "timing_scope": "full transcribe pass including file loading and decoding; four-batch warmup excluded; CUDA synchronized"
-      }
-    },
-    "gigaspeech_cleaned": {
-      "completion_sha256": "49e46bcdd03efe2eb8398417a6427ebb34babec98a17b8f30990fb676757b128",
-      "manifest_sha256": "b2904a6c1e25d4146ec46f416f6837ee6b751d8d2f5b412bd8f0e990357f6c6c",
-      "run_sha256": "e9ac3a765118471602ef13709383dae3ba4f316c083bc3f231cc8b37d4cbaa99",
-      "inputs_sha256": "91a933d2a4128e506f7f7faec4c0a218b804fd3d1298d775092729e9520ff6b3",
-      "model_revision": "555136b50265a132d4cea0d35560c26fc4f657ab",
-      "model_sha256": "031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56",
-      "dataset": {
-        "dataset_path": "hf-audio/open-asr-leaderboard",
-        "dataset": "gigaspeech_cleaned",
-        "split": "test",
-        "dataset_revision": "b6bdcd0beb34f8975dc659796176d88f43aff502",
-        "parent_dataset_revision": null
-      },
-      "counts": {
-        "source_count": 18768,
-        "eligible_count": 18757,
-        "prediction_count": 18757
-      },
-      "runtime": {
-        "gpu_name": "NVIDIA H200",
-        "device": "cuda:0",
-        "dtype": "torch.bfloat16",
-        "batch_size": 128,
-        "num_workers": 1,
-        "packages": {
-          "torch": "2.8.0+cu128",
-          "nemo_toolkit": "2.7.2",
-          "datasets": "5.0.0",
-          "torchcodec": "0.6.0+cu128",
-          "soundfile": "0.14.0",
-          "numpy": "2.4.6",
-          "huggingface_hub": "0.36.2"
-        },
-        "decoding": {
-          "model_type": "tdt",
-          "strategy": "greedy_batch",
-          "compute_hypothesis_token_set": false,
-          "preserve_alignments": null,
-          "tdt_include_token_duration": null,
-          "confidence_cfg": {
-            "preserve_frame_confidence": false,
-            "preserve_token_confidence": false,
-            "preserve_word_confidence": false,
-            "exclude_blank": true,
-            "aggregation": "min",
-            "tdt_include_duration": false,
-            "method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            }
-          },
-          "fused_batch_size": null,
-          "compute_timestamps": null,
-          "compute_langs": false,
-          "word_seperator": " ",
-          "segment_seperators": [
-            ".",
-            "!",
-            "?"
-          ],
-          "segment_gap_threshold": null,
-          "rnnt_timestamp_type": "all",
-          "greedy": {
-            "max_symbols_per_step": 10,
-            "preserve_alignments": false,
-            "preserve_frame_confidence": false,
-            "tdt_include_token_duration": false,
-            "tdt_include_duration_confidence": false,
-            "confidence_method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            },
-            "loop_labels": true,
-            "use_cuda_graph_decoder": true,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "enable_per_stream_biasing": false,
-            "max_symbols": 10
-          },
-          "beam": {
-            "beam_size": 2,
-            "search_type": "default",
-            "score_norm": true,
-            "return_best_hypothesis": false,
-            "tsd_max_sym_exp_per_step": 50,
-            "alsd_max_target_len": 2.0,
-            "nsc_max_timesteps_expansion": 1,
-            "nsc_prefix_alpha": 1,
-            "maes_num_steps": 2,
-            "maes_prefix_alpha": 1,
-            "maes_expansion_gamma": 2.3,
-            "maes_expansion_beta": 2,
-            "language_model": null,
-            "softmax_temperature": 1.0,
-            "preserve_alignments": false,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "hat_subtract_ilm": false,
-            "hat_ilm_weight": 0.0,
-            "max_symbols_per_step": 10,
-            "blank_lm_score_mode": "LM_WEIGHTED_FULL",
-            "pruning_mode": "LATE",
-            "allow_cuda_graphs": true,
-            "tsd_max_sym_exp": 50
-          },
-          "temperature": 1.0,
-          "durations": [
-            0,
-            1,
-            2,
-            3,
-            4
-          ],
-          "big_blank_durations": []
-        },
-        "parameter_dtypes": [
-          "torch.bfloat16"
-        ],
-        "parameter_devices": [
-          "cuda:0"
-        ],
-        "warmup_batches": 4,
-        "cuda_synchronized": true,
-        "clock": "time.perf_counter",
-        "segmentation": "unchanged source clips; no added chunking or truncation"
-      },
-      "timing": {
-        "audio_seconds": 126515.196,
-        "transcription_seconds": 23.663223169045523,
-        "rtfx_unrounded": 5346.490420861086,
-        "timing_scope": "full transcribe pass including file loading and decoding; four-batch warmup excluded; CUDA synchronized"
-      }
-    },
-    "librispeech_test_clean": {
-      "completion_sha256": "454e79fc47aeac5e5c3f20a0e333abb63aef21bab587d52fb881ed1b132c3ad6",
-      "manifest_sha256": "ef41dea7fd36011ea044679c1d5f5dd419fd2660a86408db1265f72f381ba8d0",
-      "run_sha256": "035e817cb45c3b2ff60a0501196871ec830e192b0706352376980bf5cf586952",
-      "inputs_sha256": "d3d9dc61b5ac17c6c66f66e0da3d25c15acbbc36fb69ee4fd0cd60ee1ddd4c6b",
-      "model_revision": "555136b50265a132d4cea0d35560c26fc4f657ab",
-      "model_sha256": "031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56",
-      "dataset": {
-        "dataset_path": "hf-audio/open-asr-leaderboard",
-        "dataset": "librispeech",
-        "split": "test.clean",
-        "dataset_revision": "b6bdcd0beb34f8975dc659796176d88f43aff502",
-        "parent_dataset_revision": null
-      },
-      "counts": {
-        "source_count": 2620,
-        "eligible_count": 2620,
-        "prediction_count": 2620
-      },
-      "runtime": {
-        "gpu_name": "NVIDIA H200",
-        "device": "cuda:0",
-        "dtype": "torch.bfloat16",
-        "batch_size": 128,
-        "num_workers": 1,
-        "packages": {
-          "torch": "2.8.0+cu128",
-          "nemo_toolkit": "2.7.2",
-          "datasets": "5.0.0",
-          "torchcodec": "0.6.0+cu128",
-          "soundfile": "0.14.0",
-          "numpy": "2.4.6",
-          "huggingface_hub": "0.36.2"
-        },
-        "decoding": {
-          "model_type": "tdt",
-          "strategy": "greedy_batch",
-          "compute_hypothesis_token_set": false,
-          "preserve_alignments": null,
-          "tdt_include_token_duration": null,
-          "confidence_cfg": {
-            "preserve_frame_confidence": false,
-            "preserve_token_confidence": false,
-            "preserve_word_confidence": false,
-            "exclude_blank": true,
-            "aggregation": "min",
-            "tdt_include_duration": false,
-            "method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            }
-          },
-          "fused_batch_size": null,
-          "compute_timestamps": null,
-          "compute_langs": false,
-          "word_seperator": " ",
-          "segment_seperators": [
-            ".",
-            "!",
-            "?"
-          ],
-          "segment_gap_threshold": null,
-          "rnnt_timestamp_type": "all",
-          "greedy": {
-            "max_symbols_per_step": 10,
-            "preserve_alignments": false,
-            "preserve_frame_confidence": false,
-            "tdt_include_token_duration": false,
-            "tdt_include_duration_confidence": false,
-            "confidence_method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            },
-            "loop_labels": true,
-            "use_cuda_graph_decoder": true,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "enable_per_stream_biasing": false,
-            "max_symbols": 10
-          },
-          "beam": {
-            "beam_size": 2,
-            "search_type": "default",
-            "score_norm": true,
-            "return_best_hypothesis": false,
-            "tsd_max_sym_exp_per_step": 50,
-            "alsd_max_target_len": 2.0,
-            "nsc_max_timesteps_expansion": 1,
-            "nsc_prefix_alpha": 1,
-            "maes_num_steps": 2,
-            "maes_prefix_alpha": 1,
-            "maes_expansion_gamma": 2.3,
-            "maes_expansion_beta": 2,
-            "language_model": null,
-            "softmax_temperature": 1.0,
-            "preserve_alignments": false,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "hat_subtract_ilm": false,
-            "hat_ilm_weight": 0.0,
-            "max_symbols_per_step": 10,
-            "blank_lm_score_mode": "LM_WEIGHTED_FULL",
-            "pruning_mode": "LATE",
-            "allow_cuda_graphs": true,
-            "tsd_max_sym_exp": 50
-          },
-          "temperature": 1.0,
-          "durations": [
-            0,
-            1,
-            2,
-            3,
-            4
-          ],
-          "big_blank_durations": []
-        },
-        "parameter_dtypes": [
-          "torch.bfloat16"
-        ],
-        "parameter_devices": [
-          "cuda:0"
-        ],
-        "warmup_batches": 4,
-        "cuda_synchronized": true,
-        "clock": "time.perf_counter",
-        "segmentation": "unchanged source clips; no added chunking or truncation"
-      },
-      "timing": {
-        "audio_seconds": 19452.480625,
-        "transcription_seconds": 4.653575702803209,
-        "rtfx_unrounded": 4180.114790715076,
-        "timing_scope": "full transcribe pass including file loading and decoding; four-batch warmup excluded; CUDA synchronized"
-      }
-    },
-    "librispeech_test_other": {
-      "completion_sha256": "b996b7eddeacb73cb4bd4574098a79cac229697b2d5f517fc8984cb4276bf484",
-      "manifest_sha256": "0358a449ca0adbd031371a8c342189b67330d3ec66f8dbb8ca898829521c215e",
-      "run_sha256": "4c5ecfb0c6ce492a92634607b786e9292b5c43df654d9bf13b27f91b3aa2391f",
-      "inputs_sha256": "c695e3fea3f000b019ed0a7a5bea10a3036f0f4952927887174f0f1466337c63",
-      "model_revision": "555136b50265a132d4cea0d35560c26fc4f657ab",
-      "model_sha256": "031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56",
-      "dataset": {
-        "dataset_path": "hf-audio/open-asr-leaderboard",
-        "dataset": "librispeech",
-        "split": "test.other",
-        "dataset_revision": "b6bdcd0beb34f8975dc659796176d88f43aff502",
-        "parent_dataset_revision": null
-      },
-      "counts": {
-        "source_count": 2939,
-        "eligible_count": 2939,
-        "prediction_count": 2939
-      },
-      "runtime": {
-        "gpu_name": "NVIDIA H200",
-        "device": "cuda:0",
-        "dtype": "torch.bfloat16",
-        "batch_size": 128,
-        "num_workers": 1,
-        "packages": {
-          "torch": "2.8.0+cu128",
-          "nemo_toolkit": "2.7.2",
-          "datasets": "5.0.0",
-          "torchcodec": "0.6.0+cu128",
-          "soundfile": "0.14.0",
-          "numpy": "2.4.6",
-          "huggingface_hub": "0.36.2"
-        },
-        "decoding": {
-          "model_type": "tdt",
-          "strategy": "greedy_batch",
-          "compute_hypothesis_token_set": false,
-          "preserve_alignments": null,
-          "tdt_include_token_duration": null,
-          "confidence_cfg": {
-            "preserve_frame_confidence": false,
-            "preserve_token_confidence": false,
-            "preserve_word_confidence": false,
-            "exclude_blank": true,
-            "aggregation": "min",
-            "tdt_include_duration": false,
-            "method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            }
-          },
-          "fused_batch_size": null,
-          "compute_timestamps": null,
-          "compute_langs": false,
-          "word_seperator": " ",
-          "segment_seperators": [
-            ".",
-            "!",
-            "?"
-          ],
-          "segment_gap_threshold": null,
-          "rnnt_timestamp_type": "all",
-          "greedy": {
-            "max_symbols_per_step": 10,
-            "preserve_alignments": false,
-            "preserve_frame_confidence": false,
-            "tdt_include_token_duration": false,
-            "tdt_include_duration_confidence": false,
-            "confidence_method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            },
-            "loop_labels": true,
-            "use_cuda_graph_decoder": true,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "enable_per_stream_biasing": false,
-            "max_symbols": 10
-          },
-          "beam": {
-            "beam_size": 2,
-            "search_type": "default",
-            "score_norm": true,
-            "return_best_hypothesis": false,
-            "tsd_max_sym_exp_per_step": 50,
-            "alsd_max_target_len": 2.0,
-            "nsc_max_timesteps_expansion": 1,
-            "nsc_prefix_alpha": 1,
-            "maes_num_steps": 2,
-            "maes_prefix_alpha": 1,
-            "maes_expansion_gamma": 2.3,
-            "maes_expansion_beta": 2,
-            "language_model": null,
-            "softmax_temperature": 1.0,
-            "preserve_alignments": false,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "hat_subtract_ilm": false,
-            "hat_ilm_weight": 0.0,
-            "max_symbols_per_step": 10,
-            "blank_lm_score_mode": "LM_WEIGHTED_FULL",
-            "pruning_mode": "LATE",
-            "allow_cuda_graphs": true,
-            "tsd_max_sym_exp": 50
-          },
-          "temperature": 1.0,
-          "durations": [
-            0,
-            1,
-            2,
-            3,
-            4
-          ],
-          "big_blank_durations": []
-        },
-        "parameter_dtypes": [
-          "torch.bfloat16"
-        ],
-        "parameter_devices": [
-          "cuda:0"
-        ],
-        "warmup_batches": 4,
-        "cuda_synchronized": true,
-        "clock": "time.perf_counter",
-        "segmentation": "unchanged source clips; no added chunking or truncation"
-      },
-      "timing": {
-        "audio_seconds": 19229.570125,
-        "transcription_seconds": 4.6231324940454215,
-        "rtfx_unrounded": 4159.424405371816,
-        "timing_scope": "full transcribe pass including file loading and decoding; four-batch warmup excluded; CUDA synchronized"
-      }
-    },
-    "spgispeech": {
-      "completion_sha256": "1e24a2e56d62cb37ea9b5d320aed0a536f83924fdaf7d6bcca7f3441f8c0f370",
-      "manifest_sha256": "04cee652d2ec9734bbf22918f6d848a3e812465e09e3dd5e4d49f373e189f4e3",
-      "run_sha256": "014659bc814f3b8a61e234d65fdeeb0ce0095ff23d9a267f933ec02309870963",
-      "inputs_sha256": "7a60c54feeedfe78479c1f2782948dd902374e3fda3d7a1082e16cea20191ca2",
-      "model_revision": "555136b50265a132d4cea0d35560c26fc4f657ab",
-      "model_sha256": "031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56",
-      "dataset": {
-        "dataset_path": "hf-audio/open-asr-leaderboard",
-        "dataset": "spgispeech",
-        "split": "test",
-        "dataset_revision": "b6bdcd0beb34f8975dc659796176d88f43aff502",
-        "parent_dataset_revision": null
-      },
-      "counts": {
-        "source_count": 39341,
-        "eligible_count": 39341,
-        "prediction_count": 39341
-      },
-      "runtime": {
-        "gpu_name": "NVIDIA H200",
-        "device": "cuda:0",
-        "dtype": "torch.bfloat16",
-        "batch_size": 128,
-        "num_workers": 1,
-        "packages": {
-          "torch": "2.8.0+cu128",
-          "nemo_toolkit": "2.7.2",
-          "datasets": "5.0.0",
-          "torchcodec": "0.6.0+cu128",
-          "soundfile": "0.14.0",
-          "numpy": "2.4.6",
-          "huggingface_hub": "0.36.2"
-        },
-        "decoding": {
-          "model_type": "tdt",
-          "strategy": "greedy_batch",
-          "compute_hypothesis_token_set": false,
-          "preserve_alignments": null,
-          "tdt_include_token_duration": null,
-          "confidence_cfg": {
-            "preserve_frame_confidence": false,
-            "preserve_token_confidence": false,
-            "preserve_word_confidence": false,
-            "exclude_blank": true,
-            "aggregation": "min",
-            "tdt_include_duration": false,
-            "method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            }
-          },
-          "fused_batch_size": null,
-          "compute_timestamps": null,
-          "compute_langs": false,
-          "word_seperator": " ",
-          "segment_seperators": [
-            ".",
-            "!",
-            "?"
-          ],
-          "segment_gap_threshold": null,
-          "rnnt_timestamp_type": "all",
-          "greedy": {
-            "max_symbols_per_step": 10,
-            "preserve_alignments": false,
-            "preserve_frame_confidence": false,
-            "tdt_include_token_duration": false,
-            "tdt_include_duration_confidence": false,
-            "confidence_method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            },
-            "loop_labels": true,
-            "use_cuda_graph_decoder": true,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "enable_per_stream_biasing": false,
-            "max_symbols": 10
-          },
-          "beam": {
-            "beam_size": 2,
-            "search_type": "default",
-            "score_norm": true,
-            "return_best_hypothesis": false,
-            "tsd_max_sym_exp_per_step": 50,
-            "alsd_max_target_len": 2.0,
-            "nsc_max_timesteps_expansion": 1,
-            "nsc_prefix_alpha": 1,
-            "maes_num_steps": 2,
-            "maes_prefix_alpha": 1,
-            "maes_expansion_gamma": 2.3,
-            "maes_expansion_beta": 2,
-            "language_model": null,
-            "softmax_temperature": 1.0,
-            "preserve_alignments": false,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "hat_subtract_ilm": false,
-            "hat_ilm_weight": 0.0,
-            "max_symbols_per_step": 10,
-            "blank_lm_score_mode": "LM_WEIGHTED_FULL",
-            "pruning_mode": "LATE",
-            "allow_cuda_graphs": true,
-            "tsd_max_sym_exp": 50
-          },
-          "temperature": 1.0,
-          "durations": [
-            0,
-            1,
-            2,
-            3,
-            4
-          ],
-          "big_blank_durations": []
-        },
-        "parameter_dtypes": [
-          "torch.bfloat16"
-        ],
-        "parameter_devices": [
-          "cuda:0"
-        ],
-        "warmup_batches": 4,
-        "cuda_synchronized": true,
-        "clock": "time.perf_counter",
-        "segmentation": "unchanged source clips; no added chunking or truncation"
-      },
-      "timing": {
-        "audio_seconds": 360007.4995,
-        "transcription_seconds": 56.08279880089685,
-        "rtfx_unrounded": 6419.214218928084,
-        "timing_scope": "full transcribe pass including file loading and decoding; four-batch warmup excluded; CUDA synchronized"
-      }
-    },
-    "voicearena_monsoon_en_in": {
-      "completion_sha256": "cd6b3b08c5a98e5c9c981305790c7c862a3f808575fa53c6c1129e2fb815d756",
-      "manifest_sha256": "c4d6afde9f6506cfa313d99b9c3019368c88c947ef05e95240fcc780c7f1b34b",
-      "run_sha256": "13fe7db18c2c944c0def9456fedd5b6adff6c2e5ed15c06dbcf80ba1143e7e42",
-      "inputs_sha256": "13845efbda33d013912cc7ffeba467c7a763fbb4ab17fb45770d26256414f512",
-      "model_revision": "555136b50265a132d4cea0d35560c26fc4f657ab",
-      "model_sha256": "031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56",
-      "dataset": {
-        "dataset_path": "VoiceArena/Monsoon_en_IN_test",
-        "dataset": "",
-        "split": "test",
-        "dataset_revision": "bc1da7b42ef6e2853123c97bf6d22067e4802d11",
-        "parent_dataset_revision": null
-      },
-      "counts": {
-        "source_count": 2102,
-        "eligible_count": 2102,
-        "prediction_count": 2102
-      },
-      "runtime": {
-        "gpu_name": "NVIDIA H200",
-        "device": "cuda:0",
-        "dtype": "torch.bfloat16",
-        "batch_size": 128,
-        "num_workers": 1,
-        "packages": {
-          "torch": "2.8.0+cu128",
-          "nemo_toolkit": "2.7.2",
-          "datasets": "5.0.0",
-          "torchcodec": "0.6.0+cu128",
-          "soundfile": "0.14.0",
-          "numpy": "2.4.6",
-          "huggingface_hub": "0.36.2"
-        },
-        "decoding": {
-          "model_type": "tdt",
-          "strategy": "greedy_batch",
-          "compute_hypothesis_token_set": false,
-          "preserve_alignments": null,
-          "tdt_include_token_duration": null,
-          "confidence_cfg": {
-            "preserve_frame_confidence": false,
-            "preserve_token_confidence": false,
-            "preserve_word_confidence": false,
-            "exclude_blank": true,
-            "aggregation": "min",
-            "tdt_include_duration": false,
-            "method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            }
-          },
-          "fused_batch_size": null,
-          "compute_timestamps": null,
-          "compute_langs": false,
-          "word_seperator": " ",
-          "segment_seperators": [
-            ".",
-            "!",
-            "?"
-          ],
-          "segment_gap_threshold": null,
-          "rnnt_timestamp_type": "all",
-          "greedy": {
-            "max_symbols_per_step": 10,
-            "preserve_alignments": false,
-            "preserve_frame_confidence": false,
-            "tdt_include_token_duration": false,
-            "tdt_include_duration_confidence": false,
-            "confidence_method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            },
-            "loop_labels": true,
-            "use_cuda_graph_decoder": true,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "enable_per_stream_biasing": false,
-            "max_symbols": 10
-          },
-          "beam": {
-            "beam_size": 2,
-            "search_type": "default",
-            "score_norm": true,
-            "return_best_hypothesis": false,
-            "tsd_max_sym_exp_per_step": 50,
-            "alsd_max_target_len": 2.0,
-            "nsc_max_timesteps_expansion": 1,
-            "nsc_prefix_alpha": 1,
-            "maes_num_steps": 2,
-            "maes_prefix_alpha": 1,
-            "maes_expansion_gamma": 2.3,
-            "maes_expansion_beta": 2,
-            "language_model": null,
-            "softmax_temperature": 1.0,
-            "preserve_alignments": false,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "hat_subtract_ilm": false,
-            "hat_ilm_weight": 0.0,
-            "max_symbols_per_step": 10,
-            "blank_lm_score_mode": "LM_WEIGHTED_FULL",
-            "pruning_mode": "LATE",
-            "allow_cuda_graphs": true,
-            "tsd_max_sym_exp": 50
-          },
-          "temperature": 1.0,
-          "durations": [
-            0,
-            1,
-            2,
-            3,
-            4
-          ],
-          "big_blank_durations": []
-        },
-        "parameter_dtypes": [
-          "torch.bfloat16"
-        ],
-        "parameter_devices": [
-          "cuda:0"
-        ],
-        "warmup_batches": 4,
-        "cuda_synchronized": true,
-        "clock": "time.perf_counter",
-        "segmentation": "unchanged source clips; no added chunking or truncation"
-      },
-      "timing": {
-        "audio_seconds": 20247.6074375,
-        "transcription_seconds": 4.187174894148484,
-        "rtfx_unrounded": 4835.624961784075,
-        "timing_scope": "full transcribe pass including file loading and decoding; four-batch warmup excluded; CUDA synchronized"
-      }
-    },
-    "voxpopuli_cleaned_aa": {
-      "completion_sha256": "92805a8d186411aeb61b1591222be24878c95669fa80bc52335ce95f1ad960b7",
-      "manifest_sha256": "4ea41ceff0db011f6b5caa45825a5613489a4043f1c95c7b1e2a0b7c63009519",
-      "run_sha256": "2f6ab7c29a8f5912e020444aefee81beec6239f926ed4a3185bc8528a0fdea2c",
-      "inputs_sha256": "98255266f2e023d8bb891e76c17b67494a4c3a9a6b1bdaff848739992f9cc078",
-      "model_revision": "555136b50265a132d4cea0d35560c26fc4f657ab",
-      "model_sha256": "031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56",
-      "dataset": {
-        "dataset_path": "hf-audio/open-asr-leaderboard",
-        "dataset": "voxpopuli_cleaned_aa",
-        "split": "test",
-        "dataset_revision": "b6bdcd0beb34f8975dc659796176d88f43aff502",
-        "parent_dataset_revision": null
-      },
-      "counts": {
-        "source_count": 628,
-        "eligible_count": 628,
-        "prediction_count": 628
-      },
-      "runtime": {
-        "gpu_name": "NVIDIA H200",
-        "device": "cuda:0",
-        "dtype": "torch.bfloat16",
-        "batch_size": 128,
-        "num_workers": 1,
-        "packages": {
-          "torch": "2.8.0+cu128",
-          "nemo_toolkit": "2.7.2",
-          "datasets": "5.0.0",
-          "torchcodec": "0.6.0+cu128",
-          "soundfile": "0.14.0",
-          "numpy": "2.4.6",
-          "huggingface_hub": "0.36.2"
-        },
-        "decoding": {
-          "model_type": "tdt",
-          "strategy": "greedy_batch",
-          "compute_hypothesis_token_set": false,
-          "preserve_alignments": null,
-          "tdt_include_token_duration": null,
-          "confidence_cfg": {
-            "preserve_frame_confidence": false,
-            "preserve_token_confidence": false,
-            "preserve_word_confidence": false,
-            "exclude_blank": true,
-            "aggregation": "min",
-            "tdt_include_duration": false,
-            "method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            }
-          },
-          "fused_batch_size": null,
-          "compute_timestamps": null,
-          "compute_langs": false,
-          "word_seperator": " ",
-          "segment_seperators": [
-            ".",
-            "!",
-            "?"
-          ],
-          "segment_gap_threshold": null,
-          "rnnt_timestamp_type": "all",
-          "greedy": {
-            "max_symbols_per_step": 10,
-            "preserve_alignments": false,
-            "preserve_frame_confidence": false,
-            "tdt_include_token_duration": false,
-            "tdt_include_duration_confidence": false,
-            "confidence_method_cfg": {
-              "name": "entropy",
-              "entropy_type": "tsallis",
-              "alpha": 0.33,
-              "entropy_norm": "exp",
-              "temperature": "DEPRECATED"
-            },
-            "loop_labels": true,
-            "use_cuda_graph_decoder": true,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "enable_per_stream_biasing": false,
-            "max_symbols": 10
-          },
-          "beam": {
-            "beam_size": 2,
-            "search_type": "default",
-            "score_norm": true,
-            "return_best_hypothesis": false,
-            "tsd_max_sym_exp_per_step": 50,
-            "alsd_max_target_len": 2.0,
-            "nsc_max_timesteps_expansion": 1,
-            "nsc_prefix_alpha": 1,
-            "maes_num_steps": 2,
-            "maes_prefix_alpha": 1,
-            "maes_expansion_gamma": 2.3,
-            "maes_expansion_beta": 2,
-            "language_model": null,
-            "softmax_temperature": 1.0,
-            "preserve_alignments": false,
-            "ngram_lm_model": null,
-            "ngram_lm_alpha": 0.0,
-            "boosting_tree": {
-              "model_path": null,
-              "key_phrases_file": null,
-              "key_phrases_list": null,
-              "key_phrase_items_list": null,
-              "context_score": 1.0,
-              "depth_scaling": 2.0,
-              "unk_score": 0.0,
-              "final_eos_score": 1.0,
-              "score_per_phrase": 0.0,
-              "source_lang": "en",
-              "use_triton": true,
-              "uniform_weights": false,
-              "use_bpe_dropout": false,
-              "num_of_transcriptions": 5,
-              "bpe_alpha": 0.3
-            },
-            "boosting_tree_alpha": 0.0,
-            "hat_subtract_ilm": false,
-            "hat_ilm_weight": 0.0,
-            "max_symbols_per_step": 10,
-            "blank_lm_score_mode": "LM_WEIGHTED_FULL",
-            "pruning_mode": "LATE",
-            "allow_cuda_graphs": true,
-            "tsd_max_sym_exp": 50
-          },
-          "temperature": 1.0,
-          "durations": [
-            0,
-            1,
-            2,
-            3,
-            4
-          ],
-          "big_blank_durations": []
-        },
-        "parameter_dtypes": [
-          "torch.bfloat16"
-        ],
-        "parameter_devices": [
-          "cuda:0"
-        ],
-        "warmup_batches": 4,
-        "cuda_synchronized": true,
-        "clock": "time.perf_counter",
-        "segmentation": "unchanged source clips; no added chunking or truncation"
-      },
-      "timing": {
-        "audio_seconds": 7122.378375,
-        "transcription_seconds": 2.8625032720156014,
-        "rtfx_unrounded": 2488.1642737773545,
-        "timing_scope": "full transcribe pass including file loading and decoding; four-batch warmup excluded; CUDA synchronized"
-      }
-    }
-  },
-  "source_url": "https://huggingface.co/oruk/orukeet/blob/main/evaluation/open_asr_20260918/public-eight-h200-results.json",
-  "prepared_by_sha256": "600b58ad29df47ef1bbc833b25d99c6dd4fff3ed40cbf1ad002efa467cab0b33",
-  "bundle_sha256": "0cb403d1553052f0396770be499558cdd2a0d88dd29e73481121b54f715af5cd",
-  "task_map_sha256": "27d565b719a9f98e5cfd58e185404d3c129329ebfc5b73cb3de0399ee6a34137",
-  "hub_yaml": {
-    "path": ".eval_results/open_asr_leaderboard.yaml",
-    "sha256": "b2aa8144405d83c39278ff022e6eb16193a0b769d11b2775b52253db815d25e7"
-  },
-  "known_exposure": "LibriSpeech test-other was used for r3 training and selection. Monsoon English informed prior training-partition selection. Remaining inherited exposure is incompletely verified."
-}

