Parley speech models: standard (fp32) and compact (int8)

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+compact/diarization/mel_filters_128x257.f32 filter=lfs diff=lfs merge=lfs -text
+standard/asr/encoder.data filter=lfs diff=lfs merge=lfs -text
+standard/diarization/mel_filters_128x257.f32 filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,74 @@
+---
+license: other
+license_name: openmdw-1.1
+license_link: https://openmdw.ai/license/1-1/
+library_name: onnxruntime
+pipeline_tag: automatic-speech-recognition
+tags:
+  - onnx
+  - directml
+  - speaker-diarization
+  - streaming-asr
+  - multilingual
+base_model:
+  - nvidia/Nemotron-3-Diarization
+  - nvidia/nemotron-3.5-asr-streaming-0.6b
+---
+
+# Parley speech models
+
+ONNX models used by **Parley**, a desktop app for live speaker-aware transcription and
+translation. Parley downloads one variant on first run and runs it natively with ONNX Runtime
+(DirectML on any DirectX 12 GPU, CPU fallback).
+
+| Variant | Size | Precision | Best for |
+|---|---|---|---|
+| `standard/` | 2.99 GB | fp32 | GPUs (DirectML) |
+| `compact/` | 0.79 GB | int8 (dynamic, per-channel) | CPU-only machines |
+
+Each variant folder has a `manifest.json` with the size and SHA-256 of every file.
+
+## Contents
+
+### `diarization/` — NVIDIA Nemotron-3-Diarization (streaming Sortformer, up to 8 speakers)
+
+Exported from the 🤗 Transformers implementation of
+[nvidia/Nemotron-3-Diarization](https://huggingface.co/nvidia/Nemotron-3-Diarization):
+
+- `diar_embed.onnx` — feature stacking: log-mel `(1, T, 128)`, `T % 8 == 0` → embeddings `(1, T/8, 512)`
+- `diar_step.onnx` — encoder + head: embeddings `(1, L, 512)` and padding mask `(1, L)` →
+  speaker probabilities `(1, L, 8)` pooled to 80 ms frames. Right padding is exact (masked
+  attention, zeroed before the upsampling conv), so callers can use fixed shapes on DirectML.
+- `silence_embeds.f32` — learned silence embedding for the speaker cache
+- `diarization.json` — Arrival-Order Speaker Cache / FIFO parameters and streaming modes
+- `mel_filters_128x257.f32` — slaney mel filter bank (shared with the ASR front end)
+
+The streaming state (speaker cache + FIFO, with its score-based compression) is not part of the
+graphs; Parley implements it in Rust, ported from `Nemotron3DiarizationSpeakerCache`.
+Streaming output matches the Transformers reference exactly on the fp32 variant (100% frame
+agreement on the model card's example); the int8 variant agrees on 99.5% of frames.
+
+### `asr/` — NVIDIA Nemotron 3.5 ASR (cache-aware FastConformer RNN-T, 40 locales, 560 ms chunks)
+
+NeMo cache-aware streaming export of
+[nvidia/nemotron-3.5-asr-streaming-0.6b](https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b)
+by the [sherpa-onnx](https://github.com/k2-fsa/sherpa-onnx) project
+(`csukuangfj2/sherpa-onnx-nemotron-3.5-asr-streaming-0.6b-560ms[-int8]-2026-06-11`):
+`encoder.onnx` (+ `encoder.data`), `decoder.onnx`, `joiner.onnx`, `tokens.txt`. Chunk geometry,
+cache shapes and the language-prompt dictionary are in the encoder's metadata.
+
+## Front end
+
+Both models consume the same features: 16 kHz mono, pre-emphasis 0.97, 512-point FFT, 400-sample
+symmetric Hann window, hop 160, power spectrum, 128 slaney mel bins, `ln(x + 2^-24)`, no
+normalization.
+
+## License and attribution
+
+The model weights are NVIDIA's, released under the
+[OpenMDW License Agreement 1.1](https://openmdw.ai/license/1-1/); these are format conversions
+and quantizations of those weights, redistributed under the same terms.
+
+- Nemotron-3-Diarization and Nemotron 3.5 ASR © NVIDIA Corporation.
+- ASR ONNX export by the sherpa-onnx project (Apache-2.0 tooling).
+- Diarization export and quantization by Parley.