evaluation/open_asr_20260918/public-eight-h200-results.md

@@ -1,16 +0,0 @@
-| Dataset | WER (%) | H200 RTFx |
-|:--|--:|--:|
-| ami_cleaned | 8.68 | 3741.22 |
-| earnings22_cleaned_aa_chunked | 5.74 | 2982.17 |
-| gigaspeech_cleaned | 7.48 | 5346.49 |
-| librispeech_test_clean | 1.47 | 4180.11 |
-| librispeech_test_other | 2.84 | 4159.42 |
-| spgispeech | 3.38 | 6419.21 |
-| voicearena_monsoon_en_in | 3.84 | 4835.62 |
-| voxpopuli_cleaned_aa | 2.46 | 2488.16 |
-| **Public-eight aggregate** | **4.49** | **5545.68** |
-
-LibriSpeech test-other was used for r3 training and selection. Monsoon English informed prior training-partition selection. Remaining inherited exposure is incompletely verified.
-
-Monsoon has no registered Hub metric task. All eight results are retained above; no task ID is invented.
-RTFx aggregate is total audio divided by total measured inference time, not an arithmetic mean of per-dataset RTFx.

evaluation/open_asr_20260918/training-disclosure.md

@@ -1,15 +0,0 @@
-# Orukeet r3: training and benchmark exposure
-
-This disclosure applies to [Orukeet](https://huggingface.co/oruk/orukeet), native NeMo artifact `orukeet-v0.1.0.nemo` at revision `555136b50265a132d4cea0d35560c26fc4f657ab`. Its SHA-256 is `031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56`.
-
-**LibriSpeech test-other was used directly for training and model selection.** The released r3 checkpoint trained on all 2,939 recordings for three passes, totaling 8,817 training presentations. The same split informed candidate selection. Its reported WER is therefore a result on exposed data, not a held-out generalization estimate. Exact source-ID and decoded-waveform matching confirmed all 2,939 records against the current evaluation partition.
-
-**The preceding FT-4035 continuation used 223,452 records across 24 complete selected partitions, totaling 371.47 hours for one pass.** The previously reported 6,118 examples were a sampled follow-up fit diagnostic; they were not the full training population. Those partitions comprised EuroSpeech Bulgarian, Greek and Italian; 17 English GigaSpeechBench domains; Golos crowd Russian; Lesbos Greek; and NST Danish and Swedish. These names describe that continuation stage, not a complete accounting of upstream Parakeet or earlier Orukeet training.
-
-**Voice Arena Monsoon English has prior evaluation and selection exposure.** All 2,102 records from the same source revision used in the current benchmark participated in the earlier evaluation that informed selection of FT-4035 training partitions. Monsoon itself was not among the 24 partitions directly trained on in that continuation. LibriSpeech test-clean was also evaluated earlier; evaluation alone is not presented as proof of direct training or of an additional selection decision.
-
-An audit of the retained FT-4035 source and training waveform fingerprints found no exact whole-waveform matches to the eight current public partitions. This limited comparison does not rule out inherited exposure, partial overlap, different segmentation or transformations of the same recording. GigaSpeechBench domain training and GigaSpeech-Cleaned evaluation are distinct source definitions; their similar names do not establish exact overlap or independence.
-
-The local eight-dataset mean includes test-other. A separately reported seven-dataset sensitivity omits that split, but still includes Monsoon selection exposure and does not certify the remaining audio as unseen. We have not selected a new subset retrospectively.
-
-We ask the leaderboard maintainers how this known exposure should be represented and whether they require a different checkpoint for inclusion. We do not assume that disclosure guarantees acceptance or that the published guidelines automatically exclude this model.

evaluation/standard_asr/CURRENT.md

@@ -1,30 +0,0 @@
-# Paired evaluation for the r3 technical report
-
-The report is bound to Orukeet NeMo SHA-256 `031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56` and stock Parakeet SHA-256 `3cbdc85877e668ca7b82d0d56770eb1fac76691f55d6b97545e8d61ca588d10d`.
-
-Two manifests are evaluated independently with `run.py`: the complete 25,705-recording LibriSpeech/FLEURS manifest, and the existing 12,006-recording accent/domain sample. Both models are decoded afresh. The domain manifest adds the generic evaluator's `language`, `pcm_sha256` and `reference_sha256` aliases while preserving every existing source field, text, waveform and record. Its source manifest remains immutable.
-
-`run.py` records an initial diagnostic score. `rescore.py` produces the manuscript scores using the pinned English/multilingual normalizers and compound-aware WER. `audit_predictions.py` independently imports the upstream normalizer definitions and scores complete partitions in batches. The report builder accepts only those audited scores.
-
-```sh
-python evaluation/standard_asr/rescore.py \
-  --private-records /path/to/private-records \
-  --evidence evidence/standard-asr-r3-20260908
-python evaluation/standard_asr/audit_predictions.py \
-  --private-records /path/to/private-records \
-  --evidence evidence/standard-asr-r3-20260908 \
-  --upstream-root /path/to/pinned-scoring-source
-```
-
-Repeat with `evidence/domains-r3-20260908` and its corresponding records. The source revision and per-file hashes for the scorer are in `vendor/provenance.json`. All normalizer dependencies are recorded in each final `comparison.json`.
-
-Without private audio or transcripts, reproduce every reported score and build the PDF:
-
-```sh
-python evaluation/standard_asr/build_current_report.py
-python scripts/build_neurips_report.py
-```
-
-Pooled WER divides summed edit counts by summed reference-word counts. FLEURS pooling includes all 25 languages, including English; the 25-language macro gives each language equal weight. Non-English compound alignment can change the reference-word denominator separately for each model. Win counts compare full-precision per-partition WERs. The two evaluation samples are never merged into one pooled headline.
-
-Test-other was used for the final adaptation and checkpoint selection. The domain sample contains 6,118 recordings from the preceding adaptation and retains previously audited Greek/Italian EuroSpeech transcript spans. No recordings are dropped. Numeric evidence is transcript-free; full manifests and hypotheses are archived in the private model repository.

evaluation/standard_asr/README.md

@@ -1,39 +1,5 @@
-# LibriSpeech and FLEURS comparison
+# LibriSpeech and FLEURS evaluation
 
-This evaluation compares the released Orukeet FT-4035 NeMo checkpoint with NVIDIA Parakeet TDT 0.6B v3 on both complete LibriSpeech test partitions and all 25 supported FLEURS test languages. The [score table](../../docs/standard-asr-benchmarks.md) reports WER and CER for each partition and both models.
+The complete source-model evaluation and reproduction code is maintained in the [Orukeet source repository](https://github.com/Oruk-AI/orukeet/tree/main/evaluation/standard_asr).
 
-The runner uses identical mono 16 kHz audio, FP32 weights, BF16 CUDA autocast and matched greedy-batch TDT decoding. It retains every record and every empty hypothesis. Final scoring uses the pinned English and multilingual normalizers in [vendor/](vendor/). English normalization standardizes spelling, names and numbers. Multilingual normalization retains diacritics, expands digits by language and aligns compound boundaries; WER uses compound-aware edit distance. CER uses normalized strings before boundary alignment, including spaces. Corpus rates pool integer edit counts; language macros weight their specified languages equally.
-
-Place the original LibriSpeech `test-clean` audio and `.trans.txt` files under `data/goal_v2_sources/libri/LibriSpeech/test-clean`, `test-other` under `data/librispeech/LibriSpeech/test-other`, and original FLEURS test FLAC files under `data/fleurs/<language>/test`, relative to `AUDIO_ROOT`. Use the recorded NeMo environment from the preceding source-model comparison. FLEURS references are fetched from the immutable dataset revision recorded in `prepare.py`; the scripts record reference-file and decoded-audio hashes.
-
-```sh
-python evaluation/standard_asr/prepare.py --audio-root "$AUDIO_ROOT" --output /tmp/orukeet-standard/prepared
-python evaluation/standard_asr/run.py \
-  --manifest /tmp/orukeet-standard/prepared/manifest.jsonl \
-  --parakeet /path/to/parakeet-tdt-0.6b-v3.nemo \
-  --orukeet /path/to/orukeet-v0.1.0rc1.nemo \
-  --metric-code evaluation/unseen \
-  --output /tmp/orukeet-standard/evaluation
-```
-
-The runner verifies both checkpoint hashes before inference. Resume accepts only predictions with matching checkpoint, manifest, runner and normalizer hashes. A decoded-audio hash check precedes inference. Out-of-memory recovery recursively reduces the batch size; it does not omit or shorten recordings.
-
-The inference runner also emits diagnostic counts under the preceding evaluation's normalizer. Preserve these as `inference-comparison.json` and `inference-numeric-evidence.jsonl.gz` in the evidence directory. Place the unchanged `manifest.jsonl`, `parakeet.jsonl` and `orukeet.jsonl` in a private records directory, then generate the final scores:
-
-```sh
-python -m pip install -r evaluation/standard_asr/requirements-score.txt
-python evaluation/standard_asr/rescore.py --private-records /path/to/private-records --evidence /path/to/evidence
-python evaluation/standard_asr/audit_predictions.py --private-records /path/to/private-records --evidence /path/to/evidence
-```
-
-The scoring implementation is pinned to [source revision 48219c6](https://github.com/huggingface/open_asr_leaderboard/tree/48219c6028db0517d704600d92f31edfc96e8c23/normalizer). [Provenance](vendor/provenance.json) records source and vendored hashes, exact extracted definitions and the Apache-2.0 license. Compound alignment can change the reference word count separately for each model; all resulting denominators are retained. Scoring changes never alter predictions or checkpoint identities. For the release audit, pass `--upstream-root /path/to/pinned-source` to `audit_predictions.py`; it independently loads the original normalization definitions and verifies all 54 model/partition WERs with full-partition batch scoring.
-
-`comparison.json` holds full-precision corpus results and checkpoint identities. `numeric-evidence.jsonl.gz` contains transcript-free per-record edit counts. `preparation.json` records the complete test membership and source revisions. To reproduce the release tables from the included counts:
-
-```sh
-python evaluation/standard_asr/build_materials.py
-python scripts/build_neurips_report.py
-python evaluation/standard_asr/build_materials.py --verify-pdf
-```
-
-Corpus sources: [LibriSpeech](https://www.openslr.org/12) and [FLEURS](https://arxiv.org/abs/2205.12446). Dataset audio and reference transcripts are obtained from the original providers; the release evidence contains counts and hashes.
+The [released-checkpoint score tables](../../docs/current-checkpoint-benchmarks.md), [methods](../../docs/technical-report.md), and recorded evaluation evidence remain available in this model repository.

evaluation/standard_asr/audit_predictions.py

@@ -1,104 +0,0 @@
-"""Re-score the private hypotheses and compare every record with release counts."""
-import argparse
-import gzip
-import hashlib
-import json
-from pathlib import Path
-import sys
-import ast
-from collections import defaultdict
-from difflib import SequenceMatcher
-import importlib.util
-import re
-import num2words
-from kaldialign import batch_error_rate
-
-ROOT = Path(__file__).resolve().parents[2]
-from scoring import counts
-
-
-def sha(path):
-    return hashlib.sha256(path.read_bytes()).hexdigest()
-
-
-def main():
-    p=argparse.ArgumentParser()
-    p.add_argument('--private-records',type=Path,required=True)
-    p.add_argument('--evidence',type=Path,default=ROOT/'evidence/standard-asr-20260908')
-    p.add_argument('--upstream-root',type=Path)
-    a=p.parse_args()
-    result=json.loads((a.evidence/'comparison.json').read_text())
-    manifest=a.private_records/'manifest.jsonl'
-    assert sha(manifest)==result['manifest_sha256']
-    rows=[json.loads(line) for line in manifest.open()]
-    numeric={}
-    with gzip.open(a.evidence/'numeric-evidence.jsonl.gz','rt') as stream:
-        for line in stream:
-            row=json.loads(line)
-            assert row['record_sha256'] not in numeric
-            numeric[row['record_sha256']]=row
-    assert len(numeric)==len(rows)==result['rows']
-    empty={}
-    paired=defaultdict(lambda: [[], []])
-    for model in ['parakeet','orukeet']:
-        path=a.private_records/(model+'.jsonl')
-        assert sha(path)==result['prediction_sha256'][model]
-        records=[json.loads(line) for line in path.open()]
-        predictions={r['uid']:r for r in records}
-        assert len(predictions)==len(records)==len(rows)
-        empty[model]=0
-        for row in rows:
-            prediction=predictions[row['uid']]
-            assert prediction['model_sha256']==result['models'][model]
-            assert prediction['manifest_sha256']==result['manifest_sha256']
-            values=counts(row['text'],prediction['prediction'],row['language'])
-            expected=numeric[hashlib.sha256(row['uid'].encode()).hexdigest()]
-            assert expected['split']==row['split']
-            for metric in ['errors','words','char_errors','chars','utterance_error']:
-                assert values[metric]==expected['counts'][model][metric],(row['uid'],model,metric)
-            empty[model]+=not bool(prediction['prediction'].strip())
-            paired[(row['split'], model)][0].append(row['text'])
-            paired[(row['split'], model)][1].append(prediction['prediction'])
-    audit=dict(status='passed',publication_authorized=False,records=len(rows),paired_predictions=2*len(rows),
-               models=result['models'],empty_hypotheses=empty,
-               checks=['Every private hypothesis re-scored locally','All reference and prediction hashes match',
-                       'Every WER/CER numerator and denominator matches released counts'],
-               inputs_sha256={name:sha(a.evidence/name) for name in ['comparison.json','numeric-evidence.jsonl.gz']},
-               manifest_sha256=result['manifest_sha256'],prediction_sha256=result['prediction_sha256'],
-               script_sha256=sha(Path(__file__)))
-    if a.upstream_root:
-        provenance=json.loads((ROOT/'evaluation/standard_asr/vendor/provenance.json').read_text())
-        for name,digest in provenance['upstream_sha256'].items():
-            assert sha(a.upstream_root/name)==digest,name
-        # Load upstream normalizers independently of the release's vendored package.
-        init=a.upstream_root/'normalizer/__init__.py'
-        spec=importlib.util.spec_from_file_location('upstream_text',init,
-                         submodule_search_locations=[str(init.parent)])
-        module=importlib.util.module_from_spec(spec);sys.modules[spec.name]=module
-        spec.loader.exec_module(module)
-        namespace=dict(BasicMultilingualTextNormalizer=module.BasicMultilingualTextNormalizer,
-                       re=re,num2words=num2words,FILLER_WORDS={},SequenceMatcher=SequenceMatcher)
-        for filename,name in [('data_utils.py','MultilingualNormalizer'),
-                              ('eval_utils.py','normalize_compound_pairs')]:
-            tree=ast.parse((init.parent/filename).read_text())
-            definition=next(n for n in tree.body if isinstance(n,(ast.ClassDef,ast.FunctionDef)) and n.name==name)
-            exec(compile(ast.Module(body=[definition],type_ignores=[]),filename,'exec'),namespace)
-        english=module.EnglishTextNormalizer()
-        multilingual=namespace['MultilingualNormalizer'](remove_diacritics=False)
-        for (split,model),(refs,hyps) in paired.items():
-            lang=result['sets'][split]['language']
-            normalize=english if lang=='en' else lambda text:multilingual(text,lang=lang)
-            refs,hyps=list(map(normalize,refs)),list(map(normalize,hyps))
-            if lang!='en':refs,hyps=namespace['normalize_compound_pairs'](refs,hyps)
-            batch=batch_error_rate([tuple(r.split()) for r in refs],[tuple(h.split()) for h in hyps],merge_compounds=True)
-            target=result['sets'][split]['models'][model]
-            assert batch['total']==target['errors'] and batch['ref_len']==target['words']
-            assert abs(100*batch['err_rate']-target['wer'])<1e-12
-        audit['upstream_scoring_verified']=dict(revision=provenance['revision'],model_partition_pairs=len(paired),
-            check=f'Unmodified upstream normalizer definitions and full-partition batch scoring reproduce all {len(paired)} paired WERs.')
-        audit['checks'].append(f'All {len(paired)} model/partition WERs match independent upstream batch scoring')
-    (a.evidence/'hypotheses-audit.json').write_text(json.dumps(audit,indent=2)+'\n')
-    print(json.dumps(audit))
-
-
-if __name__=='__main__':main()

evaluation/standard_asr/build_current_report.py

@@ -1,179 +0,0 @@
-"""Bind the short report to one checkpoint and reproduce every aggregate from counts."""
-from collections import Counter, defaultdict
-import csv
-import gzip
-import hashlib
-import json
-from pathlib import Path
-import subprocess
-import sys
-import unicodedata
-
-ROOT = Path(__file__).resolve().parents[2]
-LANGUAGES = dict(bg='Bulgarian', cs='Czech', da='Danish', de='German', el='Greek', en='English',
-    es='Spanish', et='Estonian', fi='Finnish', fr='French', hr='Croatian', hu='Hungarian',
-    it='Italian', lt='Lithuanian', lv='Latvian', mt='Maltese', nl='Dutch', pl='Polish',
-    pt='Portuguese', ro='Romanian', ru='Russian', sk='Slovak', sl='Slovenian', sv='Swedish', uk='Ukrainian')
-MODEL = '031c8ddab4845aeced904a7cde8e8aa57993b2e344716cf83a545b079c473b56'
-BASE = '3cbdc85877e668ca7b82d0d56770eb1fac76691f55d6b97545e8d61ca588d10d'
-MODELS = ['parakeet', 'orukeet']
-COUNTS = ['errors','words','substitutions','deletions','insertions','chars','char_errors','utterance_error']
-
-def sha(path): return hashlib.sha256(path.read_bytes()).hexdigest()
-def read(path): return json.loads(path.read_text())
-def write(path, value): path.write_text(json.dumps(value,indent=2,ensure_ascii=False)+'\n')
-
-def validate(folder, nrows, nsets):
-    d=ROOT/'evidence'/folder; result=read(d/'comparison.json');audit=read(d/'hypotheses-audit.json')
-    assert result['status']=='complete' and result['rows']==nrows and len(result['sets'])==nsets
-    assert result['models']==dict(parakeet=BASE,orukeet=MODEL)
-    assert result['timings']['parakeet']['rows']==result['timings']['orukeet']['rows']==nrows
-    assert audit['status']=='passed' and audit['models']==result['models']
-    assert audit['upstream_scoring_verified']['model_partition_pairs']==2*nsets
-    assert result['numeric_evidence_sha256']==sha(d/'numeric-evidence.jsonl.gz')
-    for name,digest in result['scoring']['code_sha256'].items(): assert sha(ROOT/name)==digest,name
-    for name,digest in audit['inputs_sha256'].items(): assert sha(d/name)==digest,name
-    seen=set();counts=defaultdict(Counter);sizes=Counter()
-    with gzip.open(d/'numeric-evidence.jsonl.gz','rt') as f:
-        for line in f:
-            row=json.loads(line);uid=row['record_sha256'];assert uid not in seen;seen.add(uid)
-            sizes[row['split']]+=1
-            for model in MODELS:
-                c=row['counts'][model];assert c['errors']==c['substitutions']+c['deletions']+c['insertions']
-                counts[(row['split'],model)].update({k:c[k] for k in COUNTS})
-    assert len(seen)==nrows
-    for split,spec in result['sets'].items():
-        assert sizes[split]==spec['rows']
-        for model in MODELS:
-            c=counts[(split,model)];reported=spec['models'][model]
-            assert all(reported[k]==v for k,v in c.items())
-            for metric,num,den in [('wer','errors','words'),('cer','char_errors','chars')]:
-                assert abs(reported[metric]-100*c[num]/c[den])<1e-12
-    return result
-
-def pool(result, splits):
-    members=[result['sets'][s] for s in splits];models={}
-    for model in MODELS:
-        c=Counter()
-        for spec in members: c.update({k:spec['models'][model][k] for k in COUNTS})
-        models[model]=dict(c,wer=100*c['errors']/c['words'],cer=100*c['char_errors']/c['chars'])
-    wins=sum(s['models']['orukeet']['wer']<s['models']['parakeet']['wer'] for s in members)
-    ties=sum(s['models']['orukeet']['wer']==s['models']['parakeet']['wer'] for s in members)
-    return dict(splits=list(splits),rows=sum(s['rows'] for s in members),models=models,wins=wins,ties=ties,
-                losses=len(members)-wins-ties,relative_wer_reduction_percent=100*(1-models['orukeet']['wer']/models['parakeet']['wer']))
-
-def label(split):
-    if split.startswith('librispeech_'):return 'LibriSpeech '+split.removeprefix('librispeech_').replace('_','-')
-    if split.startswith('fleurs_'):return 'FLEURS '+LANGUAGES[split.rsplit('_',1)[1]]
-    gsb=dict(agr='agriculture',ait='AI',art='arts',bio='biology',chn='Chinese accent',ecm='economics',
-        eng='engineering',ent='entertainment',fin='finance',hum='humanities',ind='Indian accent',jpn='Japanese accent',law='law',
-        med='medicine',mil='military',phl='Filipino accent',sct='Scottish accent',sgp='Singaporean accent')
-    if split.startswith('gigaspeechbench_'):return 'GSB '+gsb[split.split('_')[1]]
-    if split.startswith('eurospeech_'):return 'EuroSpeech '+split.rsplit('_',1)[1].upper()
-    if split.startswith('voxpopuli_'):return 'VoxPopuli '+split.rsplit('_',1)[1].upper()
-    return dict(golos_crowd_ru='Golos crowd RU',golos_farfield_ru='Golos far-field RU',
-        nst_da_da='NST Danish',nst_sv_sv='NST Swedish',monsoon_en_in='Monsoon India',lesbos_el='Lesbos Greek')[split]
-
-def wer_tex(spec,model):
-    value=f"{spec['models'][model]['wer']:.2f}"
-    other=MODELS[1-MODELS.index(model)]
-    if spec['models'][model]['wer']<spec['models'][other]['wer']:value=r'\textbf{'+value+'}'
-    return value
-
-def main():
-    standard=validate('standard-asr-r3-20260908',25705,27)
-    domains=validate('domains-r3-20260908',12006,47)
-    expected={'librispeech_test_clean','librispeech_test_other'}|{'fleurs_'+l for l in LANGUAGES}
-    assert set(standard['sets'])==expected
-    assert all(s['rows']==(230 if name=='lesbos_el' else 256) for name,s in domains['sets'].items())
-    freeze=read(ROOT/'evidence/librispeech-ft-20260908/r3/export-audit.json')
-    assert freeze['status']=='pass' and freeze['candidate_sha256']==MODEL
-    assert freeze['frozen_gabor_rows_exact']==12288 and freeze['gabor_values_match_parent_and_original_functions']
-    summaries={'fleurs':pool(standard,sorted(s for s in expected if s.startswith('fleurs_'))),
-               'standard':pool(standard,sorted(expected)),
-               'domains':pool(domains,sorted(domains['sets'])),
-               'domain_english':pool(domains,sorted(s for s,v in domains['sets'].items() if v['language']=='en'))}
-    assert summaries['fleurs']['rows']==20146 and summaries['domain_english']['rows']==5120
-    macros={}
-    for prefix,source,split in [('LibriClean',standard,'librispeech_test_clean'),('LibriOther',standard,'librispeech_test_other'),('FleursEnglish',standard,'fleurs_en')]:
-        for model in MODELS:macros[prefix+model.title()+'WER']=f"{source['sets'][split]['models'][model]['wer']:.2f}"
-    for prefix,key in [('FleursPooled','fleurs'),('DomainPooled','domains'),('DomainEnglish','domain_english')]:
-        for model in MODELS:macros[prefix+model.title()+'WER']=f"{summaries[key]['models'][model]['wer']:.2f}"
-    for model in MODELS:
-        macros['FleursMacro'+model.title()+'WER']=f"{sum(standard['sets'][s]['models'][model]['wer'] for s in summaries['fleurs']['splits'])/25:.2f}"
-    for name,key in [('FleursWins','fleurs'),('StandardWins','standard'),('DomainWins','domains'),('DomainEnglishWins','domain_english')]:macros[name]=str(summaries[key]['wins'])
-    assert set(standard['sets']).isdisjoint(domains['sets'])
-    macros['TestedWins']=str(summaries['standard']['wins']+summaries['domains']['wins'])
-    macros['TestedSplits']=str(len(standard['sets'])+len(domains['sets']))
-    macros['FleursPooledReduction']=f"{summaries['fleurs']['relative_wer_reduction_percent']:.1f}"
-    (ROOT/'report/current-benchmark-values.tex').write_text('% Generated from audited r3 edit counts.\n'+''.join('\\newcommand{\\'+k+'}{'+v+'}\n' for k,v in macros.items()))
-    order=['librispeech_test_clean','librispeech_test_other']+sorted((s for s in expected if s.startswith('fleurs_')),key=label)
-    table=[r'\begin{table}[!ht]',r'\centering',r'\caption{Complete LibriSpeech and FLEURS test partitions. WER and CER are percentages; bold identifies lower WER. The pooled FLEURS row sums errors and reference words over all 25 languages, including English. The macro row weights languages equally.}',r'\label{tab:current-standard}',r'\small',r'\setlength{\tabcolsep}{5pt}',r'\renewcommand{\arraystretch}{1.02}',r'\begin{tabular}{lrrrrr}',r'\toprule',r'Benchmark & Clips & \multicolumn{2}{c}{Parakeet} & \multicolumn{2}{c}{Orukeet}\\',r' & & WER & CER & WER & CER\\',r'\midrule']
-    table_rows=[]
-    for split in order:
-        spec=standard['sets'][split];table_rows.append((label(split),spec))
-    table_rows.append(('FLEURS pooled',summaries['fleurs']))
-    macro=dict(rows=20146,models={m:{metric:sum(standard['sets'][s]['models'][m][metric] for s in summaries['fleurs']['splits'])/25 for metric in ['wer','cer']} for m in MODELS})
-    table_rows.append(('FLEURS language macro',macro))
-    for name,spec in table_rows:
-        if name=='FLEURS pooled':table.append(r'\midrule')
-        values=[wer_tex(spec,m)+' & '+f"{spec['models'][m]['cer']:.2f}" for m in MODELS]
-        table.append(name+' & '+f"{spec['rows']:,}"+' & '+' & '.join(values)+r'\\')
-        if name=='LibriSpeech test-other':table.append(r'\midrule')
-    table += [r'\bottomrule',r'\end{tabular}',r'\end{table}']
-    (ROOT/'report/current-standard-table.tex').write_text('\n'.join(table)+'\n')
-    domain_order=sorted(domains['sets'],key=label)
-    table=[r'\begin{table}[!p]',r'\centering',r'\caption{WER (\%) on the fixed accent and domain sample. Each partition contains 256 recordings, except Lesbos (230). GSB denotes GigaSpeechBench; two-letter suffixes identify languages. Bold identifies lower WER. Both models use the same decoding and scoring protocol as Table~\ref{tab:current-standard}.}',r'\label{tab:current-domains}',r'\small']
-    for idx,subset in enumerate([domain_order[:24],domain_order[24:]]):
-        table += [r'\begin{minipage}[t]{0.49\linewidth}',r'\vspace{0pt}',r'\centering',r'\setlength{\tabcolsep}{3pt}',r'\renewcommand{\arraystretch}{1.10}',r'\begin{tabular}{lrr}',r'\toprule',r'Partition & Parakeet & Orukeet\\',r'\midrule']
-        for split in subset:
-            spec=domains['sets'][split]
-            table.append(label(split)+' & '+' & '.join(wer_tex(spec,m) for m in MODELS)+r'\\')
-        table += [r'\bottomrule',r'\end{tabular}',r'\end{minipage}'+(r'\hfill%' if idx==0 else '')]
-    table += [r'\par\vspace{12pt}',r'\begin{tabular}{lrrr}',r'\toprule',r'Pooled comparison & Clips & Parakeet & Orukeet\\',r'\midrule']
-    for name,key in [('All 47 partitions','domains'),('All 20 English partitions','domain_english')]:
-        spec=summaries[key];table.append(name+' & '+f"{spec['rows']:,}"+' & '+' & '.join(wer_tex(spec,m) for m in MODELS)+r'\\')
-    table += [r'\bottomrule',r'\end{tabular}',r'\end{table}']
-    (ROOT/'report/current-domains-table.tex').write_text('\n'.join(table)+'\n')
-    md=['# Orukeet r3: paired recognition scores','',f'All Orukeet scores refer to NeMo SHA-256 `{MODEL}`. Parakeet is `{BASE}`. Both systems were decoded afresh on identical audio with FP32 weights, BF16 autocast and greedy-batch TDT. Lower WER is better.','',
-        'Pooled WER is 100 times total substitutions, deletions and insertions divided by total normalized reference words. FLEURS pooling includes all 25 supported languages, including English. It is not an average of language WERs. Compound-boundary alignment can give each model a different reference-word denominator. CER uses normalized strings before compound alignment.','',
-        'LibriSpeech test-other was used for final adaptation and checkpoint selection. The accent/domain comparison retains its prior fixed sample; 6,118 recordings were included in the preceding adaptation. Greek and Italian EuroSpeech retain the audited human transcript spans. No records are dropped from either comparison.','']
-    for title,source,order,folder in [('Complete read-speech partitions',standard,order,'standard-asr-r3-20260908'),('Accent and domain sample',domains,domain_order,'domains-r3-20260908')]:
-        md += ['## '+title,'','| Partition | Clips | Parakeet WER / CER | Orukeet WER / CER |','|:--|--:|--:|--:|']
-        with (ROOT/'evidence'/folder/'scores.csv').open('w',newline='') as f:
-            writer=csv.writer(f);writer.writerow(['split','clips','hours']+[m+'_'+k for m in MODELS for k in ['wer','cer','errors','words']])
-            for s in order:
-                spec=source['sets'][s];md.append('| '+label(s)+' | '+str(spec['rows'])+' | '+' | '.join(f"{spec['models'][m]['wer']:.2f} / {spec['models'][m]['cer']:.2f}" for m in MODELS)+' |')
-                writer.writerow([s,spec['rows'],spec['hours']]+[spec['models'][m][k] for m in MODELS for k in ['wer','cer','errors','words']])
-        md += ['',f'[Full precision](../evidence/{folder}/scores.csv) · [Counts](../evidence/{folder}/numeric-evidence.jsonl.gz) · [Independent scoring audit](../evidence/{folder}/hypotheses-audit.json)','']
-    md += ['## Pooled comparisons','','| Comparison | Clips | Parakeet errors / words | WER | Orukeet errors / words | WER | Wins / partitions |','|:--|--:|--:|--:|--:|--:|--:|']
-    for name,key in [('FLEURS, 25 languages','fleurs'),('Accents/domains, 25 languages','domains'),('Accents/domains, English','domain_english')]:
-        spec=summaries[key];md.append('| '+name+' | '+str(spec['rows'])+' | '+' | '.join(f"{spec['models'][m]['errors']:,} / {spec['models'][m]['words']:,} | {spec['models'][m]['wer']:.2f}" for m in MODELS)+f" | {spec['wins']} / {len(spec['splits'])} |")
-    (ROOT/'docs/current-checkpoint-benchmarks.md').write_text('\n'.join(md)+'\n')
-    receipt=dict(status='passed',publication_authorized=False,models=standard['models'],summaries=summaries,
-                 checks=['Fresh matched decoding of both checkpoints','All 74 paired split results recomputed from per-record counts','Pooled WER recomputed from summed errors and reference words','Independent upstream scoring matches all 148 model/partition pairs'],
-                 inputs_sha256={str(p.relative_to(ROOT)):sha(p) for folder in ['standard-asr-r3-20260908','domains-r3-20260908'] for p in [ROOT/'evidence'/folder/f for f in ['comparison.json','hypotheses-audit.json','numeric-evidence.jsonl.gz']]})
-    if '--verify-pdf' in sys.argv:
-        text=unicodedata.normalize('NFKC',subprocess.check_output(['pdftotext','-layout',str(ROOT/'output/pdf/orukeet-technical-report.pdf'),'-'],text=True))
-        lines=[' '.join(l.split()) for l in text.splitlines()]
-        for name,spec in table_rows:
-            target=' '.join([name,f"{spec['rows']:,}"]+[f"{spec['models'][m][metric]:.2f}" for m in MODELS for metric in ['wer','cer']])
-            assert target in lines,target
-        for s in domain_order:
-            spec=domains['sets'][s];target=' '.join([label(s)]+[f"{spec['models'][m]['wer']:.2f}" for m in MODELS])
-            assert any(target in line for line in lines),target
-        abstract=text.split('Abstract',1)[1].split('A fixed structure',1)[0]
-        assert all(macros[k] in abstract for k in ['FleursPooledParakeetWER','FleursPooledOrukeetWER','LibriCleanParakeetWER','LibriCleanOrukeetWER','LibriOtherParakeetWER','LibriOtherOrukeetWER','FleursEnglishParakeetWER','FleursEnglishOrukeetWER'])
-        assert 'leaderboard' not in text.lower() and 'gabormer' not in text.lower()
-        # Ignore line-break hyphenation when checking the rendered claim and setup.
-        compact_abstract=''.join(abstract.split()).replace('-','')
-        assert '031c8ddab484' in text
-        for phrase in ['Final adaptation and checkpoint selection use LibriSpeech test-other.',
-                       f"Orukeet outperforms Parakeet on {macros['TestedWins']} out of {macros['TestedSplits']} tested splits"]:
-            assert ''.join(phrase.split()).replace('-','') in compact_abstract,phrase
-        receipt['checks'].append('Rendered abstract and all 74 benchmark rows match the current checkpoint evidence')
-        receipt['pdf_sha256']=sha(ROOT/'output/pdf/orukeet-technical-report.pdf')
-    write(ROOT/'report/current-benchmark-validation.json',receipt)
-    print(json.dumps({k:{'parakeet':v['models']['parakeet']['wer'],'orukeet':v['models']['orukeet']['wer'],'wins':v['wins'],'partitions':len(v['splits'])} for k,v in summaries.items()}))
-
-if __name__=='__main__':main()

evaluation/standard_asr/build_materials.py

@@ -1,149 +0,0 @@
-"""Reproduce complete-test scores from edit counts and generate report tables."""
-from collections import Counter, defaultdict
-import csv
-import gzip
-import hashlib
-import json
-import math
-from pathlib import Path
-import subprocess
-import sys
-
-ROOT = Path(__file__).resolve().parents[2]
-OUT = ROOT / 'evidence/standard-asr-20260908'
-LANGUAGES = dict(bg='Bulgarian', cs='Czech', da='Danish', de='German', el='Greek', en='English',
-                 es='Spanish', et='Estonian', fi='Finnish', fr='French', hr='Croatian', hu='Hungarian',
-                 it='Italian', lt='Lithuanian', lv='Latvian', mt='Maltese', nl='Dutch', pl='Polish',
-                 pt='Portuguese', ro='Romanian', ru='Russian', sk='Slovak', sl='Slovenian',
-                 sv='Swedish', uk='Ukrainian')
-
-
-def sha(path):
-    return hashlib.sha256(path.read_bytes()).hexdigest()
-
-
-def main():
-    source = json.loads((OUT/'comparison.json').read_text())
-    prep = json.loads((OUT/'preparation.json').read_text())
-    assert source['status'] == 'complete' and prep['status'] == 'prepared'
-    canonical = json.loads((ROOT/'release/model-stages.json').read_text())['canonical']['source_sha256']
-    assert source['models']['orukeet'] == canonical
-    assert source['models']['parakeet'] == '3cbdc85877e668ca7b82d0d56770eb1fac76691f55d6b97545e8d61ca588d10d'
-    assert source['manifest_sha256'] == prep['manifest_sha256']
-    assert source['script_sha256'] == sha(ROOT/'evaluation/standard_asr/run.py')
-    assert prep['script_sha256'] == sha(ROOT/'evaluation/standard_asr/prepare.py')
-    assert source['normalizer_sha256'] == sha(ROOT/'evaluation/unseen/metrics.py')
-    assert source['numeric_evidence_sha256'] == sha(OUT/'numeric-evidence.jsonl.gz')
-    for name, digest in source['scoring']['code_sha256'].items():
-        assert sha(ROOT/name) == digest, name
-    audit = json.loads((OUT/'hypotheses-audit.json').read_text())
-    assert audit['status'] == 'passed' and audit['models'] == source['models']
-    assert audit['upstream_scoring_verified']['model_partition_pairs'] == 54
-    assert audit['script_sha256'] == sha(ROOT/'evaluation/standard_asr/audit_predictions.py')
-    for name, digest in audit['inputs_sha256'].items():
-        assert sha(OUT/name) == digest, name
-    expected = {f'fleurs_{lang}' for lang in LANGUAGES} | {'librispeech_test_clean','librispeech_test_other'}
-    assert set(source['sets']) == set(prep['sources']) == expected
-    totals, sizes, durations, seen = defaultdict(Counter), Counter(), Counter(), set()
-    with gzip.open(OUT/'numeric-evidence.jsonl.gz', 'rt') as stream:
-        for line in stream:
-            row = json.loads(line)
-            assert row['record_sha256'] not in seen
-            seen.add(row['record_sha256'])
-            sizes[row['split']] += 1
-            durations[row['split']] += row['duration']
-            for model in ['parakeet','orukeet']:
-                c = row['counts'][model]
-                assert c['errors'] == c['substitutions'] + c['deletions'] + c['insertions']
-                totals[(row['split'], model)].update(c)
-            assert row['counts']['parakeet']['chars'] == row['counts']['orukeet']['chars']
-    assert len(seen) == source['rows'] == prep['rows']
-    for split, spec in source['sets'].items():
-        assert sizes[split] == spec['rows'] == prep['sources'][split]['rows']
-        assert math.isclose(durations[split]/3600, spec['hours'], abs_tol=1e-8)
-        for model in source['models']:
-            c = totals[(split, model)]
-            assert all(spec['models'][model][k] == v for k,v in c.items())
-            assert math.isclose(spec['models'][model]['wer'], 100*c['errors']/c['words'], abs_tol=1e-12)
-            assert math.isclose(spec['models'][model]['cer'], 100*c['char_errors']/c['chars'], abs_tol=1e-12)
-    summaries = {}
-    for name, languages in [('fleurs_five', 'de es fr it pt'.split()), ('fleurs_all', list(LANGUAGES))]:
-        members = [source['sets'][f'fleurs_{lang}'] for lang in languages]
-        summaries[name] = dict(languages=languages, rows=sum(s['rows'] for s in members),
-                               models={model:{metric:sum(s['models'][model][metric] for s in members)/len(members)
-                                              for metric in ['wer','cer']} for model in source['models']})
-    order = ['librispeech_test_clean','librispeech_test_other'] + [f'fleurs_{k}' for k in sorted(LANGUAGES,key=LANGUAGES.get)]
-    names = {'librispeech_test_clean':'LibriSpeech test-clean', 'librispeech_test_other':'LibriSpeech test-other',
-             **{f'fleurs_{k}':'FLEURS '+v for k,v in LANGUAGES.items()}}
-    rows = [dict(split=s, label=names[s], **source['sets'][s]) for s in order]
-    macros = {}
-    for name, split in [('LibriClean','librispeech_test_clean'),('LibriOther','librispeech_test_other'),('FleursEnglish','fleurs_en')]:
-        for model, prefix in [('parakeet','Parakeet'),('orukeet','Orukeet')]:
-            macros[name+prefix+'WER'] = f"{source['sets'][split]['models'][model]['wer']:.2f}"
-    for name, key in [('FleursFive','fleurs_five'),('FleursAll','fleurs_all')]:
-        for model, prefix in [('parakeet','Parakeet'),('orukeet','Orukeet')]:
-            macros[name+prefix+'WER'] = f"{summaries[key]['models'][model]['wer']:.2f}"
-    macros['StandardTestRows'] = f"{source['rows']:,}"
-    macros['FleursTestRows'] = f"{summaries['fleurs_all']['rows']:,}"
-    (ROOT/'report/standard-benchmark-values.tex').write_text('% Generated from complete-test edit counts.\n'+''.join(
-        '\\newcommand{\\'+key+'}{'+value+'}\n' for key,value in macros.items()))
-    table = [r'\begin{table}[!ht]',r'\centering',
-             r'\caption{Complete LibriSpeech and FLEURS test partitions. Both models use matched NeMo greedy decoding; WER and CER are percentages. Macro rows weight languages equally. Lower is better.}',
-             r'\label{tab:standard-benchmarks}',r'\small',r'\setlength{\tabcolsep}{5pt}',
-             r'\renewcommand{\arraystretch}{1.05}',r'\begin{tabular}{lrrrrr}',r'\toprule',
-             r'Benchmark & Clips & \multicolumn{2}{c}{Parakeet} & \multicolumn{2}{c}{Orukeet}\\',
-             r' & & WER & CER & WER & CER\\',r'\midrule']
-    md = ['| Benchmark | Clips | Parakeet WER / CER | Orukeet WER / CER |','|:--|--:|--:|--:|']
-    for row in rows:
-        values = [f"{row['models'][model][metric]:.2f}" for model in ['parakeet','orukeet'] for metric in ['wer','cer']]
-        table.append(row['label']+' & '+f"{row['rows']:,}"+' & '+' & '.join(values)+r'\\')
-        md.append('| '+row['label']+' | '+f"{row['rows']:,}"+' | '+values[0]+' / '+values[1]+' | '+values[2]+' / '+values[3]+' |')
-        if row['split'] == 'librispeech_test_other': table.append(r'\midrule')
-    table.append(r'\midrule')
-    for key, label in [('fleurs_five','FLEURS five-language macro'),('fleurs_all','FLEURS 25-language macro')]:
-        row=summaries[key]
-        values=[f"{row['models'][model][metric]:.2f}" for model in ['parakeet','orukeet'] for metric in ['wer','cer']]
-        table.append(label+' & '+f"{row['rows']:,}"+' & '+' & '.join(values)+r'\\')
-        md.append('| '+label+' | '+f"{row['rows']:,}"+' | '+values[0]+' / '+values[1]+' | '+values[2]+' / '+values[3]+' |')
-    table += [r'\bottomrule',r'\end{tabular}',r'\end{table}']
-    (ROOT/'report/standard-benchmark-table.tex').write_text('\n'.join(table)+'\n')
-    with (OUT/'scores.csv').open('w',newline='') as stream:
-        writer=csv.writer(stream);writer.writerow(['split','clips','hours','parakeet_wer','parakeet_cer','orukeet_wer','orukeet_cer'])
-        for row in rows:writer.writerow([row['split'],row['rows'],row['hours']]+[row['models'][m][x] for m in ['parakeet','orukeet'] for x in ['wer','cer']])
-    method = ('Both checkpoints use identical mono 16 kHz audio, NeMo greedy-batch TDT decoding, FP32 weights and BF16 CUDA autocast. '
-              'English uses the pinned English text normalizer. Multilingual normalization retains diacritics and expands numbers by language. '
-              'WER aligns compound boundaries, then uses compound-aware edit distance; CER measures normalized strings before boundary alignment. '
-              'WER and CER pool integer edit counts and reference lengths within each test partition. Every test record is retained, including empty hypotheses. '
-              'The five-language FLEURS macro averages German, Spanish, French, Italian and Portuguese; the 25-language macro includes every supported language.')
-    doc = '# LibriSpeech and FLEURS benchmarks\n\n'+method+'\n\n'+'\n'.join(md)+'\n\n'+(
-        '[Evaluation and reproduction](../evaluation/standard_asr/README.md) · '
-        '[Full-precision scores](../evidence/standard-asr-20260908/scores.csv) · '
-        '[Per-record edit counts](../evidence/standard-asr-20260908/numeric-evidence.jsonl.gz)\n')
-    (ROOT/'docs/standard-asr-benchmarks.md').write_text(doc)
-    receipt = dict(status='passed', publication_authorized=False, rows=len(seen), splits=len(rows),
-                   model_sha256=canonical, summaries=summaries, checks=['Unique complete test membership','All paired integer counts','WER and CER recomputation','Pinned checkpoint and source hashes'],
-                   inputs_sha256={name:sha(OUT/name) for name in ['comparison.json','preparation.json','numeric-evidence.jsonl.gz','hypotheses-audit.json']},
-                   outputs_sha256={name:sha(ROOT/name) for name in ['report/standard-benchmark-values.tex','report/standard-benchmark-table.tex','docs/standard-asr-benchmarks.md','evidence/standard-asr-20260908/scores.csv']})
-    if '--verify-pdf' in sys.argv:
-        text=subprocess.check_output(['pdftotext','-layout',str(ROOT/'output/pdf/orukeet-technical-report.pdf'),'-'],text=True)
-        lines={' '.join(line.split()) for line in text.splitlines()}
-        for row in rows:
-            values=[f"{row['models'][model][metric]:.2f}" for model in ['parakeet','orukeet'] for metric in ['wer','cer']]
-            assert ' '.join([row['label'],f"{row['rows']:,}"]+values) in lines, row['split']
-        normalized_lines={line.replace('\ufb01','fi') for line in lines}
-        for key,label in [('fleurs_five','FLEURS five-language macro'),('fleurs_all','FLEURS 25-language macro')]:
-            summary=summaries[key]
-            values=[f"{summary['models'][model][metric]:.2f}" for model in ['parakeet','orukeet'] for metric in ['wer','cer']]
-            assert ' '.join([label,f"{summary['rows']:,}"]+values) in normalized_lines,key
-        import re
-        abstract=re.split(r'\n\s*1\s+Introduction',text.split('Abstract',1)[1],maxsplit=1)[0]
-        assert 'Open ASR' not in text and 'leaderboard' not in text.lower()
-        for key in ['LibriClean','LibriOther','FleursEnglish','FleursFive']:
-            assert macros[key+'ParakeetWER'] in abstract and macros[key+'OrukeetWER'] in abstract
-        receipt['checks'].append('Rendered abstract and all 27 test table rows match measurements')
-    (OUT/'materials-validation.json').write_text(json.dumps(receipt,indent=2)+'\n')
-    print(json.dumps({k:v for k,v in receipt.items() if k not in ['inputs_sha256','outputs_sha256','summaries']}))
-
-
-if __name__ == '__main__':
-    main()