compact/asr/decoder.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19f9c98fc6d0a2c33a65a43b36fdb2e914c26c0aa9764be3aebc502a1e982fb0
+size 14978075

compact/asr/encoder.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:012e9321373af99021415e0b0eb3ec827b4be3153be6f30d9b448fe65e896e68
+size 657601403

compact/asr/joiner.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4101c7c679a0bc30483794b27a059e34e79232aa2068d78d51231a22c8b0d7ce
+size 9504438

compact/diarization/diar_embed.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11712eaf7b478f9c630c49272e56b98bb51d7f52c854cc684ebcf340a0a684df
+size 2103637

compact/diarization/diar_step.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c07b1af8344dd969875785ac71ee109c090cfbf86921881459e339489fcebf93
+size 101373191

compact/diarization/diarization.json

@@ -0,0 +1,28 @@
+{
+  "num_speakers": 8,
+  "subsampling_factor": 8,
+  "speaker_cache_length": 264,
+  "fifo_length": 264,
+  "speaker_cache_update_period": 222,
+  "silence_frames_per_speaker": 1,
+  "prediction_score_threshold": 0.25,
+  "latest_frames_score_boost": 0.05,
+  "min_positive_scores_rate": 0.5,
+  "strong_boost_rate": 0.75,
+  "weak_boost_rate": 1.5,
+  "hidden_size": 512,
+  "streaming_modes": {
+    "low_latency": [
+      9,
+      4
+    ],
+    "very_low_latency": [
+      6,
+      2
+    ],
+    "ultra_low_latency": [
+      3,
+      1
+    ]
+  }
+}

compact/diarization/mel_filters_128x257.f32

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5608e00763a20a191022fbb00a43346db5f1ec06828a97a6f4157f7d02b76c8
+size 131584

compact/manifest.json

@@ -0,0 +1,50 @@
+{
+  "variant": "compact",
+  "files": [
+    {
+      "path": "asr/decoder.onnx",
+      "size": 14978075,
+      "sha256": "19f9c98fc6d0a2c33a65a43b36fdb2e914c26c0aa9764be3aebc502a1e982fb0"
+    },
+    {
+      "path": "asr/encoder.onnx",
+      "size": 657601403,
+      "sha256": "012e9321373af99021415e0b0eb3ec827b4be3153be6f30d9b448fe65e896e68"
+    },
+    {
+      "path": "asr/joiner.onnx",
+      "size": 9504438,
+      "sha256": "4101c7c679a0bc30483794b27a059e34e79232aa2068d78d51231a22c8b0d7ce"
+    },
+    {
+      "path": "asr/tokens.txt",
+      "size": 131440,
+      "sha256": "729cc103155bafa785f9cd45746cd41cabe97eab7182fc04d594129587958f8a"
+    },
+    {
+      "path": "diarization/diar_embed.onnx",
+      "size": 2103637,
+      "sha256": "11712eaf7b478f9c630c49272e56b98bb51d7f52c854cc684ebcf340a0a684df"
+    },
+    {
+      "path": "diarization/diar_step.onnx",
+      "size": 101373191,
+      "sha256": "c07b1af8344dd969875785ac71ee109c090cfbf86921881459e339489fcebf93"
+    },
+    {
+      "path": "diarization/diarization.json",
+      "size": 563,
+      "sha256": "e191770b8264f145fc3c0359385ffd7e4fd3047ad87aabb9031c0e203dad744e"
+    },
+    {
+      "path": "diarization/mel_filters_128x257.f32",
+      "size": 131584,
+      "sha256": "f5608e00763a20a191022fbb00a43346db5f1ec06828a97a6f4157f7d02b76c8"
+    },
+    {
+      "path": "diarization/silence_embeds.f32",
+      "size": 2048,
+      "sha256": "d4417b3c0eabdf7c47032fac2b5b5a7ee83d819a6ddda8fd8eaf74e2b5cc4ac7"
+    }
+  ]
+}

standard/asr/decoder.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f9c59ee6fa130bc2ba349dbcbba7c74a4a960a98d4196295ba4e8e5d6bde6b68
+size 59764944