evaluation/standard_asr/prepare.py

@@ -1,99 +0,0 @@
-"""Pair pinned FLEURS references and official LibriSpeech transcripts with audio."""
-import argparse
-from concurrent.futures import ThreadPoolExecutor
-import hashlib
-import json
-from pathlib import Path
-
-FLEURS_REPO = 'hf-audio/open-asr-leaderboard-multilingual-datasets'
-FLEURS_REVISION = 'b791fc9151221e5b7e59c6c2dfa4dee09dda3cb7'
-LANGUAGES = 'bg cs da de el en es et fi fr hr hu it lt lv mt nl pl pt ro ru sk sl sv uk'.split()
-
-
-def sha(path):
-    return hashlib.sha256(path.read_bytes()).hexdigest()
-
-
-def main():
-    p = argparse.ArgumentParser()
-    p.add_argument('--audio-root', type=Path, required=True)
-    p.add_argument('--output', type=Path, required=True)
-    a = p.parse_args()
-    import numpy as np
-    import soundfile as sf
-    import pyarrow.parquet as pq
-    from huggingface_hub import HfFileSystem
-    a.output.mkdir(parents=True, exist_ok=True)
-
-    def fleurs(language):
-        path = f'datasets/{FLEURS_REPO}@{FLEURS_REVISION}/data/fleurs/{language}_test.parquet'
-        target = a.output / f'fleurs-{language}-references.json'
-        if target.exists():
-            return json.loads(target.read_text())
-        with HfFileSystem().open(path, 'rb', block_size=1024*1024) as stream:
-            parquet = pq.ParquetFile(stream)
-            records = parquet.read(columns=['file_name', 'duration', 'text']).to_pylist()
-            assert len(records) == parquet.metadata.num_rows
-        target.write_text(json.dumps(records, ensure_ascii=False) + '\n')
-        print('REFERENCES', language, len(records), flush=True)
-        return records
-
-    with ThreadPoolExecutor(max_workers=6) as pool:
-        reference_sets = dict(zip(LANGUAGES, pool.map(fleurs, LANGUAGES)))
-    rows, sources = [], {}
-
-    def add(split, language, uid, path, text, duration=None):
-        assert path.is_file(), path
-        audio, rate = sf.read(path, dtype='float32')
-        assert rate == 16000 and audio.ndim == 1 and np.isfinite(audio).all()
-        seconds = len(audio) / rate
-        if duration is not None:
-            assert abs(seconds - duration) < 0.002, (path, seconds, duration)
-        rows.append(dict(uid=uid, split=split, language=language, text=text,
-                         audio_filepath=str(path), duration=seconds,
-                         pcm_sha256=hashlib.sha256(audio.astype('<f4').tobytes()).hexdigest(),
-                         reference_sha256=hashlib.sha256(text.encode()).hexdigest()))
-
-    for language, records in reference_sets.items():
-        split = f'fleurs_{language}'
-        assert len(records) == len({r['file_name'] for r in records})
-        for row in records:
-            path = a.audio_root / 'data/fleurs' / language / 'test' / (Path(row['file_name']).stem + '.flac')
-            add(split, language, split+':'+Path(row['file_name']).stem, path, row['text'], row['duration'])
-        sources[split] = dict(repository=FLEURS_REPO, revision=FLEURS_REVISION,
-                             path=f'data/fleurs/{language}_test.parquet', rows=len(records),
-                             reference_file_sha256=sha(a.output/f'fleurs-{language}-references.json'))
-
-    for split, folder, expected in [
-        ('librispeech_test_clean', 'data/goal_v2_sources/libri/LibriSpeech/test-clean', 2620),
-        ('librispeech_test_other', 'data/librispeech/LibriSpeech/test-other', 2939),
-    ]:
-        transcripts = sorted((a.audio_root / folder).rglob('*.trans.txt'))
-        count = 0
-        for transcript in transcripts:
-            for line in transcript.read_text().splitlines():
-                uid, text = line.split(' ', 1)
-                add(split, 'en', split+':'+uid, transcript.parent/(uid+'.flac'), text)
-                count += 1
-        assert count == expected, (split, count)
-        sources[split] = dict(source='Official LibriSpeech test partition and original .trans.txt transcripts',
-                             rows=count, transcript_files_sha256={str(x.relative_to(a.audio_root)):sha(x) for x in transcripts})
-    rows.sort(key=lambda r: (r['split'], r['uid']))
-    assert len(rows) == len({r['uid'] for r in rows})
-    manifest = a.output/'manifest.jsonl'
-    with manifest.open('w') as stream:
-        for row in rows:
-            stream.write(json.dumps(row, ensure_ascii=False)+'\n')
-    receipt = dict(status='prepared', publication_authorized=False, rows=len(rows), splits=len(sources),
-                   selection='Complete published test partitions; all records retained, no selection by model output.',
-                   sources=sources, manifest_sha256=sha(manifest), script_sha256=sha(Path(__file__)),
-                   hours=sum(r['duration'] for r in rows)/3600,
-                   audio='Existing mono 16 kHz FLAC files, decoded float32 PCM hashes checked before inference.',
-                   abstract_comparisons=['librispeech_test_clean','librispeech_test_other','fleurs_en'],
-                   additional_abstract_summary='Equal-language macro over de, es, fr, it, pt FLEURS test partitions')
-    (a.output/'preparation.json').write_text(json.dumps(receipt, indent=2)+'\n')
-    print('PREPARED', len(rows), receipt['hours'], flush=True)
-
-
-if __name__ == '__main__':
-    main()

evaluation/standard_asr/requirements-score.txt

@@ -1,4 +0,0 @@
-kaldialign==0.12.0
-num2words==0.5.14
-rapidfuzz==3.14.6
-regex==2026.9.3

evaluation/standard_asr/rescore.py

@@ -1,70 +0,0 @@
-"""Score fixed private transcripts with pinned compound-aware normalization."""
-import argparse
-from collections import Counter, defaultdict
-import gzip
-import hashlib
-from importlib.metadata import version
-import json
-from pathlib import Path
-from scoring import counts
-
-ROOT = Path(__file__).resolve().parents[2]
-
-
-def sha(path):
-    return hashlib.sha256(path.read_bytes()).hexdigest()
-
-
-def main():
-    parser = argparse.ArgumentParser()
-    parser.add_argument('--private-records', type=Path, required=True)
-    parser.add_argument('--evidence', type=Path, default=ROOT/'evidence/standard-asr-20260908')
-    a = parser.parse_args()
-    source = json.loads((a.evidence/'inference-comparison.json').read_text())
-    assert sha(a.private_records/'manifest.jsonl') == source['manifest_sha256']
-    rows = [json.loads(line) for line in (a.private_records/'manifest.jsonl').open()]
-    predictions = {}
-    for model in source['models']:
-        path = a.private_records/(model+'.jsonl')
-        assert sha(path) == source['prediction_sha256'][model]
-        records = [json.loads(line) for line in path.open()]
-        predictions[model] = {r['uid']:r['prediction'] for r in records}
-        assert len(records) == len(predictions[model]) == len(rows)
-    totals, seen = defaultdict(Counter), set()
-    with gzip.open(a.evidence/'numeric-evidence.jsonl.gz', 'wt') as stream:
-        for row in rows:
-            assert row['uid'] not in seen
-            seen.add(row['uid'])
-            values = {model:counts(row['text'], predictions[model][row['uid']], row['language'])
-                      for model in source['models']}
-            for model, value in values.items():
-                totals[(row['split'], model)].update(value)
-            stream.write(json.dumps(dict(record_sha256=hashlib.sha256(row['uid'].encode()).hexdigest(),
-                                         split=row['split'], language=row['language'],
-                                         duration=row['duration'], counts=values))+'\n')
-    assert len(seen) == source['rows']
-    for split, spec in source['sets'].items():
-        spec['models'] = {model:dict(totals[(split,model)],
-                                    wer=100*totals[(split,model)]['errors']/totals[(split,model)]['words'],
-                                    cer=100*totals[(split,model)]['char_errors']/totals[(split,model)]['chars'])
-                          for model in source['models']}
-    paths = [Path(__file__), ROOT/'evaluation/standard_asr/scoring.py']
-    paths += [p for p in (ROOT/'evaluation/standard_asr/vendor').iterdir() if p.is_file()]
-    source['scoring'] = dict(
-        protocol='Pinned English and multilingual normalization; compound-aware WER',
-        dependencies={name:version(name) for name in ['kaldialign','num2words','regex','rapidfuzz']},
-        code_sha256={p.relative_to(ROOT).as_posix():sha(p) for p in paths},
-        reference_word_counts='Non-English word-boundary alignment may change reference token counts separately for each model.',
-        cer='Character distance before compound alignment, including normalized spaces.',
-        inference_comparison_sha256=sha(a.evidence/'inference-comparison.json'))
-    source['normalizers'] = dict(en='Pinned EnglishTextNormalizer with spelling/name/compound maps',
-                                other='Pinned MultilingualNormalizer; diacritics retained; language-specific numbers; compound-boundary alignment')
-    source['numeric_evidence_sha256'] = sha(a.evidence/'numeric-evidence.jsonl.gz')
-    (a.evidence/'comparison.json').write_text(json.dumps(source,indent=2,ensure_ascii=False)+'\n')
-    for split in ['librispeech_test_clean','librispeech_test_other','fleurs_en','fleurs_de','fleurs_es','fleurs_fr','fleurs_it','fleurs_pt']:
-        if split in source['sets']:
-            print(split,{m:round(v['wer'],3) for m,v in source['sets'][split]['models'].items()})
-
-
-if __name__ == '__main__':
-    main()

evaluation/standard_asr/run.py

@@ -1,158 +0,0 @@
-"""Compare two fixed NeMo checkpoints on complete LibriSpeech and FLEURS tests.
-
-Input is a JSONL manifest with uid, split, language, text, audio_filepath,
-duration, and pcm_sha256. Every record is decoded by both checkpoints; empty
-hypotheses remain in the score. Resume verifies model, manifest and code hashes.
-"""
-import argparse
-from collections import Counter, defaultdict
-import gc
-import gzip
-import hashlib
-import json
-from pathlib import Path
-import sys
-import time
-
-
-def sha(path):
-    h = hashlib.sha256()
-    with open(path, 'rb') as stream:
-        for block in iter(lambda: stream.read(1 << 20), b''):
-            h.update(block)
-    return h.hexdigest()
-
-
-def write(path, value):
-    tmp = path.with_suffix('.tmp')
-    tmp.write_text(json.dumps(value, indent=2, ensure_ascii=False) + '\n')
-    tmp.replace(path)
-
-
-def main():
-    p = argparse.ArgumentParser()
-    p.add_argument('--manifest', type=Path, required=True)
-    p.add_argument('--parakeet', type=Path, required=True)
-    p.add_argument('--orukeet', type=Path, required=True)
-    p.add_argument('--parakeet-sha256', default='3cbdc85877e668ca7b82d0d56770eb1fac76691f55d6b97545e8d61ca588d10d')
-    p.add_argument('--orukeet-sha256', default='0ccfefcd1894871cb0850bd3c464adf5397752840de2a76d1d2d075c4141a945')
-    p.add_argument('--metric-code', type=Path, required=True)
-    p.add_argument('--output', type=Path, required=True)
-    a = p.parse_args()
-    import numpy as np
-    import soundfile as sf
-    import torch
-    from omegaconf import OmegaConf
-    from nemo.collections.asr.models import ASRModel
-    from nemo.utils import logging
-    sys.path.insert(0, str(a.metric_code))
-    from metrics import counts, normalize, ENGLISH
-
-    logging.set_verbosity(logging.ERROR)
-    torch.set_num_threads(8)
-    torch.manual_seed(20260908)
-    np.random.seed(20260908)
-    torch.backends.cuda.matmul.allow_tf32 = False
-    torch.backends.cudnn.allow_tf32 = True
-    a.output.mkdir(parents=True, exist_ok=True)
-    rows = [json.loads(line) for line in a.manifest.open()]
-    assert len(rows) == len({r['uid'] for r in rows})
-    rows.sort(key=lambda r: (r['duration'], r['uid']))
-    identities = dict(manifest_sha256=sha(a.manifest), script_sha256=sha(__file__),
-                      normalizer_sha256=sha(a.metric_code / 'metrics.py'))
-    for row in rows:
-        audio, rate = sf.read(row['audio_filepath'], dtype='float32')
-        assert rate == 16000 and audio.ndim == 1
-        assert abs(len(audio) / rate - row['duration']) < 1 / rate
-        assert hashlib.sha256(audio.astype('<f4').tobytes()).hexdigest() == row['pcm_sha256']
-    print('AUDIO_VERIFIED', len(rows), flush=True)
-    decoding, predictions, timings, models = None, {}, {}, {}
-    for label, path in [('parakeet', a.parakeet), ('orukeet', a.orukeet)]:
-        identity = dict(identities, model_sha256=sha(path))
-        assert identity['model_sha256'] == getattr(a, label + '_sha256')
-        models[label] = identity['model_sha256']
-        target = a.output / (label + '.jsonl')
-        prior = [json.loads(line) for line in target.open()] if target.exists() else []
-        assert len(prior) == len({r['uid'] for r in prior})
-        for record in prior:
-            assert all(record[k] == v for k, v in identity.items())
-        done = {r['uid']: r for r in prior}
-        assert set(done) <= {r['uid'] for r in rows}
-        pending = [r for r in rows if r['uid'] not in done]
-        if pending:
-            model = ASRModel.restore_from(str(path), map_location='cuda').eval()
-            model.freeze()
-            config = OmegaConf.to_container(model.cfg.decoding, resolve=True)
-            assert config['strategy'] == 'greedy_batch' and config['greedy']['max_symbols'] == 10
-            if decoding is None:
-                decoding = config
-                write(a.output / 'decoding.json', config)
-            assert config == decoding
-            model.change_decoding_strategy(OmegaConf.create(config))
-            started = time.monotonic()
-
-            def infer(batch):
-                try:
-                    with torch.inference_mode(), torch.autocast('cuda', dtype=torch.bfloat16):
-                        hypotheses = model.transcribe([r['audio_filepath'] for r in batch],
-                                                      batch_size=16, num_workers=4, verbose=False)
-                    assert len(hypotheses) == len(batch)
-                    return [h.text if hasattr(h, 'text') else str(h) for h in hypotheses]
-                except torch.cuda.OutOfMemoryError:
-                    torch.cuda.empty_cache()
-                    gc.collect()
-                    if len(batch) == 1:
-                        raise
-                    mid = len(batch) // 2
-                    return infer(batch[:mid]) + infer(batch[mid:])
-
-            with target.open('a') as stream:
-                for offset in range(0, len(pending), 256):
-                    batch = pending[offset:offset + 256]
-                    for row, text in zip(batch, infer(batch)):
-                        record = dict(uid=row['uid'], prediction=text, **identity)
-                        done[row['uid']] = record
-                        stream.write(json.dumps(record, ensure_ascii=False) + '\n')
-                    stream.flush()
-                    print('EVAL', label, len(done), '/', len(rows), flush=True)
-            torch.cuda.synchronize()
-            timings[label] = dict(seconds=time.monotonic() - started, rows=len(pending))
-            del model
-            gc.collect()
-            torch.cuda.empty_cache()
-        predictions[label] = done
-        assert len(done) == len(rows)
-    totals, sizes, durations = defaultdict(lambda: defaultdict(Counter)), Counter(), Counter()
-    with gzip.open(a.output / 'numeric-evidence.jsonl.gz', 'wt') as stream:
-        for row in rows:
-            split = row['split']
-            sizes[split] += 1
-            durations[split] += row['duration']
-            normalizer = ENGLISH if row['language'] == 'en' else normalize
-            values = {label: counts(row['text'], predictions[label][row['uid']]['prediction'], normalizer)
-                      for label in predictions}
-            for label, count in values.items():
-                totals[split][label].update(count)
-            stream.write(json.dumps(dict(record_sha256=hashlib.sha256(row['uid'].encode()).hexdigest(),
-                                         split=split, language=row['language'],
-                                         duration=row['duration'], counts=values)) + '\n')
-    sets = {}
-    for split in sorted(sizes):
-        sets[split] = dict(rows=sizes[split], hours=durations[split] / 3600,
-                           language=next(r['language'] for r in rows if r['split'] == split),
-                           models={label: dict(c, wer=100*c['errors']/c['words'],
-                                               cer=100*c['char_errors']/c['chars'])
-                                   for label, c in totals[split].items()})
-    result = dict(status='complete', publication_authorized=False, rows=len(rows), sets=sets,
-                  models=models, **identities, decoding=json.loads((a.output/'decoding.json').read_text()),
-                  precision='FP32 weights, BF16 CUDA autocast, TF32 matrix multiplication disabled',
-                  normalizers={'en':'Whisper EnglishTextNormalizer', 'other':'Recorded multilingual normalizer'},
-                  empty_output_handling='Every manifest record is scored; empty hypotheses remain.',
-                  timings=timings, prediction_sha256={k:sha(a.output/(k+'.jsonl')) for k in models},
-                  numeric_evidence_sha256=sha(a.output/'numeric-evidence.jsonl.gz'))
-    write(a.output / 'comparison.json', result)
-    print('COMPLETE', len(rows), flush=True)
-
-
-if __name__ == '__main__':
-    main()

evaluation/standard_asr/scoring.py

@@ -1,27 +0,0 @@
-"""Pinned English/multilingual normalization and compound-aware WER."""
-from kaldialign import batch_error_rate
-from rapidfuzz.distance import Levenshtein
-from vendor.normalizer import EnglishTextNormalizer
-from vendor.multilingual import MultilingualNormalizer, normalize_compound_pairs
-
-ENGLISH = EnglishTextNormalizer()
-MULTILINGUAL = MultilingualNormalizer(remove_diacritics=False)
-
-
-def normalized_pair(reference, prediction, language):
-    normalizer = ENGLISH if language == 'en' else lambda text: MULTILINGUAL(text, lang=language)
-    return normalizer(reference), normalizer(prediction)
-
-
-def counts(reference, prediction, language):
-    ref, hyp = normalized_pair(reference, prediction, language)
-    # CER uses the normalized strings before pair-dependent word-boundary changes.
-    chars, char_errors = len(ref), Levenshtein.distance(ref, hyp)
-    if language != 'en':
-        refs, hyps = normalize_compound_pairs([ref], [hyp])
-        ref, hyp = refs[0], hyps[0]
-    result = batch_error_rate([tuple(ref.split())], [tuple(hyp.split())], merge_compounds=True)
-    return dict(words=len(ref.split()), errors=result['ins'] + result['del'] + result['sub'],
-                substitutions=result['sub'], deletions=result['del'], insertions=result['ins'],
-                chars=chars, char_errors=char_errors,
-                utterance_error=int(bool(result['ins'] + result['del'] + result['sub'])))

evaluation/standard_asr/vendor/LICENSE

@@ -1,201 +0,0 @@
-                                 Apache License
-                           Version 2.0, January 2004
-                        http://www.apache.org/licenses/
-
-   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
-
-   1. Definitions.
-
-      "License" shall mean the terms and conditions for use, reproduction,
-      and distribution as defined by Sections 1 through 9 of this document.
-
-      "Licensor" shall mean the copyright owner or entity authorized by
-      the copyright owner that is granting the License.
-
-      "Legal Entity" shall mean the union of the acting entity and all
-      other entities that control, are controlled by, or are under common
-      control with that entity. For the purposes of this definition,
-      "control" means (i) the power, direct or indirect, to cause the
-      direction or management of such entity, whether by contract or
-      otherwise, or (ii) ownership of fifty percent (50%) or more of the
-      outstanding shares, or (iii) beneficial ownership of such entity.
-
-      "You" (or "Your") shall mean an individual or Legal Entity
-      exercising permissions granted by this License.
-
-      "Source" form shall mean the preferred form for making modifications,
-      including but not limited to software source code, documentation
-      source, and configuration files.
-
-      "Object" form shall mean any form resulting from mechanical
-      transformation or translation of a Source form, including but
-      not limited to compiled object code, generated documentation,
-      and conversions to other media types.
-
-      "Work" shall mean the work of authorship, whether in Source or
-      Object form, made available under the License, as indicated by a
-      copyright notice that is included in or attached to the work
-      (an example is provided in the Appendix below).
-
-      "Derivative Works" shall mean any work, whether in Source or Object
-      form, that is based on (or derived from) the Work and for which the
-      editorial revisions, annotations, elaborations, or other modifications
-      represent, as a whole, an original work of authorship. For the purposes
-      of this License, Derivative Works shall not include works that remain
-      separable from, or merely link (or bind by name) to the interfaces of,
-      the Work and Derivative Works thereof.
-
-      "Contribution" shall mean any work of authorship, including
-      the original version of the Work and any modifications or additions
-      to that Work or Derivative Works thereof, that is intentionally
-      submitted to Licensor for inclusion in the Work by the copyright owner
-      or by an individual or Legal Entity authorized to submit on behalf of
-      the copyright owner. For the purposes of this definition, "submitted"
-      means any form of electronic, verbal, or written communication sent
-      to the Licensor or its representatives, including but not limited to
-      communication on electronic mailing lists, source code control systems,
-      and issue tracking systems that are managed by, or on behalf of, the
-      Licensor for the purpose of discussing and improving the Work, but
-      excluding communication that is conspicuously marked or otherwise
-      designated in writing by the copyright owner as "Not a Contribution."
-
-      "Contributor" shall mean Licensor and any individual or Legal Entity
-      on behalf of whom a Contribution has been received by Licensor and
-      subsequently incorporated within the Work.
-
-   2. Grant of Copyright License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      copyright license to reproduce, prepare Derivative Works of,
-      publicly display, publicly perform, sublicense, and distribute the
-      Work and such Derivative Works in Source or Object form.
-
-   3. Grant of Patent License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      (except as stated in this section) patent license to make, have made,
-      use, offer to sell, sell, import, and otherwise transfer the Work,
-      where such license applies only to those patent claims licensable
-      by such Contributor that are necessarily infringed by their
-      Contribution(s) alone or by combination of their Contribution(s)
-      with the Work to which such Contribution(s) was submitted. If You
-      institute patent litigation against any entity (including a
-      cross-claim or counterclaim in a lawsuit) alleging that the Work
-      or a Contribution incorporated within the Work constitutes direct
-      or contributory patent infringement, then any patent licenses
-      granted to You under this License for that Work shall terminate
-      as of the date such litigation is filed.
-
-   4. Redistribution. You may reproduce and distribute copies of the
-      Work or Derivative Works thereof in any medium, with or without
-      modifications, and in Source or Object form, provided that You
-      meet the following conditions:
-
-      (a) You must give any other recipients of the Work or
-          Derivative Works a copy of this License; and
-
-      (b) You must cause any modified files to carry prominent notices
-          stating that You changed the files; and
-
-      (c) You must retain, in the Source form of any Derivative Works
-          that You distribute, all copyright, patent, trademark, and
-          attribution notices from the Source form of the Work,
-          excluding those notices that do not pertain to any part of
-          the Derivative Works; and
-
-      (d) If the Work includes a "NOTICE" text file as part of its
-          distribution, then any Derivative Works that You distribute must
-          include a readable copy of the attribution notices contained
-          within such NOTICE file, excluding those notices that do not
-          pertain to any part of the Derivative Works, in at least one
-          of the following places: within a NOTICE text file distributed
-          as part of the Derivative Works; within the Source form or
-          documentation, if provided along with the Derivative Works; or,
-          within a display generated by the Derivative Works, if and
-          wherever such third-party notices normally appear. The contents
-          of the NOTICE file are for informational purposes only and
-          do not modify the License. You may add Your own attribution
-          notices within Derivative Works that You distribute, alongside
-          or as an addendum to the NOTICE text from the Work, provided
-          that such additional attribution notices cannot be construed
-          as modifying the License.
-
-      You may add Your own copyright statement to Your modifications and
-      may provide additional or different license terms and conditions
-      for use, reproduction, or distribution of Your modifications, or
-      for any such Derivative Works as a whole, provided Your use,
-      reproduction, and distribution of the Work otherwise complies with
-      the conditions stated in this License.
-
-   5. Submission of Contributions. Unless You explicitly state otherwise,
-      any Contribution intentionally submitted for inclusion in the Work
-      by You to the Licensor shall be under the terms and conditions of
-      this License, without any additional terms or conditions.
-      Notwithstanding the above, nothing herein shall supersede or modify
-      the terms of any separate license agreement you may have executed
-      with Licensor regarding such Contributions.
-
-   6. Trademarks. This License does not grant permission to use the trade
-      names, trademarks, service marks, or product names of the Licensor,
-      except as required for reasonable and customary use in describing the
-      origin of the Work and reproducing the content of the NOTICE file.
-
-   7. Disclaimer of Warranty. Unless required by applicable law or
-      agreed to in writing, Licensor provides the Work (and each
-      Contributor provides its Contributions) on an "AS IS" BASIS,
-      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
-      implied, including, without limitation, any warranties or conditions
-      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
-      PARTICULAR PURPOSE. You are solely responsible for determining the
-      appropriateness of using or redistributing the Work and assume any
-      risks associated with Your exercise of permissions under this License.
-
-   8. Limitation of Liability. In no event and under no legal theory,
-      whether in tort (including negligence), contract, or otherwise,
-      unless required by applicable law (such as deliberate and grossly
-      negligent acts) or agreed to in writing, shall any Contributor be
-      liable to You for damages, including any direct, indirect, special,
-      incidental, or consequential damages of any character arising as a
-      result of this License or out of the use or inability to use the
-      Work (including but not limited to damages for loss of goodwill,
-      work stoppage, computer failure or malfunction, or any and all
-      other commercial damages or losses), even if such Contributor
-      has been advised of the possibility of such damages.
-
-   9. Accepting Warranty or Additional Liability. While redistributing
-      the Work or Derivative Works thereof, You may choose to offer,
-      and charge a fee for, acceptance of support, warranty, indemnity,
-      or other liability obligations and/or rights consistent with this
-      License. However, in accepting such obligations, You may act only
-      on Your own behalf and on Your sole responsibility, not on behalf
-      of any other Contributor, and only if You agree to indemnify,
-      defend, and hold each Contributor harmless for any liability
-      incurred by, or claims asserted against, such Contributor by reason
-      of your accepting any such warranty or additional liability.
-
-   END OF TERMS AND CONDITIONS
-
-   APPENDIX: How to apply the Apache License to your work.
-
-      To apply the Apache License to your work, attach the following
-      boilerplate notice, with the fields enclosed by brackets "[]"
-      replaced with your own identifying information. (Don't include
-      the brackets!)  The text should be enclosed in the appropriate
-      comment syntax for the file format. We also recommend that a
-      file or class name and description of purpose be included on the
-      same "printed page" as the copyright notice for easier
-      identification within third-party archives.
-
-   Copyright [2024] [The HuggingFace Inc. team]
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-       http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License.