standard/asr/encoder.data

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7584f85df76bc9ae6fbdfa53aa8d97b07a842525d1c501d536d77fd9e4f57ac7
+size 2454405120

standard/asr/encoder.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1e540e03cb2cc57f5c0f6416b893768cfda23e756bfc0a2c4d502a1350325af
+size 42247484

standard/asr/joiner.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6bd74c0a31cbde0da0368c1e29d171752108c7ad1231630e079a7d97ceee6f0
+size 37824291

standard/diarization/diar_embed.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11712eaf7b478f9c630c49272e56b98bb51d7f52c854cc684ebcf340a0a684df
+size 2103637

standard/diarization/diar_step.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9e8403251f71d7307134e36f0d9b62cdc8a310def5ae650d19f142da5d232a6
+size 395677177

standard/diarization/diarization.json

@@ -0,0 +1,28 @@
+{
+  "num_speakers": 8,
+  "subsampling_factor": 8,
+  "speaker_cache_length": 264,
+  "fifo_length": 264,
+  "speaker_cache_update_period": 222,
+  "silence_frames_per_speaker": 1,
+  "prediction_score_threshold": 0.25,
+  "latest_frames_score_boost": 0.05,
+  "min_positive_scores_rate": 0.5,
+  "strong_boost_rate": 0.75,
+  "weak_boost_rate": 1.5,
+  "hidden_size": 512,
+  "streaming_modes": {
+    "low_latency": [
+      9,
+      4
+    ],
+    "very_low_latency": [
+      6,
+      2
+    ],
+    "ultra_low_latency": [
+      3,
+      1
+    ]
+  }
+}

standard/diarization/mel_filters_128x257.f32

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5608e00763a20a191022fbb00a43346db5f1ec06828a97a6f4157f7d02b76c8
+size 131584

standard/manifest.json

@@ -0,0 +1,55 @@
+{
+  "variant": "standard",
+  "files": [
+    {
+      "path": "asr/decoder.onnx",
+      "size": 59764944,
+      "sha256": "f9c59ee6fa130bc2ba349dbcbba7c74a4a960a98d4196295ba4e8e5d6bde6b68"
+    },
+    {
+      "path": "asr/encoder.data",
+      "size": 2454405120,
+      "sha256": "7584f85df76bc9ae6fbdfa53aa8d97b07a842525d1c501d536d77fd9e4f57ac7"
+    },
+    {
+      "path": "asr/encoder.onnx",
+      "size": 42247484,
+      "sha256": "f1e540e03cb2cc57f5c0f6416b893768cfda23e756bfc0a2c4d502a1350325af"
+    },
+    {
+      "path": "asr/joiner.onnx",
+      "size": 37824291,
+      "sha256": "a6bd74c0a31cbde0da0368c1e29d171752108c7ad1231630e079a7d97ceee6f0"
+    },
+    {
+      "path": "asr/tokens.txt",
+      "size": 131440,
+      "sha256": "729cc103155bafa785f9cd45746cd41cabe97eab7182fc04d594129587958f8a"
+    },
+    {
+      "path": "diarization/diar_embed.onnx",
+      "size": 2103637,
+      "sha256": "11712eaf7b478f9c630c49272e56b98bb51d7f52c854cc684ebcf340a0a684df"
+    },
+    {
+      "path": "diarization/diar_step.onnx",
+      "size": 395677177,
+      "sha256": "d9e8403251f71d7307134e36f0d9b62cdc8a310def5ae650d19f142da5d232a6"
+    },
+    {
+      "path": "diarization/diarization.json",
+      "size": 563,
+      "sha256": "e191770b8264f145fc3c0359385ffd7e4fd3047ad87aabb9031c0e203dad744e"
+    },
+    {
+      "path": "diarization/mel_filters_128x257.f32",
+      "size": 131584,
+      "sha256": "f5608e00763a20a191022fbb00a43346db5f1ec06828a97a6f4157f7d02b76c8"
+    },
+    {
+      "path": "diarization/silence_embeds.f32",
+      "size": 2048,
+      "sha256": "d4417b3c0eabdf7c47032fac2b5b5a7ee83d819a6ddda8fd8eaf74e2b5cc4ac7"
+    }
+  ]
+}