evaluation/standard_asr/vendor/__init__.py

@@ -1 +0,0 @@
-"""Pinned text-scoring dependencies; see provenance.json and LICENSE."""

evaluation/standard_asr/vendor/english_abbreviations.py

@@ -1,1934 +0,0 @@
-english_spelling_normalizer = {
-  "ok": "okay",
-  "kay": "okay",
-  "etcetera": "etc",
-  "accessorise": "accessorize",
-  "accessorised": "accessorized",
-  "accessorises": "accessorizes",
-  "accessorising": "accessorizing",
-  "acclimatisation": "acclimatization",
-  "acclimatise": "acclimatize",
-  "acclimatised": "acclimatized",
-  "acclimatises": "acclimatizes",
-  "acclimatising": "acclimatizing",
-  "accoutrements": "accouterments",
-  "aeon": "eon",
-  "aeons": "eons",
-  "aerogramme": "aerogram",
-  "aerogrammes": "aerograms",
-  "aeroplane": "airplane",
-  "aeroplanes": "airplanes",
-  "aesthete": "esthete",
-  "aesthetes": "esthetes",
-  "aesthetic": "esthetic",
-  "aesthetically": "esthetically",
-  "aesthetics": "esthetics",
-  "aetiology": "etiology",
-  "ageing": "aging",
-  "aggrandisement": "aggrandizement",
-  "agonise": "agonize",
-  "agonised": "agonized",
-  "agonises": "agonizes",
-  "agonising": "agonizing",
-  "agonisingly": "agonizingly",
-  "almanack": "almanac",
-  "almanacks": "almanacs",
-  "aluminium": "aluminum",
-  "amortisable": "amortizable",
-  "amortisation": "amortization",
-  "amortisations": "amortizations",
-  "amortise": "amortize",
-  "amortised": "amortized",
-  "amortises": "amortizes",
-  "amortising": "amortizing",
-  "amphitheatre": "amphitheater",
-  "amphitheatres": "amphitheaters",
-  "anaemia": "anemia",
-  "anaemic": "anemic",
-  "anaesthesia": "anesthesia",
-  "anaesthetic": "anesthetic",
-  "anaesthetics": "anesthetics",
-  "anaesthetise": "anesthetize",
-  "anaesthetised": "anesthetized",
-  "anaesthetises": "anesthetizes",
-  "anaesthetising": "anesthetizing",
-  "anaesthetist": "anesthetist",
-  "anaesthetists": "anesthetists",
-  "anaesthetize": "anesthetize",
-  "anaesthetized": "anesthetized",
-  "anaesthetizes": "anesthetizes",
-  "anaesthetizing": "anesthetizing",
-  "analogue": "analog",
-  "analogues": "analogs",
-  "analyse": "analyze",
-  "analysed": "analyzed",
-  "analyses": "analyzes",
-  "analysing": "analyzing",
-  "anglicise": "anglicize",
-  "anglicised": "anglicized",
-  "anglicises": "anglicizes",
-  "anglicising": "anglicizing",
-  "annualised": "annualized",
-  "antagonise": "antagonize",
-  "antagonised": "antagonized",
-  "antagonises": "antagonizes",
-  "antagonising": "antagonizing",
-  "apologise": "apologize",
-  "apologised": "apologized",
-  "apologises": "apologizes",
-  "apologising": "apologizing",
-  "appal": "appall",
-  "appals": "appalls",
-  "appetiser": "appetizer",
-  "appetisers": "appetizers",
-  "appetising": "appetizing",
-  "appetisingly": "appetizingly",
-  "arbour": "arbor",
-  "arbours": "arbors",
-  "archaeologically": "archeologically",
-  "archaeologist": "archeologist",
-  "archaeologists": "archeologists",
-  "archaeology": "archeology",
-  "archaeological": "archeological",
-  "ardour": "ardor",
-  "armour": "armor",
-  "armoured": "armored",
-  "armourer": "armorer",
-  "armourers": "armorers",
-  "armouries": "armories",
-  "armoury": "armory",
-  "artefact": "artifact",
-  "artefacts": "artifacts",
-  "authorise": "authorize",
-  "authorised": "authorized",
-  "authorises": "authorizes",
-  "authorising": "authorizing",
-  "axe": "ax",
-  "backpedalled": "backpedaled",
-  "backpedalling": "backpedaling",
-  "bannister": "banister",
-  "bannisters": "banisters",
-  "baptise": "baptize",
-  "baptised": "baptized",
-  "baptises": "baptizes",
-  "baptising": "baptizing",
-  "bastardise": "bastardize",
-  "bastardised": "bastardized",
-  "bastardises": "bastardizes",
-  "bastardising": "bastardizing",
-  "battleax": "battleaxe",
-  "baulk": "balk",
-  "baulked": "balked",
-  "baulking": "balking",
-  "baulks": "balks",
-  "bedevilled": "bedeviled",
-  "bedevilling": "bedeviling",
-  "behaviour": "behavior",
-  "behavioural": "behavioral",
-  "behaviourism": "behaviorism",
-  "behaviourist": "behaviorist",
-  "behaviourists": "behaviorists",
-  "behaviours": "behaviors",
-  "behove": "behoove",
-  "behoved": "behooved",
-  "behoves": "behooves",
-  "bejewelled": "bejeweled",
-  "belabour": "belabor",
-  "belaboured": "belabored",
-  "belabouring": "belaboring",
-  "belabours": "belabors",
-  "bevelled": "beveled",
-  "bevvies": "bevies",
-  "bevvy": "bevy",
-  "biassed": "biased",
-  "biassing": "biasing",
-  "bingeing": "binging",
-  "bougainvillaea": "bougainvillea",
-  "bougainvillaeas": "bougainvilleas",
-  "bowdlerise": "bowdlerize",
-  "bowdlerised": "bowdlerized",
-  "bowdlerises": "bowdlerizes",
-  "bowdlerising": "bowdlerizing",
-  "breathalyse": "breathalyze",
-  "breathalysed": "breathalyzed",
-  "breathalyser": "breathalyzer",
-  "breathalysers": "breathalyzers",
-  "breathalyses": "breathalyzes",
-  "breathalysing": "breathalyzing",
-  "brutalise": "brutalize",
-  "brutalised": "brutalized",
-  "brutalises": "brutalizes",
-  "brutalising": "brutalizing",
-  "busses": "buses",
-  "bussing": "busing",
-  "caesarean": "cesarean",
-  "caesareans": "cesareans",
-  "calibre": "caliber",
-  "calibres": "calibers",
-  "calliper": "caliper",
-  "callipers": "calipers",
-  "callisthenics": "calisthenics",
-  "canalise": "canalize",
-  "canalised": "canalized",
-  "canalises": "canalizes",
-  "canalising": "canalizing",
-  "cancellation": "cancelation",
-  "cancellations": "cancelations",
-  "cancelled": "canceled",
-  "cancelling": "canceling",
-  "candour": "candor",
-  "cannibalise": "cannibalize",
-  "cannibalised": "cannibalized",
-  "cannibalises": "cannibalizes",
-  "cannibalising": "cannibalizing",
-  "canonise": "canonize",
-  "canonised": "canonized",
-  "canonises": "canonizes",
-  "canonising": "canonizing",
-  "capitalise": "capitalize",
-  "capitalised": "capitalized",
-  "capitalises": "capitalizes",
-  "capitalising": "capitalizing",
-  "caramelise": "caramelize",
-  "caramelised": "caramelized",
-  "caramelises": "caramelizes",
-  "caramelising": "caramelizing",
-  "carbonise": "carbonize",
-  "carbonised": "carbonized",
-  "carbonises": "carbonizes",
-  "carbonising": "carbonizing",
-  "carolled": "caroled",
-  "carolling": "caroling",
-  "catalogue": "catalog",
-  "catalogued": "cataloged",
-  "catalogues": "catalogs",
-  "cataloguing": "cataloging",
-  "catalyse": "catalyze",
-  "catalysed": "catalyzed",
-  "catalyses": "catalyzes",
-  "catalysing": "catalyzing",
-  "categorise": "categorize",
-  "categorised": "categorized",
-  "categorises": "categorizes",
-  "categorising": "categorizing",
-  "cauterise": "cauterize",
-  "cauterised": "cauterized",
-  "cauterises": "cauterizes",
-  "cauterising": "cauterizing",
-  "cavilled": "caviled",
-  "cavilling": "caviling",
-  "centigramme": "centigram",
-  "centigrammes": "centigrams",
-  "centilitre": "centiliter",
-  "centilitres": "centiliters",
-  "centimetre": "centimeter",
-  "centimetres": "centimeters",
-  "centralise": "centralize",
-  "centralised": "centralized",
-  "centralises": "centralizes",
-  "centralising": "centralizing",
-  "centre": "center",
-  "centred": "centered",
-  "centrefold": "centerfold",
-  "centrefolds": "centerfolds",
-  "centrepiece": "centerpiece",
-  "centrepieces": "centerpieces",
-  "centres": "centers",
-  "channelled": "channeled",
-  "channelling": "channeling",
-  "characterise": "characterize",
-  "characterised": "characterized",
-  "characterises": "characterizes",
-  "characterising": "characterizing",
-  "cheque": "check",
-  "chequebook": "checkbook",
-  "chequebooks": "checkbooks",
-  "chequered": "checkered",
-  "cheques": "checks",
-  "chilli": "chili",
-  "chimaera": "chimera",
-  "chimaeras": "chimeras",
-  "chiselled": "chiseled",
-  "chiselling": "chiseling",
-  "circularise": "circularize",
-  "circularised": "circularized",
-  "circularises": "circularizes",
-  "circularising": "circularizing",
-  "civilise": "civilize",
-  "civilised": "civilized",
-  "civilises": "civilizes",
-  "civilising": "civilizing",
-  "clamour": "clamor",
-  "clamoured": "clamored",
-  "clamouring": "clamoring",
-  "clamours": "clamors",
-  "clangour": "clangor",
-  "clarinettist": "clarinetist",
-  "clarinettists": "clarinetists",
-  "collectivise": "collectivize",
-  "collectivised": "collectivized",
-  "collectivises": "collectivizes",
-  "collectivising": "collectivizing",
-  "colonisation": "colonization",
-  "colonise": "colonize",
-  "colonised": "colonized",
-  "coloniser": "colonizer",
-  "colonisers": "colonizers",
-  "colonises": "colonizes",
-  "colonising": "colonizing",
-  "colour": "color",
-  "colourant": "colorant",
-  "colourants": "colorants",
-  "coloured": "colored",
-  "coloureds": "coloreds",
-  "colourful": "colorful",
-  "colourfully": "colorfully",
-  "colouring": "coloring",
-  "colourize": "colorize",
-  "colourized": "colorized",
-  "colourizes": "colorizes",
-  "colourizing": "colorizing",
-  "colourless": "colorless",
-  "colours": "colors",
-  "commercialise": "commercialize",
-  "commercialised": "commercialized",
-  "commercialises": "commercializes",
-  "commercialising": "commercializing",
-  "compartmentalise": "compartmentalize",
-  "compartmentalised": "compartmentalized",
-  "compartmentalises": "compartmentalizes",
-  "compartmentalising": "compartmentalizing",
-  "computerise": "computerize",
-  "computerised": "computerized",
-  "computerises": "computerizes",
-  "computerising": "computerizing",
-  "conceptualise": "conceptualize",
-  "conceptualised": "conceptualized",
-  "conceptualises": "conceptualizes",
-  "conceptualising": "conceptualizing",
-  "connexion": "connection",
-  "connexions": "connections",
-  "contextualise": "contextualize",
-  "contextualised": "contextualized",
-  "contextualises": "contextualizes",
-  "contextualising": "contextualizing",
-  "cosier": "cozier",
-  "cosies": "cozies",
-  "cosiest": "coziest",
-  "cosily": "cozily",
-  "cosiness": "coziness",
-  "cosy": "cozy",
-  "councillor": "councilor",
-  "councillors": "councilors",
-  "counselled": "counseled",
-  "counselling": "counseling",
-  "counsellor": "counselor",
-  "counsellors": "counselors",
-  "crenelated": "crenellated",
-  "criminalise": "criminalize",
-  "criminalised": "criminalized",
-  "criminalises": "criminalizes",
-  "criminalising": "criminalizing",
-  "criticise": "criticize",
-  "criticised": "criticized",
-  "criticises": "criticizes",
-  "criticising": "criticizing",
-  "crueller": "crueler",
-  "cruellest": "cruelest",
-  "crystallisation": "crystallization",
-  "crystallise": "crystallize",
-  "crystallised": "crystallized",
-  "crystallises": "crystallizes",
-  "crystallising": "crystallizing",
-  "cudgelled": "cudgeled",
-  "cudgelling": "cudgeling",
-  "customise": "customize",
-  "customised": "customized",
-  "customises": "customizes",
-  "customising": "customizing",
-  "cypher": "cipher",
-  "cyphers": "ciphers",
-  "decentralisation": "decentralization",
-  "decentralise": "decentralize",
-  "decentralised": "decentralized",
-  "decentralises": "decentralizes",
-  "decentralising": "decentralizing",
-  "decriminalisation": "decriminalization",
-  "decriminalise": "decriminalize",
-  "decriminalised": "decriminalized",
-  "decriminalises": "decriminalizes",
-  "decriminalising": "decriminalizing",
-  "defence": "defense",
-  "defenceless": "defenseless",
-  "defences": "defenses",
-  "dehumanisation": "dehumanization",
-  "dehumanise": "dehumanize",
-  "dehumanised": "dehumanized",
-  "dehumanises": "dehumanizes",
-  "dehumanising": "dehumanizing",
-  "demeanour": "demeanor",
-  "demilitarisation": "demilitarization",
-  "demilitarise": "demilitarize",
-  "demilitarised": "demilitarized",
-  "demilitarises": "demilitarizes",
-  "demilitarising": "demilitarizing",
-  "demobilisation": "demobilization",
-  "demobilise": "demobilize",
-  "demobilised": "demobilized",
-  "demobilises": "demobilizes",
-  "demobilising": "demobilizing",
-  "democratisation": "democratization",
-  "democratise": "democratize",
-  "democratised": "democratized",
-  "democratises": "democratizes",
-  "democratising": "democratizing",
-  "demonise": "demonize",
-  "demonised": "demonized",
-  "demonises": "demonizes",
-  "demonising": "demonizing",
-  "demoralisation": "demoralization",
-  "demoralise": "demoralize",
-  "demoralised": "demoralized",
-  "demoralises": "demoralizes",
-  "demoralising": "demoralizing",
-  "denationalisation": "denationalization",
-  "denationalise": "denationalize",
-  "denationalised": "denationalized",
-  "denationalises": "denationalizes",
-  "denationalising": "denationalizing",
-  "deodorise": "deodorize",
-  "deodorised": "deodorized",
-  "deodorises": "deodorizes",
-  "deodorising": "deodorizing",
-  "depersonalise": "depersonalize",
-  "depersonalised": "depersonalized",
-  "depersonalises": "depersonalizes",
-  "depersonalising": "depersonalizing",
-  "deputise": "deputize",
-  "deputised": "deputized",
-  "deputises": "deputizes",
-  "deputising": "deputizing",
-  "desensitisation": "desensitization",
-  "desensitise": "desensitize",
-  "desensitised": "desensitized",
-  "desensitises": "desensitizes",
-  "desensitising": "desensitizing",
-  "destabilisation": "destabilization",
-  "destabilise": "destabilize",
-  "destabilised": "destabilized",
-  "destabilises": "destabilizes",
-  "destabilising": "destabilizing",
-  "dialled": "dialed",
-  "dialling": "dialing",
-  "dialogue": "dialog",
-  "dialogues": "dialogs",
-  "diarrhoea": "diarrhea",
-  "digitise": "digitize",
-  "digitised": "digitized",
-  "digitises": "digitizes",
-  "digitising": "digitizing",
-  "disc": "disk",
-  "discolour": "discolor",
-  "discoloured": "discolored",
-  "discolouring": "discoloring",
-  "discolours": "discolors",
-  "discs": "disks",
-  "disembowelled": "disemboweled",
-  "disembowelling": "disemboweling",
-  "disfavour": "disfavor",
-  "dishevelled": "disheveled",
-  "dishonour": "dishonor",
-  "dishonourable": "dishonorable",
-  "dishonourably": "dishonorably",
-  "dishonoured": "dishonored",
-  "dishonouring": "dishonoring",
-  "dishonours": "dishonors",
-  "disorganisation": "disorganization",
-  "disorganised": "disorganized",
-  "distil": "distill",
-  "distils": "distills",
-  "dramatisation": "dramatization",
-  "dramatisations": "dramatizations",
-  "dramatise": "dramatize",
-  "dramatised": "dramatized",
-  "dramatises": "dramatizes",
-  "dramatising": "dramatizing",
-  "draught": "draft",
-  "draughtboard": "draftboard",
-  "draughtboards": "draftboards",
-  "draughtier": "draftier",
-  "draughtiest": "draftiest",
-  "draughts": "drafts",
-  "draughtsman": "draftsman",
-  "draughtsmanship": "draftsmanship",
-  "draughtsmen": "draftsmen",
-  "draughtswoman": "draftswoman",
-  "draughtswomen": "draftswomen",
-  "draughty": "drafty",
-  "drivelled": "driveled",
-  "drivelling": "driveling",
-  "duelled": "dueled",
-  "duelling": "dueling",
-  "economise": "economize",
-  "economised": "economized",
-  "economises": "economizes",
-  "economising": "economizing",
-  "editorialise": "editorialize",
-  "editorialised": "editorialized",
-  "editorialises": "editorializes",
-  "editorialising": "editorializing",
-  "edoema": "edema",
-  "empathise": "empathize",
-  "empathised": "empathized",
-  "empathises": "empathizes",
-  "empathising": "empathizing",
-  "emphasise": "emphasize",
-  "emphasised": "emphasized",
-  "emphasises": "emphasizes",
-  "emphasising": "emphasizing",
-  "enamelled": "enameled",
-  "enamelling": "enameling",
-  "enamoured": "enamored",
-  "encyclopaedia": "encyclopedia",
-  "encyclopaedias": "encyclopedias",
-  "encyclopaedic": "encyclopedic",
-  "endeavour": "endeavor",
-  "endeavoured": "endeavored",
-  "endeavouring": "endeavoring",
-  "endeavours": "endeavors",
-  "energise": "energize",
-  "energised": "energized",
-  "energises": "energizes",
-  "energising": "energizing",
-  "enrol": "enroll",
-  "enrols": "enrolls",
-  "enthral": "enthrall",
-  "enthrals": "enthralls",
-  "epaulette": "epaulet",
-  "epaulettes": "epaulets",
-  "epicentre": "epicenter",
-  "epicentres": "epicenters",
-  "epilogue": "epilog",
-  "epilogues": "epilogs",
-  "epitomise": "epitomize",
-  "epitomised": "epitomized",
-  "epitomises": "epitomizes",
-  "epitomising": "epitomizing",
-  "equalisation": "equalization",
-  "equalise": "equalize",
-  "equalised": "equalized",
-  "equaliser": "equalizer",
-  "equalisers": "equalizers",
-  "equalises": "equalizes",
-  "equalising": "equalizing",
-  "eulogise": "eulogize",
-  "eulogised": "eulogized",
-  "eulogises": "eulogizes",
-  "eulogising": "eulogizing",
-  "evangelise": "evangelize",
-  "evangelised": "evangelized",
-  "evangelises": "evangelizes",
-  "evangelising": "evangelizing",
-  "exorcise": "exorcize",
-  "exorcised": "exorcized",
-  "exorcises": "exorcizes",
-  "exorcising": "exorcizing",
-  "extemporisation": "extemporization",
-  "extemporise": "extemporize",
-  "extemporised": "extemporized",
-  "extemporises": "extemporizes",
-  "extemporising": "extemporizing",
-  "externalisation": "externalization",
-  "externalisations": "externalizations",
-  "externalise": "externalize",
-  "externalised": "externalized",
-  "externalises": "externalizes",
-  "externalising": "externalizing",
-  "factorise": "factorize",
-  "factorised": "factorized",
-  "factorises": "factorizes",
-  "factorising": "factorizing",
-  "faecal": "fecal",
-  "faeces": "feces",
-  "familiarisation": "familiarization",
-  "familiarise": "familiarize",
-  "familiarised": "familiarized",
-  "familiarises": "familiarizes",
-  "familiarising": "familiarizing",
-  "fantasise": "fantasize",
-  "fantasised": "fantasized",
-  "fantasises": "fantasizes",
-  "fantasising": "fantasizing",
-  "favour": "favor",
-  "favourable": "favorable",
-  "favourably": "favorably",
-  "favoured": "favored",
-  "favouring": "favoring",
-  "favourite": "favorite",
-  "favourites": "favorites",
-  "favouritism": "favoritism",
-  "favours": "favors",
-  "feminise": "feminize",
-  "feminised": "feminized",
-  "feminises": "feminizes",
-  "feminising": "feminizing",
-  "fertilisation": "fertilization",
-  "fertilise": "fertilize",
-  "fertilised": "fertilized",
-  "fertiliser": "fertilizer",
-  "fertilisers": "fertilizers",
-  "fertilises": "fertilizes",
-  "fertilising": "fertilizing",
-  "fervour": "fervor",
-  "fibre": "fiber",
-  "fibreglass": "fiberglass",
-  "fibres": "fibers",
-  "fictionalisation": "fictionalization",
-  "fictionalisations": "fictionalizations",
-  "fictionalise": "fictionalize",
-  "fictionalised": "fictionalized",
-  "fictionalises": "fictionalizes",
-  "fictionalising": "fictionalizing",
-  "fillet": "filet",
-  "filleted": "fileted",
-  "filleting": "fileting",
-  "fillets": "filets",
-  "finalisation": "finalization",
-  "finalise": "finalize",
-  "finalised": "finalized",
-  "finalises": "finalizes",
-  "finalising": "finalizing",
-  "flautist": "flutist",
-  "flautists": "flutists",
-  "flavour": "flavor",
-  "flavoured": "flavored",
-  "flavouring": "flavoring",
-  "flavourings": "flavorings",
-  "flavourless": "flavorless",
-  "flavours": "flavors",
-  "flavoursome": "flavorsome",
-  "flyer": "flier",
-  "foetal": "fetal",
-  "foetid": "fetid",
-  "foetus": "fetus",
-  "foetuses": "fetuses",
-  "formalisation": "formalization",
-  "formalise": "formalize",
-  "formalised": "formalized",
-  "formalises": "formalizes",
-  "formalising": "formalizing",
-  "fossilisation": "fossilization",
-  "fossilise": "fossilize",
-  "fossilised": "fossilized",
-  "fossilises": "fossilizes",
-  "fossilising": "fossilizing",
-  "fraternisation": "fraternization",
-  "fraternise": "fraternize",
-  "fraternised": "fraternized",
-  "fraternises": "fraternizes",
-  "fraternising": "fraternizing",
-  "fulfil": "fulfill",
-  "fulfilment": "fulfillment",
-  "fulfils": "fulfills",
-  "funnelled": "funneled",
-  "funnelling": "funneling",
-  "gage": "gauge",
-  "gaged": "gauged",
-  "gages": "gauges",
-  "gaging": "gauging",
-  "galvanise": "galvanize",
-  "galvanised": "galvanized",
-  "galvanises": "galvanizes",
-  "galvanising": "galvanizing",
-  "gambolled": "gamboled",
-  "gambolling": "gamboling",
-  "gaol": "jail",
-  "gaolbird": "jailbird",
-  "gaolbirds": "jailbirds",
-  "gaolbreak": "jailbreak",
-  "gaolbreaks": "jailbreaks",
-  "gaoled": "jailed",
-  "gaoler": "jailer",
-  "gaolers": "jailers",
-  "gaoling": "jailing",
-  "gaols": "jails",
-  "gasses": "gases",
-  "generalisation": "generalization",
-  "generalisations": "generalizations",
-  "generalise": "generalize",
-  "generalised": "generalized",
-  "generalises": "generalizes",
-  "generalising": "generalizing",
-  "ghettoise": "ghettoize",
-  "ghettoised": "ghettoized",
-  "ghettoises": "ghettoizes",
-  "ghettoising": "ghettoizing",
-  "gipsies": "gypsies",
-  "glamor": "glamour",
-  "glamorise": "glamorize",
-  "glamorised": "glamorized",
-  "glamorises": "glamorizes",
-  "glamorising": "glamorizing",
-  "globalisation": "globalization",
-  "globalise": "globalize",
-  "globalised": "globalized",
-  "globalises": "globalizes",
-  "globalising": "globalizing",
-  "glueing": "gluing",
-  "goitre": "goiter",
-  "goitres": "goiters",
-  "gonorrhoea": "gonorrhea",
-  "gramme": "gram",
-  "grammes": "grams",
-  "gravelled": "graveled",
-  "grey": "gray",
-  "greyed": "grayed",
-  "greying": "graying",
-  "greyish": "grayish",
-  "greyness": "grayness",
-  "greys": "grays",
-  "grovelled": "groveled",
-  "grovelling": "groveling",
-  "groyne": "groin",
-  "groynes": "groins",
-  "gruelling": "grueling",
-  "gruellingly": "gruelingly",
-  "gryphon": "griffin",
-  "gryphons": "griffins",
-  "gynaecological": "gynecological",
-  "gynaecologist": "gynecologist",
-  "gynaecologists": "gynecologists",
-  "gynaecology": "gynecology",
-  "haematological": "hematological",
-  "haematologist": "hematologist",
-  "haematologists": "hematologists",
-  "haematology": "hematology",
-  "haemoglobin": "hemoglobin",
-  "haemophilia": "hemophilia",
-  "haemophiliac": "hemophiliac",
-  "haemophiliacs": "hemophiliacs",
-  "haemorrhage": "hemorrhage",
-  "haemorrhaged": "hemorrhaged",
-  "haemorrhages": "hemorrhages",
-  "haemorrhaging": "hemorrhaging",
-  "haemorrhoids": "hemorrhoids",
-  "harbour": "harbor",
-  "harboured": "harbored",
-  "harbouring": "harboring",
-  "harbours": "harbors",
-  "harmonisation": "harmonization",
-  "harmonise": "harmonize",
-  "harmonised": "harmonized",
-  "harmonises": "harmonizes",
-  "harmonising": "harmonizing",
-  "homoeopath": "homeopath",
-  "homoeopathic": "homeopathic",
-  "homoeopaths": "homeopaths",
-  "homoeopathy": "homeopathy",
-  "homogenise": "homogenize",
-  "homogenised": "homogenized",
-  "homogenises": "homogenizes",
-  "homogenising": "homogenizing",
-  "honour": "honor",
-  "honourable": "honorable",
-  "honourably": "honorably",
-  "honoured": "honored",
-  "honouring": "honoring",
-  "honours": "honors",
-  "hospitalisation": "hospitalization",
-  "hospitalise": "hospitalize",
-  "hospitalised": "hospitalized",
-  "hospitalises": "hospitalizes",
-  "hospitalising": "hospitalizing",
-  "humanise": "humanize",
-  "humanised": "humanized",
-  "humanises": "humanizes",
-  "humanising": "humanizing",
-  "humour": "humor",
-  "humoured": "humored",
-  "humouring": "humoring",
-  "humourless": "humorless",
-  "humours": "humors",
-  "hybridise": "hybridize",
-  "hybridised": "hybridized",
-  "hybridises": "hybridizes",
-  "hybridising": "hybridizing",
-  "hypnotise": "hypnotize",
-  "hypnotised": "hypnotized",
-  "hypnotises": "hypnotizes",
-  "hypnotising": "hypnotizing",
-  "hypothesise": "hypothesize",
-  "hypothesised": "hypothesized",
-  "hypothesises": "hypothesizes",
-  "hypothesising": "hypothesizing",
-  "idealisation": "idealization",
-  "idealise": "idealize",
-  "idealised": "idealized",
-  "idealises": "idealizes",
-  "idealising": "idealizing",
-  "idolise": "idolize",
-  "idolised": "idolized",
-  "idolises": "idolizes",
-  "idolising": "idolizing",
-  "immobilisation": "immobilization",
-  "immobilise": "immobilize",
-  "immobilised": "immobilized",
-  "immobiliser": "immobilizer",
-  "immobilisers": "immobilizers",
-  "immobilises": "immobilizes",
-  "immobilising": "immobilizing",
-  "immortalise": "immortalize",
-  "immortalised": "immortalized",
-  "immortalises": "immortalizes",
-  "immortalising": "immortalizing",
-  "immunisation": "immunization",
-  "immunise": "immunize",
-  "immunised": "immunized",
-  "immunises": "immunizes",
-  "immunising": "immunizing",
-  "impanelled": "impaneled",
-  "impanelling": "impaneling",
-  "imperilled": "imperiled",
-  "imperilling": "imperiling",
-  "individualise": "individualize",
-  "individualised": "individualized",
-  "individualises": "individualizes",
-  "individualising": "individualizing",
-  "industrialise": "industrialize",
-  "industrialised": "industrialized",
-  "industrialises": "industrializes",
-  "industrialising": "industrializing",
-  "inflexion": "inflection",
-  "inflexions": "inflections",
-  "initialise": "initialize",
-  "initialised": "initialized",
-  "initialises": "initializes",
-  "initialising": "initializing",
-  "initialled": "initialed",
-  "initialling": "initialing",
-  "instal": "install",
-  "instalment": "installment",
-  "instalments": "installments",
-  "instals": "installs",
-  "instil": "instill",
-  "instils": "instills",
-  "institutionalisation": "institutionalization",
-  "institutionalise": "institutionalize",
-  "institutionalised": "institutionalized",
-  "institutionalises": "institutionalizes",
-  "institutionalising": "institutionalizing",
-  "intellectualise": "intellectualize",
-  "intellectualised": "intellectualized",
-  "intellectualises": "intellectualizes",
-  "intellectualising": "intellectualizing",
-  "internalisation": "internalization",
-  "internalise": "internalize",
-  "internalised": "internalized",
-  "internalises": "internalizes",
-  "internalising": "internalizing",
-  "internationalisation": "internationalization",
-  "internationalise": "internationalize",
-  "internationalised": "internationalized",
-  "internationalises": "internationalizes",
-  "internationalising": "internationalizing",
-  "ionisation": "ionization",
-  "ionise": "ionize",
-  "ionised": "ionized",
-  "ioniser": "ionizer",
-  "ionisers": "ionizers",
-  "ionises": "ionizes",
-  "ionising": "ionizing",
-  "italicise": "italicize",
-  "italicised": "italicized",
-  "italicises": "italicizes",
-  "italicising": "italicizing",
-  "itemise": "itemize",
-  "itemised": "itemized",
-  "itemises": "itemizes",
-  "itemising": "itemizing",
-  "jeopardise": "jeopardize",
-  "jeopardised": "jeopardized",
-  "jeopardises": "jeopardizes",
-  "jeopardising": "jeopardizing",
-  "jewelled": "jeweled",
-  "jeweller": "jeweler",
-  "jewellers": "jewelers",
-  "jewellery": "jewelry",
-  "judgement": "judgment",
-  "kilogramme": "kilogram",
-  "kilogrammes": "kilograms",
-  "kilometre": "kilometer",
-  "kilometres": "kilometers",
-  "labelled": "labeled",
-  "labelling": "labeling",
-  "labour": "labor",
-  "laboured": "labored",
-  "labourer": "laborer",
-  "labourers": "laborers",
-  "labouring": "laboring",
-  "labours": "labors",
-  "lacklustre": "lackluster",
-  "legalisation": "legalization",
-  "legalise": "legalize",
-  "legalised": "legalized",
-  "legalises": "legalizes",
-  "legalising": "legalizing",
-  "legitimise": "legitimize",
-  "legitimised": "legitimized",
-  "legitimises": "legitimizes",
-  "legitimising": "legitimizing",
-  "leukaemia": "leukemia",
-  "levelled": "leveled",
-  "leveller": "leveler",
-  "levellers": "levelers",
-  "levelling": "leveling",
-  "libelled": "libeled",
-  "libelling": "libeling",
-  "libellous": "libelous",
-  "liberalisation": "liberalization",
-  "liberalise": "liberalize",
-  "liberalised": "liberalized",
-  "liberalises": "liberalizes",
-  "liberalising": "liberalizing",
-  "licence": "license",
-  "licenced": "licensed",
-  "licences": "licenses",
-  "licencing": "licensing",
-  "likeable": "likable",
-  "lionisation": "lionization",
-  "lionise": "lionize",
-  "lionised": "lionized",
-  "lionises": "lionizes",
-  "lionising": "lionizing",
-  "liquidise": "liquidize",
-  "liquidised": "liquidized",
-  "liquidiser": "liquidizer",
-  "liquidisers": "liquidizers",
-  "liquidises": "liquidizes",
-  "liquidising": "liquidizing",
-  "litre": "liter",
-  "litres": "liters",
-  "localise": "localize",
-  "localised": "localized",
-  "localises": "localizes",
-  "localising": "localizing",
-  "louvre": "louver",
-  "louvred": "louvered",
-  "louvres": "louvers",
-  "lustre": "luster",
-  "magnetise": "magnetize",
-  "magnetised": "magnetized",
-  "magnetises": "magnetizes",
-  "magnetising": "magnetizing",
-  "manoeuvrability": "maneuverability",
-  "manoeuvrable": "maneuverable",
-  "manoeuvre": "maneuver",
-  "manoeuvred": "maneuvered",
-  "manoeuvres": "maneuvers",
-  "manoeuvring": "maneuvering",
-  "manoeuvrings": "maneuverings",
-  "marginalisation": "marginalization",
-  "marginalise": "marginalize",
-  "marginalised": "marginalized",
-  "marginalises": "marginalizes",
-  "marginalising": "marginalizing",
-  "marshalled": "marshaled",
-  "marshalling": "marshaling",
-  "marvelled": "marveled",
-  "marvelling": "marveling",
-  "marvellous": "marvelous",
-  "marvellously": "marvelously",
-  "materialisation": "materialization",
-  "materialise": "materialize",
-  "materialised": "materialized",
-  "materialises": "materializes",
-  "materialising": "materializing",
-  "maximisation": "maximization",
-  "maximise": "maximize",
-  "maximised": "maximized",
-  "maximises": "maximizes",
-  "maximising": "maximizing",
-  "meagre": "meager",
-  "mechanisation": "mechanization",
-  "mechanise": "mechanize",
-  "mechanised": "mechanized",
-  "mechanises": "mechanizes",
-  "mechanising": "mechanizing",
-  "mediaeval": "medieval",
-  "memorialise": "memorialize",
-  "memorialised": "memorialized",
-  "memorialises": "memorializes",
-  "memorialising": "memorializing",
-  "memorise": "memorize",
-  "memorised": "memorized",
-  "memorises": "memorizes",
-  "memorising": "memorizing",
-  "mesmerise": "mesmerize",
-  "mesmerised": "mesmerized",
-  "mesmerises": "mesmerizes",
-  "mesmerising": "mesmerizing",
-  "metabolise": "metabolize",
-  "metabolised": "metabolized",
-  "metabolises": "metabolizes",
-  "metabolising": "metabolizing",
-  "metre": "meter",
-  "metres": "meters",
-  "mhm": "hmm",
-  "micrometre": "micrometer",
-  "micrometres": "micrometers",
-  "militarise": "militarize",
-  "militarised": "militarized",
-  "militarises": "militarizes",
-  "militarising": "militarizing",
-  "milligramme": "milligram",
-  "milligrammes": "milligrams",
-  "millilitre": "milliliter",
-  "millilitres": "milliliters",
-  "millimetre": "millimeter",
-  "millimetres": "millimeters",
-  "miniaturisation": "miniaturization",
-  "miniaturise": "miniaturize",
-  "miniaturised": "miniaturized",
-  "miniaturises": "miniaturizes",
-  "miniaturising": "miniaturizing",
-  "minibusses": "minibuses",
-  "minimise": "minimize",
-  "minimised": "minimized",
-  "minimises": "minimizes",
-  "minimising": "minimizing",
-  "misbehaviour": "misbehavior",
-  "misdemeanour": "misdemeanor",
-  "misdemeanours": "misdemeanors",
-  "misspelt": "misspelled",
-  "mitre": "miter",
-  "mitres": "miters",
-  "mm": "hmm",
-  "mmm": "hmm",
-  "mobilisation": "mobilization",
-  "mobilise": "mobilize",
-  "mobilised": "mobilized",
-  "mobilises": "mobilizes",
-  "mobilising": "mobilizing",
-  "modelled": "modeled",
-  "modeller": "modeler",
-  "modellers": "modelers",
-  "modelling": "modeling",
-  "modernise": "modernize",
-  "modernised": "modernized",
-  "modernises": "modernizes",
-  "modernising": "modernizing",
-  "moisturise": "moisturize",
-  "moisturised": "moisturized",
-  "moisturiser": "moisturizer",
-  "moisturisers": "moisturizers",
-  "moisturises": "moisturizes",
-  "moisturising": "moisturizing",
-  "monologue": "monolog",
-  "monologues": "monologs",
-  "monopolisation": "monopolization",
-  "monopolise": "monopolize",
-  "monopolised": "monopolized",
-  "monopolises": "monopolizes",
-  "monopolising": "monopolizing",
-  "moralise": "moralize",
-  "moralised": "moralized",
-  "moralises": "moralizes",
-  "moralising": "moralizing",
-  "motorised": "motorized",
-  "mould": "mold",
-  "moulded": "molded",
-  "moulder": "molder",
-  "mouldered": "moldered",
-  "mouldering": "moldering",
-  "moulders": "molders",
-  "mouldier": "moldier",
-  "mouldiest": "moldiest",
-  "moulding": "molding",
-  "mouldings": "moldings",
-  "moulds": "molds",
-  "mouldy": "moldy",
-  "moult": "molt",
-  "moulted": "molted",
-  "moulting": "molting",
-  "moults": "molts",
-  "moustache": "mustache",
-  "moustached": "mustached",
-  "moustaches": "mustaches",
-  "moustachioed": "mustachioed",
-  "multicoloured": "multicolored",
-  "nationalisation": "nationalization",
-  "nationalisations": "nationalizations",
-  "nationalise": "nationalize",
-  "nationalised": "nationalized",
-  "nationalises": "nationalizes",
-  "nationalising": "nationalizing",
-  "naturalisation": "naturalization",
-  "naturalise": "naturalize",
-  "naturalised": "naturalized",
-  "naturalises": "naturalizes",
-  "naturalising": "naturalizing",
-  "neighbour": "neighbor",
-  "neighbourhood": "neighborhood",
-  "neighbourhoods": "neighborhoods",
-  "neighbouring": "neighboring",
-  "neighbourliness": "neighborliness",
-  "neighbourly": "neighborly",
-  "neighbours": "neighbors",
-  "neutralisation": "neutralization",
-  "neutralise": "neutralize",
-  "neutralised": "neutralized",
-  "neutralises": "neutralizes",
-  "neutralising": "neutralizing",
-  "normalisation": "normalization",
-  "normalise": "normalize",
-  "normalised": "normalized",
-  "normalises": "normalizes",
-  "normalising": "normalizing",
-  "odour": "odor",
-  "odourless": "odorless",
-  "odours": "odors",
-  "oesophagus": "esophagus",
-  "oesophaguses": "esophaguses",
-  "oestrogen": "estrogen",
-  "offence": "offense",
-  "offences": "offenses",
-  "omelette": "omelet",
-  "omelettes": "omelets",
-  "optimise": "optimize",
-  "optimised": "optimized",
-  "optimises": "optimizes",
-  "optimising": "optimizing",
-  "organisation": "organization",
-  "organisational": "organizational",
-  "organisations": "organizations",
-  "organise": "organize",
-  "organised": "organized",
-  "organiser": "organizer",
-  "organisers": "organizers",
-  "organises": "organizes",
-  "organising": "organizing",
-  "orthopaedic": "orthopedic",
-  "orthopaedics": "orthopedics",
-  "ostracise": "ostracize",
-  "ostracised": "ostracized",
-  "ostracises": "ostracizes",
-  "ostracising": "ostracizing",
-  "outmanoeuvre": "outmaneuver",
-  "outmanoeuvred": "outmaneuvered",
-  "outmanoeuvres": "outmaneuvers",
-  "outmanoeuvring": "outmaneuvering",
-  "overemphasise": "overemphasize",
-  "overemphasised": "overemphasized",
-  "overemphasises": "overemphasizes",
-  "overemphasising": "overemphasizing",
-  "oxidisation": "oxidization",
-  "oxidise": "oxidize",
-  "oxidised": "oxidized",
-  "oxidises": "oxidizes",
-  "oxidising": "oxidizing",
-  "paederast": "pederast",
-  "paederasts": "pederasts",
-  "paediatric": "pediatric",
-  "paediatrician": "pediatrician",
-  "paediatricians": "pediatricians",
-  "paediatrics": "pediatrics",
-  "paedophile": "pedophile",
-  "paedophiles": "pedophiles",
-  "paedophilia": "pedophilia",
-  "palaeolithic": "paleolithic",
-  "palaeontologist": "paleontologist",
-  "palaeontologists": "paleontologists",
-  "palaeontology": "paleontology",
-  "panelled": "paneled",
-  "panelling": "paneling",
-  "panellist": "panelist",
-  "panellists": "panelists",
-  "paralyse": "paralyze",
-  "paralysed": "paralyzed",
-  "paralyses": "paralyzes",
-  "paralysing": "paralyzing",
-  "parcelled": "parceled",
-  "parcelling": "parceling",
-  "parlour": "parlor",
-  "parlours": "parlors",
-  "particularise": "particularize",
-  "particularised": "particularized",
-  "particularises": "particularizes",
-  "particularising": "particularizing",
-  "passivisation": "passivization",
-  "passivise": "passivize",
-  "passivised": "passivized",
-  "passivises": "passivizes",
-  "passivising": "passivizing",
-  "pasteurisation": "pasteurization",
-  "pasteurise": "pasteurize",
-  "pasteurised": "pasteurized",
-  "pasteurises": "pasteurizes",
-  "pasteurising": "pasteurizing",
-  "patronise": "patronize",
-  "patronised": "patronized",
-  "patronises": "patronizes",
-  "patronising": "patronizing",
-  "patronisingly": "patronizingly",
-  "pedalled": "pedaled",
-  "pedalling": "pedaling",
-  "pedestrianisation": "pedestrianization",
-  "pedestrianise": "pedestrianize",
-  "pedestrianised": "pedestrianized",
-  "pedestrianises": "pedestrianizes",
-  "pedestrianising": "pedestrianizing",
-  "penalise": "penalize",
-  "penalised": "penalized",
-  "penalises": "penalizes",
-  "penalising": "penalizing",
-  "pencilled": "penciled",
-  "pencilling": "penciling",
-  "personalise": "personalize",
-  "personalised": "personalized",
-  "personalises": "personalizes",
-  "personalising": "personalizing",
-  "pharmacopoeia": "pharmacopeia",
-  "pharmacopoeias": "pharmacopeias",
-  "philosophise": "philosophize",
-  "philosophised": "philosophized",
-  "philosophises": "philosophizes",
-  "philosophising": "philosophizing",
-  "philtre": "filter",
-  "philtres": "filters",
-  "phoney": "phony",
-  "plagiarise": "plagiarize",
-  "plagiarised": "plagiarized",
-  "plagiarises": "plagiarizes",
-  "plagiarising": "plagiarizing",
-  "plough": "plow",
-  "ploughed": "plowed",
-  "ploughing": "plowing",
-  "ploughman": "plowman",
-  "ploughmen": "plowmen",
-  "ploughs": "plows",
-  "ploughshare": "plowshare",
-  "ploughshares": "plowshares",
-  "polarisation": "polarization",
-  "polarise": "polarize",
-  "polarised": "polarized",
-  "polarises": "polarizes",
-  "polarising": "polarizing",
-  "politicisation": "politicization",
-  "politicise": "politicize",
-  "politicised": "politicized",
-  "politicises": "politicizes",
-  "politicising": "politicizing",
-  "popularisation": "popularization",
-  "popularise": "popularize",
-  "popularised": "popularized",
-  "popularises": "popularizes",
-  "popularising": "popularizing",
-  "pouffe": "pouf",
-  "pouffes": "poufs",
-  "practise": "practice",
-  "practised": "practiced",
-  "practises": "practices",
-  "practising": "practicing",
-  "praesidium": "presidium",
-  "praesidiums": "presidiums",
-  "pressurisation": "pressurization",
-  "pressurise": "pressurize",
-  "pressurised": "pressurized",
-  "pressurises": "pressurizes",
-  "pressurising": "pressurizing",
-  "pretence": "pretense",
-  "pretences": "pretenses",
-  "primaeval": "primeval",
-  "prioritisation": "prioritization",
-  "prioritise": "prioritize",
-  "prioritised": "prioritized",
-  "prioritises": "prioritizes",
-  "prioritising": "prioritizing",
-  "privatisation": "privatization",
-  "privatisations": "privatizations",
-  "privatise": "privatize",
-  "privatised": "privatized",
-  "privatises": "privatizes",
-  "privatising": "privatizing",
-  "professionalisation": "professionalization",
-  "professionalise": "professionalize",
-  "professionalised": "professionalized",
-  "professionalises": "professionalizes",
-  "professionalising": "professionalizing",
-  "programme": "program",
-  "programmes": "programs",
-  "prologue": "prolog",
-  "prologues": "prologs",
-  "propagandise": "propagandize",
-  "propagandised": "propagandized",
-  "propagandises": "propagandizes",
-  "propagandising": "propagandizing",
-  "proselytise": "proselytize",
-  "proselytised": "proselytized",
-  "proselytiser": "proselytizer",
-  "proselytisers": "proselytizers",
-  "proselytises": "proselytizes",
-  "proselytising": "proselytizing",
-  "psychoanalyse": "psychoanalyze",
-  "psychoanalysed": "psychoanalyzed",
-  "psychoanalyses": "psychoanalyzes",
-  "psychoanalysing": "psychoanalyzing",
-  "publicise": "publicize",
-  "publicised": "publicized",
-  "publicises": "publicizes",
-  "publicising": "publicizing",
-  "pulverisation": "pulverization",
-  "pulverise": "pulverize",
-  "pulverised": "pulverized",
-  "pulverises": "pulverizes",
-  "pulverising": "pulverizing",
-  "pummelled": "pummeled",
-  "pummelling": "pummeling",
-  "pyjama": "pajama",
-  "pyjamas": "pajamas",
-  "pzazz": "pizzazz",
-  "quarrelled": "quarreled",
-  "quarrelling": "quarreling",
-  "radicalise": "radicalize",
-  "radicalised": "radicalized",
-  "radicalises": "radicalizes",
-  "radicalising": "radicalizing",
-  "rancour": "rancor",
-  "randomise": "randomize",
-  "randomised": "randomized",
-  "randomises": "randomizes",
-  "randomising": "randomizing",
-  "rationalisation": "rationalization",
-  "rationalisations": "rationalizations",
-  "rationalise": "rationalize",
-  "rationalised": "rationalized",
-  "rationalises": "rationalizes",
-  "rationalising": "rationalizing",
-  "ravelled": "raveled",
-  "ravelling": "raveling",
-  "realisable": "realizable",
-  "realisation": "realization",
-  "realisations": "realizations",
-  "realise": "realize",
-  "realised": "realized",
-  "realises": "realizes",
-  "realising": "realizing",
-  "recognisable": "recognizable",
-  "recognisably": "recognizably",
-  "recognisance": "recognizance",
-  "recognise": "recognize",
-  "recognised": "recognized",
-  "recognises": "recognizes",
-  "recognising": "recognizing",
-  "reconnoitre": "reconnoiter",
-  "reconnoitred": "reconnoitered",
-  "reconnoitres": "reconnoiters",
-  "reconnoitring": "reconnoitering",
-  "refuelled": "refueled",
-  "refuelling": "refueling",
-  "regularisation": "regularization",
-  "regularise": "regularize",
-  "regularised": "regularized",
-  "regularises": "regularizes",
-  "regularising": "regularizing",
-  "remodelled": "remodeled",
-  "remodelling": "remodeling",
-  "remould": "remold",
-  "remoulded": "remolded",
-  "remoulding": "remolding",
-  "remoulds": "remolds",
-  "reorganisation": "reorganization",
-  "reorganisations": "reorganizations",
-  "reorganise": "reorganize",
-  "reorganised": "reorganized",
-  "reorganises": "reorganizes",
-  "reorganising": "reorganizing",
-  "revelled": "reveled",
-  "reveller": "reveler",
-  "revellers": "revelers",
-  "revelling": "reveling",
-  "revitalise": "revitalize",
-  "revitalised": "revitalized",
-  "revitalises": "revitalizes",
-  "revitalising": "revitalizing",
-  "revolutionise": "revolutionize",
-  "revolutionised": "revolutionized",
-  "revolutionises": "revolutionizes",
-  "revolutionising": "revolutionizing",
-  "rhapsodise": "rhapsodize",
-  "rhapsodised": "rhapsodized",
-  "rhapsodises": "rhapsodizes",
-  "rhapsodising": "rhapsodizing",
-  "rigour": "rigor",
-  "rigours": "rigors",
-  "ritualised": "ritualized",
-  "rivalled": "rivaled",
-  "rivalling": "rivaling",
-  "romanticise": "romanticize",
-  "romanticised": "romanticized",
-  "romanticises": "romanticizes",
-  "romanticising": "romanticizing",
-  "rumour": "rumor",
-  "rumoured": "rumored",
-  "rumours": "rumors",
-  "sabre": "saber",
-  "sabres": "sabers",
-  "saltpetre": "saltpeter",
-  "sanitise": "sanitize",
-  "sanitised": "sanitized",
-  "sanitises": "sanitizes",
-  "sanitising": "sanitizing",
-  "satirise": "satirize",
-  "satirised": "satirized",
-  "satirises": "satirizes",
-  "satirising": "satirizing",
-  "saviour": "savior",
-  "saviours": "saviors",
-  "savour": "savor",
-  "savoured": "savored",
-  "savouries": "savories",
-  "savouring": "savoring",
-  "savours": "savors",
-  "savoury": "savory",
-  "scandalise": "scandalize",
-  "scandalised": "scandalized",
-  "scandalises": "scandalizes",
-  "scandalising": "scandalizing",
-  "sceptic": "skeptic",
-  "sceptical": "skeptical",
-  "sceptically": "skeptically",
-  "scepticism": "skepticism",
-  "sceptics": "skeptics",
-  "sceptre": "scepter",
-  "sceptres": "scepters",
-  "scrutinise": "scrutinize",
-  "scrutinised": "scrutinized",
-  "scrutinises": "scrutinizes",
-  "scrutinising": "scrutinizing",
-  "secularisation": "secularization",
-  "secularise": "secularize",
-  "secularised": "secularized",
-  "secularises": "secularizes",
-  "secularising": "secularizing",
-  "sensationalise": "sensationalize",
-  "sensationalised": "sensationalized",
-  "sensationalises": "sensationalizes",
-  "sensationalising": "sensationalizing",
-  "sensitise": "sensitize",
-  "sensitised": "sensitized",
-  "sensitises": "sensitizes",
-  "sensitising": "sensitizing",
-  "sentimentalise": "sentimentalize",
-  "sentimentalised": "sentimentalized",
-  "sentimentalises": "sentimentalizes",
-  "sentimentalising": "sentimentalizing",
-  "sepulchre": "sepulcher",
-  "sepulchres": "sepulchers",
-  "serialisation": "serialization",
-  "serialisations": "serializations",
-  "serialise": "serialize",
-  "serialised": "serialized",
-  "serialises": "serializes",
-  "serialising": "serializing",
-  "sermonise": "sermonize",
-  "sermonised": "sermonized",
-  "sermonises": "sermonizes",
-  "sermonising": "sermonizing",
-  "sheikh": "sheik",
-  "shovelled": "shoveled",
-  "shovelling": "shoveling",
-  "shrivelled": "shriveled",
-  "shrivelling": "shriveling",
-  "signalise": "signalize",
-  "signalised": "signalized",
-  "signalises": "signalizes",
-  "signalising": "signalizing",
-  "signalled": "signaled",
-  "signalling": "signaling",
-  "smoulder": "smolder",
-  "smouldered": "smoldered",
-  "smouldering": "smoldering",
-  "smoulders": "smolders",
-  "snivelled": "sniveled",
-  "snivelling": "sniveling",
-  "snorkelled": "snorkeled",
-  "snorkelling": "snorkeling",
-  "snowplough": "snowplow",
-  "snowploughs": "snowplows",
-  "socialisation": "socialization",
-  "socialise": "socialize",
-  "socialised": "socialized",
-  "socialises": "socializes",
-  "socialising": "socializing",
-  "sodomise": "sodomize",
-  "sodomised": "sodomized",
-  "sodomises": "sodomizes",
-  "sodomising": "sodomizing",
-  "solemnise": "solemnize",
-  "solemnised": "solemnized",
-  "solemnises": "solemnizes",
-  "solemnising": "solemnizing",
-  "sombre": "somber",
-  "specialisation": "specialization",
-  "specialisations": "specializations",
-  "specialise": "specialize",
-  "specialised": "specialized",
-  "specialises": "specializes",
-  "specialising": "specializing",
-  "spectre": "specter",
-  "spectres": "specters",
-  "spiralled": "spiraled",
-  "spiralling": "spiraling",
-  "splendour": "splendor",
-  "splendours": "splendors",
-  "squirrelled": "squirreled",
-  "squirrelling": "squirreling",
-  "stabilisation": "stabilization",
-  "stabilise": "stabilize",
-  "stabilised": "stabilized",
-  "stabiliser": "stabilizer",
-  "stabilisers": "stabilizers",
-  "stabilises": "stabilizes",
-  "stabilising": "stabilizing",
-  "standardisation": "standardization",
-  "standardise": "standardize",
-  "standardised": "standardized",
-  "standardises": "standardizes",
-  "standardising": "standardizing",
-  "stencilled": "stenciled",
-  "stencilling": "stenciling",
-  "sterilisation": "sterilization",
-  "sterilisations": "sterilizations",
-  "sterilise": "sterilize",
-  "sterilised": "sterilized",
-  "steriliser": "sterilizer",
-  "sterilisers": "sterilizers",
-  "sterilises": "sterilizes",
-  "sterilising": "sterilizing",
-  "stigmatisation": "stigmatization",
-  "stigmatise": "stigmatize",
-  "stigmatised": "stigmatized",
-  "stigmatises": "stigmatizes",
-  "stigmatising": "stigmatizing",
-  "storey": "story",
-  "storeys": "stories",
-  "subsidisation": "subsidization",
-  "subsidise": "subsidize",
-  "subsidised": "subsidized",
-  "subsidiser": "subsidizer",
-  "subsidisers": "subsidizers",
-  "subsidises": "subsidizes",
-  "subsidising": "subsidizing",
-  "succour": "succor",
-  "succoured": "succored",
-  "succouring": "succoring",
-  "succours": "succors",
-  "sulphate": "sulfate",
-  "sulphates": "sulfates",
-  "sulphide": "sulfide",
-  "sulphides": "sulfides",
-  "sulphur": "sulfur",
-  "sulphurous": "sulfurous",
-  "summarise": "summarize",
-  "summarised": "summarized",
-  "summarises": "summarizes",
-  "summarising": "summarizing",
-  "swivelled": "swiveled",
-  "swivelling": "swiveling",
-  "symbolise": "symbolize",
-  "symbolised": "symbolized",
-  "symbolises": "symbolizes",
-  "symbolising": "symbolizing",
-  "sympathise": "sympathize",
-  "sympathised": "sympathized",
-  "sympathiser": "sympathizer",
-  "sympathisers": "sympathizers",
-  "sympathises": "sympathizes",
-  "sympathising": "sympathizing",
-  "synchronisation": "synchronization",
-  "synchronise": "synchronize",
-  "synchronised": "synchronized",
-  "synchronises": "synchronizes",
-  "synchronising": "synchronizing",
-  "synthesise": "synthesize",
-  "synthesised": "synthesized",
-  "synthesiser": "synthesizer",
-  "synthesisers": "synthesizers",
-  "synthesises": "synthesizes",
-  "synthesising": "synthesizing",
-  "syphon": "siphon",
-  "syphoned": "siphoned",
-  "syphoning": "siphoning",
-  "syphons": "siphons",
-  "systematisation": "systematization",
-  "systematise": "systematize",
-  "systematised": "systematized",
-  "systematises": "systematizes",
-  "systematising": "systematizing",
-  "tantalise": "tantalize",
-  "tantalised": "tantalized",
-  "tantalises": "tantalizes",
-  "tantalising": "tantalizing",
-  "tantalisingly": "tantalizingly",
-  "tasselled": "tasseled",
-  "technicolour": "technicolor",
-  "temporise": "temporize",
-  "temporised": "temporized",
-  "temporises": "temporizes",
-  "temporising": "temporizing",
-  "tenderise": "tenderize",
-  "tenderised": "tenderized",
-  "tenderises": "tenderizes",
-  "tenderising": "tenderizing",
-  "terrorise": "terrorize",
-  "terrorised": "terrorized",
-  "terrorises": "terrorizes",
-  "terrorising": "terrorizing",
-  "theatre": "theater",
-  "theatregoer": "theatergoer",
-  "theatregoers": "theatergoers",
-  "theatres": "theaters",
-  "theorise": "theorize",
-  "theorised": "theorized",
-  "theorises": "theorizes",
-  "theorising": "theorizing",
-  "tonne": "ton",
-  "tonnes": "tons",
-  "towelled": "toweled",
-  "towelling": "toweling",
-  "toxaemia": "toxemia",
-  "tranquillise": "tranquilize",
-  "tranquillised": "tranquilized",
-  "tranquilliser": "tranquilizer",
-  "tranquillisers": "tranquilizers",
-  "tranquillises": "tranquilizes",
-  "tranquillising": "tranquilizing",
-  "tranquillity": "tranquility",
-  "tranquillize": "tranquilize",
-  "tranquillized": "tranquilized",
-  "tranquillizer": "tranquilizer",
-  "tranquillizers": "tranquilizers",
-  "tranquillizes": "tranquilizes",
-  "tranquillizing": "tranquilizing",
-  "tranquilly": "tranquility",
-  "transistorised": "transistorized",
-  "traumatise": "traumatize",
-  "traumatised": "traumatized",
-  "traumatises": "traumatizes",
-  "traumatising": "traumatizing",
-  "travelled": "traveled",
-  "traveller": "traveler",
-  "travellers": "travelers",
-  "travelling": "traveling",
-  "travelogue": "travelog",
-  "travelogues": "travelogs",
-  "trialled": "trialed",
-  "trialling": "trialing",
-  "tricolour": "tricolor",
-  "tricolours": "tricolors",
-  "trivialise": "trivialize",
-  "trivialised": "trivialized",
-  "trivialises": "trivializes",
-  "trivialising": "trivializing",
-  "tumour": "tumor",
-  "tumours": "tumors",
-  "tunnelled": "tunneled",
-  "tunnelling": "tunneling",
-  "tyrannise": "tyrannize",
-  "tyrannised": "tyrannized",
-  "tyrannises": "tyrannizes",
-  "tyrannising": "tyrannizing",
-  "tyre": "tire",
-  "tyres": "tires",
-  "unauthorised": "unauthorized",
-  "uncivilised": "uncivilized",
-  "underutilised": "underutilized",
-  "unequalled": "unequaled",
-  "unfavourable": "unfavorable",
-  "unfavourably": "unfavorably",
-  "unionisation": "unionization",
-  "unionise": "unionize",
-  "unionised": "unionized",
-  "unionises": "unionizes",
-  "unionising": "unionizing",
-  "unorganised": "unorganized",
-  "unravelled": "unraveled",
-  "unravelling": "unraveling",
-  "unrecognisable": "unrecognizable",
-  "unrecognised": "unrecognized",
-  "unrivalled": "unrivaled",
-  "unsavoury": "unsavory",
-  "untrammelled": "untrammeled",
-  "urbanisation": "urbanization",
-  "urbanise": "urbanize",
-  "urbanised": "urbanized",
-  "urbanises": "urbanizes",
-  "urbanising": "urbanizing",
-  "utilisable": "utilizable",
-  "utilisation": "utilization",
-  "utilise": "utilize",
-  "utilised": "utilized",
-  "utilises": "utilizes",
-  "utilising": "utilizing",
-  "valour": "valor",
-  "vandalise": "vandalize",
-  "vandalised": "vandalized",
-  "vandalises": "vandalizes",
-  "vandalising": "vandalizing",
-  "vaporisation": "vaporization",
-  "vaporise": "vaporize",
-  "vaporised": "vaporized",
-  "vaporises": "vaporizes",
-  "vaporising": "vaporizing",
-  "vapour": "vapor",
-  "vapours": "vapors",
-  "verbalise": "verbalize",
-  "verbalised": "verbalized",
-  "verbalises": "verbalizes",
-  "verbalising": "verbalizing",
-  "victimisation": "victimization",
-  "victimise": "victimize",
-  "victimised": "victimized",
-  "victimises": "victimizes",
-  "victimising": "victimizing",
-  "videodisc": "videodisk",
-  "videodiscs": "videodisks",
-  "vigour": "vigor",
-  "visualisation": "visualization",
-  "visualisations": "visualizations",
-  "visualise": "visualize",
-  "visualised": "visualized",
-  "visualises": "visualizes",
-  "visualising": "visualizing",
-  "vocalisation": "vocalization",
-  "vocalisations": "vocalizations",
-  "vocalise": "vocalize",
-  "vocalised": "vocalized",
-  "vocalises": "vocalizes",
-  "vocalising": "vocalizing",
-  "vulcanised": "vulcanized",
-  "vulgarisation": "vulgarization",
-  "vulgarise": "vulgarize",
-  "vulgarised": "vulgarized",
-  "vulgarises": "vulgarizes",
-  "vulgarising": "vulgarizing",
-  "waggon": "wagon",
-  "waggons": "wagons",
-  "watercolour": "watercolor",
-  "watercolours": "watercolors",
-  "weaselled": "weaseled",
-  "weaselling": "weaseling",
-  "westernisation": "westernization",
-  "westernise": "westernize",
-  "westernised": "westernized",
-  "westernises": "westernizes",
-  "westernising": "westernizing",
-  "womanise": "womanize",
-  "womanised": "womanized",
-  "womaniser": "womanizer",
-  "womanisers": "womanizers",
-  "womanises": "womanizes",
-  "womanising": "womanizing",
-  "woollen": "woolen",
-  "woollens": "woolens",
-  "woollies": "woolies",
-  "woolly": "wooly",
-  "worshipped": "worshiped",
-  "worshipper": "worshiper",
-  "worshipping": "worshiping",
-  "yodelled": "yodeled",
-  "yodelling": "yodeling",
-  "yoghourt": "yogurt",
-  "yoghourts": "yogurts",
-  "yoghurt": "yogurt",
-  "yoghurts": "yogurts"
-}
-
-
-english_name_normalizer = {
-    # ── Double-letter variants ──────────────────────────────────────────────
-    "alan": "allen",
-    "allan": "allen",
-    "bridgette": "bridget",
-    "charly": "charlie",
-    "charley": "charlie",
-    "garry": "gary",
-    "gregg": "greg",
-    "jacky": "jackie",
-    "joann": "joanne",
-    "joane": "joanne",
-    "kellye": "kelly",
-    "kelli": "kelly",
-    "kelley": "kelly",
-    "lilly": "lily",
-    "micheal": "michael",
-    "michele": "michelle",
-    "mollie": "molly",
-    "phillip": "philip",
-    "sallie": "sally",
-    "stacey": "stacy",
-    "stacie": "stacy",
-    "tracey": "tracy",
-    "tracie": "tracy",
-    "bret": "brett",
-    "carrol": "carol",
-    "carole": "carol",
-    "carroll": "carol",
-    "allison": "alison",
-    "alyson": "alison",
-    "russel": "russell",
-    "douglass": "douglas",
-    "dominick": "dominic",
-    "robb": "rob",
-    # ── Chr/Kr variants ─────────────────────────────────────────────────────
-    "kris": "chris",
-    "kristopher": "christopher",
-    "cristopher": "christopher",
-    "kristina": "christina",
-    "kristen": "kristin",
-    # ── C/K variants ────────────────────────────────────────────────────────
-    "karl": "carl",
-    "kathy": "cathy",
-    "katherine": "catherine",
-    "kathryn": "catherine",
-    "catharine": "catherine",
-    "erik": "eric",
-    "erick": "eric",
-    "caren": "karen",
-    "caryn": "karen",
-    "karin": "karen",
-    "katelyn": "caitlin",
-    "kaitlyn": "caitlin",
-    "kaitlin": "caitlin",
-    "nikole": "nicole",
-    "veronika": "veronica",
-    "viktor": "victor",
-    "viktoria": "victoria",
-    "kevan": "kevin",
-    "patrik": "patrick",
-    "frederik": "frederick",
-    "fredrick": "frederick",
-    "lukas": "lucas",
-    # ── Silent letters / alternate spellings ───────────────────────────────
-    "ann": "anne",
-    "jon": "john",
-    "johnathan": "jonathan",
-    "jonathon": "jonathan",
-    "sara": "sarah",
-    "mathew": "matthew",
-    "nicolas": "nicholas",
-    "rachael": "rachel",
-    "rebekah": "rebecca",
-    "devorah": "deborah",
-    "theresa": "teresa",
-    "suzanne": "susanne",
-    "antony": "anthony",
-    "martyn": "martin",
-    "denis": "dennis",
-    "laurence": "lawrence",
-    "tomas": "thomas",
-    "tobey": "toby",
-    # ── Mac/Mc extensions ───────────────────────────────────────────────────
-    "macarthur": "mcarthur",
-    "macartney": "mccartney",
-    "macarthy": "mccarthy",
-    "maccarthy": "mccarthy",
-    "macdonald": "mcdonald",
-    "mackay": "mckay",
-    "mackenzie": "mckenzie",
-    "macleod": "mcleod",
-    "maclean": "mclean",
-    "macmillan": "mcmillan",
-    "macintosh": "mcintosh",
-    "macintyre": "mcintyre",
-    "macnamara": "mcnamara",
-    "macgowan": "mcgowan",
-    # ── International ─────────────────────────────────
-    "mohamad": "mohammed",
-    "mohamed": "mohammed",
-    "mohammad": "mohammed",
-    "muhammad": "mohammed",
-    "muhamad": "mohammed",
-    "muhammed": "mohammed",
-    "mouhamed": "mohammed",
-    "mouhamad": "mohammed",
-    "mahomet": "mohammed",
-    "fatimah": "fatima",
-    "yusuf": "yousef",
-    "yusef": "yousef",
-    "myriam": "miriam",
-    "rajeev": "rajiv",
-    # ── Miscellaneous homophones ────────────────────────────────────────────
-    "alphonso": "alfonso",
-    "bryan": "brian",
-    "geoffrey": "jeffrey",
-    "jeffery": "jeffrey",
-    "geoff": "jeff",
-    "neal": "neil",
-    "shaun": "sean",
-    "shawn": "sean",
-    "shayne": "shane",
-    "stephen": "steven",
-    "toni": "tony",
-    "leigh": "lee",
-    "lewis": "louis",
-    "marc": "mark",
-    "meghan": "megan",
-    "nathalie": "natalie",
-    "robyn": "robin",
-    "rodger": "roger",
-    "linsey": "lindsay",
-    "lindsey": "lindsay",
-    "zackary": "zachary",
-    "zachery": "zachary",
-    "zak": "zach",
-    "sheri": "sherry",
-    "cheri": "sherry",
-    "sherrie": "sherry",
-    "terri": "terry",
-    "lori": "laurie",
-    "jaime": "jamie",
-    "jayson": "jason",
-    "lesley": "leslie",
-    "lynda": "linda",
-    "lynne": "lynn",
-    "gayle": "gail",
-    "rhonda": "ronda",
-    "yvonne": "ivonne",
-    "stewart": "stuart",
-    "walther": "walter",
-    "symon": "simon",
-    "collin": "colin",
-    "dillon": "dylan",
-    "aron": "aaron",
-    "artur": "arthur",
-    "henri": "henry",
-    "josef": "joseph",
-    "pieter": "peter",
-}
-
-
-# Regex-based multi-word → single-token mappings, plus spellings that vary too
-# freely to enumerate in english_spelling_normalizer (e.g. elongations).
-# Applied after symbol removal, so hyphens/punctuation are already stripped.
-# Keys are used with re.sub; values are the replacement strings.
-english_compound_normalizer = {
-    r"\bet\s+cetera\b": "etc",
-    r"\bal\s+right\b": "alright",
-    r"\ball\s+right\b": "alright",
-    r"\bhow\s+ever\b": "however",
-    r"\bwi\s+fi\b": "wifi",
-    r"\bhi\s+fi\b": "hifi",
-    r"\blo\s+fi\b": "lofi",
-    r"\bsci\s+fi\b": "scifi",
-    r"\be\s+mail\b": "email",
-    r"\be\s+book\b": "ebook",
-    r"\be\s+commerce\b": "ecommerce",
-    r"\bx\s+ray\b": "xray",
-    r"\bt\s+shirt\b": "tshirt",
-    r"\ba\s+m\b": "am",
-    r"\bp\s+m\b": "pm",
-    r"\bo\s+k\b": "okay",
-    r"\bo+h+\b": "oh",
-    r"\booo+\b": "oh",
-}

evaluation/standard_asr/vendor/multilingual.py

@@ -1,94 +0,0 @@
-# Adapted from Hugging Face text scoring, Apache-2.0; see provenance.json.
-# Only dependency imports and unrelated data-loading/scoring code were removed.
-import re
-from difflib import SequenceMatcher
-import num2words
-from .normalizer import BasicMultilingualTextNormalizer
-FILLER_WORDS = {}
-
-class MultilingualNormalizer(BasicMultilingualTextNormalizer):
-    """BasicMultilingualTextNormalizer with optional number normalization.
-
-    Call with just text for standard normalization (backward-compatible).
-    Pass lang= to also convert digits to words via num2words and remove
-    language-specific filler words (see FILLER_WORDS).
-    """
-
-    def __init__(self, remove_diacritics: bool = True):
-        super().__init__(remove_diacritics)
-        # Pre-compile filler patterns. Each filler word is passed through the
-        # base normalization itself, so the pattern matches the normalized
-        # text exactly (base normalization may strip punctuation such as "…"
-        # or combining marks). Longest-first so that multi-word and longer
-        # variants match before their prefixes. Matched on whitespace
-        # boundaries ((?<!\S) / (?!\S)) rather than \b, which is unreliable
-        # next to combining marks.
-        self._filler_patterns = {}
-        base_normalize = super().__call__
-        for lang, words in FILLER_WORDS.items():
-            normalized_words = {base_normalize(w) for w in words}
-            normalized_words.discard("")
-            self._filler_patterns[lang] = re.compile(
-                r"(?<!\S)(?:"
-                + "|".join(re.escape(w) for w in sorted(normalized_words, key=len, reverse=True))
-                + r")(?!\S)"
-            )
-
-    def _remove_fillers(self, text, lang):
-        pattern = self._filler_patterns.get(lang)
-        if pattern is None:
-            return text
-        text = pattern.sub("", text)
-        return re.sub(r"\s+", " ", text).strip()
-
-    def _normalize_numbers(self, text, lang):
-        # Join space-separated thousand groups (e.g. "10 000" -> "10000")
-        text = re.sub(r"(\d)\s+(\d{3})\b", r"\1\2", text)
-
-        # Convert remaining digit sequences to words
-        def _replace(m):
-            try:
-                return num2words.num2words(int(m.group()), lang=lang)
-            except Exception:
-                return m.group()
-
-        return re.sub(r"\d+", _replace, text)
-
-    def __call__(self, s, lang=None):
-        s = super().__call__(s)
-        if lang is not None:
-            s = self._remove_fillers(s, lang)
-            s = self._normalize_numbers(s, lang)
-        return s
-
-def normalize_compound_pairs(refs, preds):
-    """Align compound word boundaries between ref/pred pairs.
-
-    When a mismatch region has identical characters ignoring whitespace,
-    normalize both sides to the joined form.
-    """
-    new_refs, new_preds = [], []
-    for ref_text, pred_text in zip(refs, preds):
-        ref_words = ref_text.split()
-        pred_words = pred_text.split()
-
-        sm = SequenceMatcher(None, ref_words, pred_words)
-        new_rw, new_pw = [], []
-
-        for tag, i1, i2, j1, j2 in sm.get_opcodes():
-            if tag == "equal":
-                new_rw.extend(ref_words[i1:i2])
-                new_pw.extend(pred_words[j1:j2])
-            else:
-                rc = "".join(ref_words[i1:i2])
-                pc = "".join(pred_words[j1:j2])
-                if rc == pc:
-                    new_rw.append(rc)
-                    new_pw.append(pc)
-                else:
-                    new_rw.extend(ref_words[i1:i2])
-                    new_pw.extend(pred_words[j1:j2])
-
-        new_refs.append(" ".join(new_rw))
-        new_preds.append(" ".join(new_pw))
-    return new_refs, new_preds

evaluation/standard_asr/vendor/normalizer.py

@@ -1,753 +0,0 @@
-# Copyright 2022 The OpenAI team and The HuggingFace Team. All rights reserved.
-# Most of the code is copy pasted from the original whisper repository
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import re
-import unicodedata
-from fractions import Fraction
-from typing import Iterator, List, Match, Optional, Union
-from .english_abbreviations import english_name_normalizer, english_spelling_normalizer, english_compound_normalizer
-
-import regex
-
-
-# non-ASCII letters that are not separated by "NFKD" normalization
-ADDITIONAL_DIACRITICS = {
-    "œ": "oe",
-    "Œ": "OE",
-    "ø": "o",
-    "Ø": "O",
-    "æ": "ae",
-    "Æ": "AE",
-    "ß": "ss",
-    "ẞ": "SS",
-    "đ": "d",
-    "Đ": "D",
-    "ð": "d",
-    "Ð": "D",
-    "þ": "th",
-    "Þ": "th",
-    "ł": "l",
-    "Ł": "L",
-}
-
-
-def remove_symbols_and_diacritics(s: str, keep=""):
-    """
-    Replace any other markers, symbols, and punctuations with a space, and drop any diacritics (category 'Mn' and some
-    manual mappings)
-    """
-
-    def replace_character(char):
-        if char in keep:
-            return char
-        elif char in ADDITIONAL_DIACRITICS:
-            return ADDITIONAL_DIACRITICS[char]
-
-        elif unicodedata.category(char) == "Mn":
-            return ""
-
-        elif unicodedata.category(char)[0] in "MSP":
-            return " "
-
-        return char
-
-    return "".join(replace_character(c) for c in unicodedata.normalize("NFKD", s))
-
-
-def remove_symbols(s: str):
-    """
-    Replace any other markers, symbols, punctuations with a space, keeping diacritics
-    """
-    return "".join(" " if unicodedata.category(c)[0] in "MSP" else c for c in unicodedata.normalize("NFKC", s))
-
-
-def remove_symbols_keep_marks(s: str):
-    """
-    Replace symbols and punctuation with a space, keeping combining marks.
-
-    Unlike `remove_symbols`, combining marks (category 'M') are preserved. This
-    is required for scripts like Devanagari, where vowel signs (matras) and the
-    virama are combining marks that are integral to words.
-    """
-    return "".join(" " if unicodedata.category(c)[0] in "SP" else c for c in unicodedata.normalize("NFKC", s))
-
-
-class BasicTextNormalizer:
-    def __init__(self, remove_diacritics: bool = False, split_letters: bool = False):
-        self.clean = remove_symbols_and_diacritics if remove_diacritics else remove_symbols
-        self.split_letters = split_letters
-
-    def __call__(self, s: str):
-        s = s.lower()
-        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
-        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
-        s = self.clean(s).lower()
-
-        if self.split_letters:
-            s = " ".join(regex.findall(r"\X", s, regex.U))
-
-        s = re.sub(r"\s+", " ", s)  # replace any successive whitespace characters with a space
-
-        return s
-
-
-class BasicMultilingualTextNormalizer:
-    def __init__(self, remove_diacritics: bool = True):
-        # When keeping diacritics, also keep combining marks (category 'M'):
-        # scripts like Devanagari encode vowel signs and the virama as
-        # combining marks, so stripping them mangles words.
-        self.clean = remove_symbols_and_diacritics if remove_diacritics else remove_symbols_keep_marks
-
-    def __call__(self, s: str):
-        s = s.lower()
-        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
-        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
-        s = self.clean(s).lower()
-
-        # Remove punctuation and extra spaces
-        s = regex.sub(r"[^\w\s]", "", s)
-        s = re.sub(r"\s+", " ", s).strip()
-
-        return s
-
-
-class EnglishNumberNormalizer:
-    """
-    Convert any spelled-out numbers into arabic numbers, while handling:
-
-    - remove any commas
-    - keep the suffixes such as: `1960s`, `274th`, `32nd`, etc.
-    - spell out currency symbols after the number. e.g. `$20 million` -> `20000000 dollars`
-    - spell out `one` and `ones`
-    - interpret successive single-digit numbers as nominal: `one oh one` -> `101`
-    """
-
-    def __init__(self):
-        super().__init__()
-
-        self.zeros = {"o", "oh", "zero"}
-        # fmt: off
-        self.ones = {
-            name: i
-            for i, name in enumerate(
-                ["one", "two", "three", "four", "five", "six", "seven", "eight", "nine", "ten", "eleven", "twelve", "thirteen", "fourteen", "fifteen", "sixteen", "seventeen", "eighteen", "nineteen"],
-                start=1,
-            )
-        }
-        # fmt: on
-        self.ones_plural = {
-            "sixes" if name == "six" else name + "s": (value, "s") for name, value in self.ones.items()
-        }
-        self.ones_ordinal = {
-            "zeroth": (0, "th"),
-            "first": (1, "st"),
-            "second": (2, "nd"),
-            "third": (3, "rd"),
-            "fifth": (5, "th"),
-            "twelfth": (12, "th"),
-            **{
-                name + ("h" if name.endswith("t") else "th"): (value, "th")
-                for name, value in self.ones.items()
-                if value > 3 and value != 5 and value != 12
-            },
-        }
-        self.ones_suffixed = {**self.ones_plural, **self.ones_ordinal}
-
-        self.tens = {
-            "twenty": 20,
-            "thirty": 30,
-            "forty": 40,
-            "fifty": 50,
-            "sixty": 60,
-            "seventy": 70,
-            "eighty": 80,
-            "ninety": 90,
-        }
-        self.tens_plural = {name.replace("y", "ies"): (value, "s") for name, value in self.tens.items()}
-        self.tens_ordinal = {name.replace("y", "ieth"): (value, "th") for name, value in self.tens.items()}
-        self.tens_suffixed = {**self.tens_plural, **self.tens_ordinal}
-
-        self.multipliers = {
-            "hundred": 100,
-            "thousand": 1_000,
-            "million": 1_000_000,
-            "billion": 1_000_000_000,
-            "trillion": 1_000_000_000_000,
-            "quadrillion": 1_000_000_000_000_000,
-            "quintillion": 1_000_000_000_000_000_000,
-            "sextillion": 1_000_000_000_000_000_000_000,
-            "septillion": 1_000_000_000_000_000_000_000_000,
-            "octillion": 1_000_000_000_000_000_000_000_000_000,
-            "nonillion": 1_000_000_000_000_000_000_000_000_000_000,
-            "decillion": 1_000_000_000_000_000_000_000_000_000_000_000,
-        }
-        self.multipliers_plural = {name + "s": (value, "s") for name, value in self.multipliers.items()}
-        self.multipliers_ordinal = {name + "th": (value, "th") for name, value in self.multipliers.items()}
-        self.multipliers_suffixed = {**self.multipliers_plural, **self.multipliers_ordinal}
-        self.decimals = {*self.ones, *self.tens, *self.zeros}
-
-        self.preceding_prefixers = {
-            "minus": "-",
-            "negative": "-",
-            "plus": "+",
-            "positive": "+",
-        }
-        self.following_prefixers = {
-            "pound": "£",
-            "pounds": "£",
-            "euro": "€",
-            "euros": "€",
-            "dollar": "$",
-            "dollars": "$",
-            "cent": "¢",
-            "cents": "¢",
-        }
-        self.prefixes = set(list(self.preceding_prefixers.values()) + list(self.following_prefixers.values()))
-        self.suffixers = {
-            "per": {"cent": "%"},
-            "percent": "%",
-        }
-        self.specials = {"and", "double", "triple", "point"}
-
-        self.words = {
-            key
-            for mapping in [
-                self.zeros,
-                self.ones,
-                self.ones_suffixed,
-                self.tens,
-                self.tens_suffixed,
-                self.multipliers,
-                self.multipliers_suffixed,
-                self.preceding_prefixers,
-                self.following_prefixers,
-                self.suffixers,
-                self.specials,
-            ]
-            for key in mapping
-        }
-        self.literal_words = {"one", "ones"}
-
-    def process_words(self, words: List[str]) -> Iterator[str]:
-        prefix: Optional[str] = None
-        value: Optional[Union[str, int]] = None
-        skip = False
-
-        def to_fraction(s: str):
-            try:
-                return Fraction(s)
-            except ValueError:
-                return None
-
-        def is_digit_token(token: Optional[str]) -> bool:
-            """True for tokens that continue a digit sequence ("four", "oh", "20")."""
-            return token is not None and bool(
-                re.match(r"^\d+$", token)
-                or token in self.zeros
-                or token in self.ones
-                or token in self.tens
-            )
-
-        def output(result: Union[str, int]):
-            nonlocal prefix, value
-            result = str(result)
-            if prefix is not None:
-                result = prefix + result
-            value = None
-            prefix = None
-            return result
-
-        if len(words) == 0:
-            return
-
-        for i, current in enumerate(words):
-            prev = words[i - 1] if i != 0 else None
-            next = words[i + 1] if i != len(words) - 1 else None
-            if skip:
-                skip = False
-                continue
-
-            next_is_numeric = next is not None and re.match(r"^\d+(\.\d+)?$", next)
-            has_prefix = current[0] in self.prefixes
-            current_without_prefix = current[1:] if has_prefix else current
-            if re.match(r"^\d+(\.\d+)?$", current_without_prefix):
-                # arabic numbers (potentially with signs and fractions)
-                f = to_fraction(current_without_prefix)
-                if f is None:
-                    raise ValueError("Converting the fraction failed")
-
-                if value is not None:
-                    if isinstance(value, str) and value.endswith("."):
-                        # concatenate decimals / ip address components
-                        value = str(value) + str(current)
-                        continue
-                    else:
-                        yield output(value)
-
-                prefix = current[0] if has_prefix else prefix
-                if f.denominator == 1:
-                    value = f.numerator  # store integers as int
-                else:
-                    value = current_without_prefix
-            elif current not in self.words:
-                # non-numeric words
-                if value is not None:
-                    yield output(value)
-                yield output(current)
-            elif current in self.zeros:
-                # "oh" is far more often the interjection than a spoken zero, so
-                # read it as a digit only inside a digit sequence: the sequence
-                # has to continue on the right ("four oh one", "nineteen oh
-                # five"), and with nothing pending on the left it takes two more
-                # digit tokens, i.e. a serial/phone-style reading ("oh seven nine
-                # eight"). On its own — including the doubled "oh oh" — it stays
-                # a word. "o" and "zero" are unchanged.
-                next2 = words[i + 2] if i + 2 < len(words) else None
-                in_number = is_digit_token(next) and (
-                    value is not None or is_digit_token(next2)
-                )
-                if current == "oh" and not in_number:
-                    if value is not None:
-                        yield output(value)  # don't drop a pending number
-                    yield output(current)
-                else:
-                    value = str(value or "") + "0"
-            elif current in self.ones:
-                ones = self.ones[current]
-
-                if value is None:
-                    value = ones
-                elif isinstance(value, str) or prev in self.ones:
-                    if prev in self.tens and ones < 10:  # replace the last zero with the digit
-                        value = value[:-1] + str(ones)
-                    else:
-                        value = str(value) + str(ones)
-                elif ones < 10:
-                    if value % 10 == 0:
-                        value += ones
-                    else:
-                        value = str(value) + str(ones)
-                else:  # eleven to nineteen
-                    if value % 100 == 0:
-                        value += ones
-                    else:
-                        value = str(value) + str(ones)
-            elif current in self.ones_suffixed:
-                # ordinal or cardinal; yield the number right away
-                ones, suffix = self.ones_suffixed[current]
-                if value is None:
-                    yield output(str(ones) + suffix)
-                elif isinstance(value, str) or prev in self.ones:
-                    if prev in self.tens and ones < 10:
-                        yield output(value[:-1] + str(ones) + suffix)
-                    else:
-                        yield output(str(value) + str(ones) + suffix)
-                elif ones < 10:
-                    if value % 10 == 0:
-                        yield output(str(value + ones) + suffix)
-                    else:
-                        yield output(str(value) + str(ones) + suffix)
-                else:  # eleven to nineteen
-                    if value % 100 == 0:
-                        yield output(str(value + ones) + suffix)
-                    else:
-                        yield output(str(value) + str(ones) + suffix)
-                value = None
-            elif current in self.tens:
-                tens = self.tens[current]
-                if value is None:
-                    value = tens
-                elif isinstance(value, str):
-                    value = str(value) + str(tens)
-                else:
-                    if value % 100 == 0:
-                        value += tens
-                    else:
-                        value = str(value) + str(tens)
-            elif current in self.tens_suffixed:
-                # ordinal or cardinal; yield the number right away
-                tens, suffix = self.tens_suffixed[current]
-                if value is None:
-                    yield output(str(tens) + suffix)
-                elif isinstance(value, str):
-                    yield output(str(value) + str(tens) + suffix)
-                else:
-                    if value % 100 == 0:
-                        yield output(str(value + tens) + suffix)
-                    else:
-                        yield output(str(value) + str(tens) + suffix)
-            elif current in self.multipliers:
-                multiplier = self.multipliers[current]
-                if value is None:
-                    value = multiplier
-                elif isinstance(value, str) or value == 0:
-                    f = to_fraction(value)
-                    p = f * multiplier if f is not None else None
-                    if f is not None and p.denominator == 1:
-                        value = p.numerator
-                    else:
-                        yield output(value)
-                        value = multiplier
-                else:
-                    before = value // 1000 * 1000
-                    residual = value % 1000
-                    value = before + residual * multiplier
-            elif current in self.multipliers_suffixed:
-                multiplier, suffix = self.multipliers_suffixed[current]
-                if value is None:
-                    yield output(str(multiplier) + suffix)
-                elif isinstance(value, str):
-                    f = to_fraction(value)
-                    p = f * multiplier if f is not None else None
-                    if f is not None and p.denominator == 1:
-                        yield output(str(p.numerator) + suffix)
-                    else:
-                        yield output(value)
-                        yield output(str(multiplier) + suffix)
-                else:  # int
-                    before = value // 1000 * 1000
-                    residual = value % 1000
-                    value = before + residual * multiplier
-                    yield output(str(value) + suffix)
-                value = None
-            elif current in self.preceding_prefixers:
-                # apply prefix (positive, minus, etc.) if it precedes a number
-                if value is not None:
-                    yield output(value)
-
-                if next in self.words or next_is_numeric:
-                    prefix = self.preceding_prefixers[current]
-                else:
-                    yield output(current)
-            elif current in self.following_prefixers:
-                # apply prefix (dollars, cents, etc.) only after a number
-                if value is not None:
-                    prefix = self.following_prefixers[current]
-                    yield output(value)
-                else:
-                    yield output(current)
-            elif current in self.suffixers:
-                # apply suffix symbols (percent -> '%')
-                if value is not None:
-                    suffix = self.suffixers[current]
-                    if isinstance(suffix, dict):
-                        if next in suffix:
-                            yield output(str(value) + suffix[next])
-                            skip = True
-                        else:
-                            yield output(value)
-                            yield output(current)
-                    else:
-                        yield output(str(value) + suffix)
-                else:
-                    yield output(current)
-            elif current in self.specials:
-                if next not in self.words and not next_is_numeric:
-                    # apply special handling only if the next word can be numeric
-                    if value is not None:
-                        yield output(value)
-                    yield output(current)
-                elif current == "and":
-                    # ignore "and" after hundreds, thousands, etc.
-                    if prev not in self.multipliers:
-                        if value is not None:
-                            yield output(value)
-                        yield output(current)
-                elif current == "double" or current == "triple":
-                    if next in self.ones or next in self.zeros:
-                        repeats = 2 if current == "double" else 3
-                        ones = self.ones.get(next, 0)
-                        value = str(value or "") + str(ones) * repeats
-                        skip = True
-                    else:
-                        if value is not None:
-                            yield output(value)
-                        yield output(current)
-                elif current == "point":
-                    if next in self.decimals or next_is_numeric:
-                        value = str(value or "") + "."
-                else:
-                    # should all have been covered at this point
-                    raise ValueError(f"Unexpected token: {current}")
-            else:
-                # all should have been covered at this point
-                raise ValueError(f"Unexpected token: {current}")
-
-        if value is not None:
-            yield output(value)
-
-    def preprocess(self, s: str):
-        # replace "<number> and a half" with "<number> point five"
-        results = []
-
-        segments = re.split(r"\band\s+a\s+half\b", s)
-        for i, segment in enumerate(segments):
-            if len(segment.strip()) == 0:
-                continue
-            if i == len(segments) - 1:
-                results.append(segment)
-            else:
-                results.append(segment)
-                last_word = segment.rsplit(maxsplit=2)[-1]
-                if last_word in self.decimals or last_word in self.multipliers:
-                    results.append("point five")
-                else:
-                    results.append("and a half")
-
-        s = " ".join(results)
-
-        # put a space at number/letter boundary
-        s = re.sub(r"([a-z])([0-9])", r"\1 \2", s)
-        s = re.sub(r"([0-9])([a-z])", r"\1 \2", s)
-
-        # but remove spaces which could be a suffix
-        s = re.sub(r"([0-9])\s+(st|nd|rd|th|s)\b", r"\1\2", s)
-
-        return s
-
-    def postprocess(self, s: str):
-        def combine_cents(m: Match):
-            try:
-                currency = m.group(1)
-                integer = m.group(2)
-                cents = int(m.group(3))
-                return f"{currency}{integer}.{cents:02d}"
-            except ValueError:
-                return m.string
-
-        def extract_cents(m: Match):
-            try:
-                return f"¢{int(m.group(1))}"
-            except ValueError:
-                return m.string
-
-        # apply currency postprocessing; "$2 and ¢7" -> "$2.07"
-        s = re.sub(r"([€£$])([0-9]+) (?:and )?¢([0-9]{1,2})\b", combine_cents, s)
-        s = re.sub(r"[€£$]0.([0-9]{1,2})\b", extract_cents, s)
-
-        # write "one(s)" instead of "1(s)", just for the readability
-        s = re.sub(r"\b1(s?)\b", r"one\1", s)
-
-        return s
-
-    def __call__(self, s: str):
-        s = self.preprocess(s)
-        s = " ".join(word for word in self.process_words(s.split()) if word is not None)
-        s = self.postprocess(s)
-
-        return s
-
-
-class EnglishSpellingNormalizer:
-    """
-    Applies British-American spelling mappings as listed in [1].
-
-    [1] https://www.tysto.com/uk-us-spelling-list.html
-    """
-
-    def __init__(self, english_spelling_mapping):
-        self.mapping = english_spelling_mapping
-
-    def __call__(self, s: str):
-        return " ".join(self.mapping.get(word, word) for word in s.split())
-
-
-class EnglishAcronymNormalizer:
-    """
-    Collapse sequences of single-character tokens (letters or digits) into single words.
-
-    This normalizes acronym spacing so that both spaced-out and joined forms match:
-        - "b b c" -> "bbc"
-        - "5 g" -> "5g"
-
-    Lone single-character words surrounded by multi-character words are left untouched
-    (e.g. "a big cat" stays "a big cat").
-    """
-
-    def __call__(self, s: str) -> str:
-        words = s.split()
-        result = []
-        i = 0
-        while i < len(words):
-            if len(words[i]) == 1 and words[i].isalnum():
-                # Start of a potential acronym run
-                run = [words[i]]
-                j = i + 1
-                while j < len(words) and len(words[j]) == 1 and words[j].isalnum():
-                    run.append(words[j])
-                    j += 1
-                # Require 3+ tokens if the run contains common words "a" or "i",
-                # otherwise 2+ is enough (e.g. "5 g" -> "5g")
-                has_common_word = any(c in ("a", "i") for c in run)
-                min_run = 3 if has_common_word else 2
-                if len(run) >= min_run:
-                    result.append("".join(run))
-                else:
-                    result.extend(run)
-                i = j
-            else:
-                result.append(words[i])
-                i += 1
-        return " ".join(result)
-
-
-class EnglishNameNormalizer:
-    """
-    Collapse common name spelling variants to a single canonical form.
-
-    This is intentionally conservative and token-based so it can be extended
-    with project-specific aliases when needed.
-    """
-
-    def __init__(self, english_name_mapping=english_name_normalizer):
-        self.mapping = english_name_mapping
-
-    def __call__(self, s: str):
-        return " ".join(self.mapping.get(word, word) for word in s.split())
-
-
-class EnglishTextNormalizer:
-    def __init__(self, english_spelling_mapping=english_spelling_normalizer):
-        # Filler words / hesitations to remove. Written as regexes so that
-        # arbitrary elongation is covered without enumerating every spelling
-        # ("uh", "uhh", "uuuh", "uhhhh", ...). Each alternative is wrapped in
-        # \b...\b below, so a shorter alternative cannot match a prefix of a
-        # longer token and the order of the single-token patterns is irrelevant.
-        # Hyphens are word boundaries too, which is why most hyphenated forms
-        # need no entry ("um-hmm" is matched as "um" + "hmm") — but any whose
-        # halves are not both fillers must be listed *before* the patterns,
-        # otherwise only the first half is matched ("ah-ha" -> "ha").
-        filler_words = [
-            "ah-ha",         # "ha" alone is not a filler, so match the pair first
-            r"a+h+m*",       # ah, aah, ahh, ahhh, aaah, ahm, ahmm
-            r"a+h+a+",       # aha, ahaa, ahaaa
-            r"e+h+m*",       # eh, ehh, eeeh, ehhh, ehm, ehmm
-            r"e+m+",         # em, emm
-            r"e+r+m*",       # er, err, errr, erm
-            r"h+a+h+",       # hah, hahh
-            r"h+e+h+",       # heh, hehh
-            r"h+m+",         # hm, hmm, hmmm, hhm
-            r"h+u+h+",       # huh, huhh
-            r"m{2,}",        # mm, mmm, mmmm
-            r"m+h+m*",       # mh, mhm, mhmm, mmhm
-            r"t+s+k+",       # tsk
-            r"u+g+h+",       # ugh, uuugh
-            r"u+h+m*",       # uh, uuh, uhh, uhhh, uuuh, uhm, uuuhm
-            r"u+h+u+[hm]*",  # uhuh, uhum
-            r"u+m+h*",       # um, umm, ummm, uuum, umh
-            # Irregular forms, not worth a pattern of their own.
-            "ahem", "eheh", "ehehe", "ehr", "hmmph", "hum", "hunh", "mhum", "mmkay",
-        ]
-        self.ignore_patterns = r"\b(" + "|".join(filler_words) + r")\b"
-        self.replacers = {
-            # Bare o'clock times: the ":00" is not spoken as words, so drop it
-            # ("2:00 AM" -> "2 am"). Applied here, while the colon is still
-            # present, so that a time is distinguishable from an unrelated
-            # digit sequence — by the time symbols are stripped "3:00" and
-            # "3 00" look alike. Without this the minutes are absorbed into the
-            # hour ("3:00" -> "30") or left as a stray token ("11:00" -> "11 0").
-            r"\b(\d{1,2}):00\b": r"\1",
-            # common contractions
-            r"\bwon't\b": "will not",
-            r"\bcan't\b": "can not",
-            r"\blet's\b": "let us",
-            r"\bain't\b": "aint",
-            r"\by'all\b": "you all",
-            r"\bwanna\b": "want to",
-            r"\bgotta\b": "got to",
-            r"\bgonna\b": "going to",
-            r"\bi'ma\b": "i am going to",
-            r"\bimma\b": "i am going to",
-            r"\bwoulda\b": "would have",
-            r"\bcoulda\b": "could have",
-            r"\bshoulda\b": "should have",
-            r"\bma'am\b": "madam",
-            # contractions in titles/prefixes
-            r"\bmr\b": "mister ",
-            r"\bmrs\b": "missus ",
-            r"\bst\b": "saint ",
-            r"\bdr\b": "doctor ",
-            r"\bprof\b": "professor ",
-            r"\bcapt\b": "captain ",
-            r"\bgov\b": "governor ",
-            r"\bald\b": "alderman ",
-            r"\bgen\b": "general ",
-            r"\bsen\b": "senator ",
-            r"\brep\b": "representative ",
-            r"\bpres\b": "president ",
-            r"\brev\b": "reverend ",
-            r"\bhon\b": "honorable ",
-            r"\basst\b": "assistant ",
-            r"\bassoc\b": "associate ",
-            r"\blt\b": "lieutenant ",
-            r"\bcol\b": "colonel ",
-            r"\bjr\b": "junior ",
-            r"\bsr\b": "senior ",
-            r"\besq\b": "esquire ",
-            # prefect tenses, ideally it should be any past participles, but it's harder..
-            r"'d been\b": " had been",
-            r"'s been\b": " has been",
-            r"'d gone\b": " had gone",
-            r"'s gone\b": " has gone",
-            r"'d done\b": " had done",  # "'s done" is ambiguous
-            r"'s got\b": " has got",
-            # general contractions
-            r"n't\b": " not",
-            r"'re\b": " are",
-            r"\b(it|he|she|what|that|who|here|there|how|when|where|why|this)'s\b": r"\1 is",
-            r"'d\b": " would",
-            r"'ll\b": " will",
-            r"'t\b": " not",
-            r"'ve\b": " have",
-            r"'m\b": " am",
-        }
-        self.standardize_numbers = EnglishNumberNormalizer()
-        self.standardize_spellings = EnglishSpellingNormalizer(english_spelling_mapping)
-        self.standardize_names = EnglishNameNormalizer()
-        self.standardize_acronyms = EnglishAcronymNormalizer()
-        # Multi-word compound mappings — defined in english_abbreviations.py
-        self.compound_words = english_compound_normalizer
-
-    def __call__(self, s: str):
-        s = s.lower()
-
-        s = re.sub(r"[<\[][^>\]]*[>\]]", "", s)  # remove words between brackets
-        s = re.sub(r"\(([^)]+?)\)", "", s)  # remove words between parenthesis
-        s = re.sub(self.ignore_patterns, "", s)
-        s = re.sub(r"\s+'", "'", s)  # standardize when there's a space before an apostrophe
-
-        for pattern, replacement in self.replacers.items():
-            s = re.sub(pattern, replacement, s)
-
-        s = re.sub(r"(\d),(\d)", r"\1\2", s)  # remove commas between digits
-        s = re.sub(r"\.([^0-9]|$)", r" \1", s)  # remove periods not followed by numbers
-        s = remove_symbols_and_diacritics(s, keep=".%$¢€£")  # keep some symbols for numerics
-
-        # Normalize hardcoded compound words (e.g. "wi fi" -> "wifi" after hyphen removal)
-        for pattern, replacement in self.compound_words.items():
-            s = re.sub(pattern, replacement, s)
-
-        s = self.standardize_numbers(s)
-        s = self.standardize_spellings(s)
-        s = self.standardize_names(s)
-        s = self.standardize_acronyms(s)
-
-        # now remove prefix/suffix symbols that are not preceded/followed by numbers
-        s = re.sub(r"[.$¢€£]([^0-9])", r" \1", s)
-        s = re.sub(r"([^0-9])%", r"\1 ", s)
-
-        s = re.sub(r"\s+", " ", s)  # replace any successive whitespace characters with a space
-
-        return s

evaluation/standard_asr/vendor/provenance.json

@@ -1,23 +0,0 @@
-{
-  "repository": "https://github.com/huggingface/open_asr_leaderboard",
-  "revision": "48219c6028db0517d704600d92f31edfc96e8c23",
-  "license": "Apache-2.0",
-  "upstream_sha256": {
-    "normalizer/normalizer.py": "490b56393484ef386b486f3679cc2261264af280b3512154431e0bdcf2778295",
-    "normalizer/english_abbreviations.py": "52997cc963e0bd6568d15554ad8d5fb0759f20f361483632a22f90d2d0e07db1",
-    "normalizer/data_utils.py": "738f362b477ee357f743a567b8bbd69fac49e56dce0a477cbdd52dec863842b8",
-    "normalizer/eval_utils.py": "27138d5884e39f5e6422e1dc48ce60dc2f5c15a56e63117115d7e6088d29e449"
-  },
-  "vendored_sha256": {
-    "LICENSE": "5ee13882fce0975f0ad3c3d5c2042af4896c2a669be2d6df5b3d256811267c85",
-    "english_abbreviations.py": "52997cc963e0bd6568d15554ad8d5fb0759f20f361483632a22f90d2d0e07db1",
-    "__init__.py": "8c39e42af3204c41c564c970e24525da160663836b54acc1b629e4342f862334",
-    "multilingual.py": "4c8db6c4171c8c2f5567d2ed625a3a1f0d45156af45ca42d9555c0c9b61a29d5",
-    "normalizer.py": "490b56393484ef386b486f3679cc2261264af280b3512154431e0bdcf2778295"
-  },
-  "modifications": {
-    "normalizer.py": "None",
-    "english_abbreviations.py": "None",
-    "multilingual.py": "Exact MultilingualNormalizer and normalize_compound_pairs definitions extracted; imports minimized, empty FILLER_WORDS preserved."
-  }
-}