chime-ami / README.md
ltuncay's picture
Publish source-preserving ami prepared candidate
7f613d1 verified
|
Raw History Blame Contribute Delete
7.36 kB
metadata
pretty_name: CHIME — AMI (prepared candidate)
license: other
tags:
  - audio
  - webdataset
  - chime
  - prepared-candidate
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test/*.tar
      - split: train
        path: data/train/*.tar
      - split: validation
        path: data/validation/*.tar
license_name: original-source-terms
license_link: https://groups.inf.ed.ac.uk/ami/download/

CHIME · AMI

This is a source-preserving prepared candidate, not a scientifically admitted CHIME benchmark release. It repackages the selected upstream release into split-specific WebDataset TAR shards with native audio and per-example JSON. A Parquet index permits metadata filtering without downloading audio shards.

Source and selected scope

  • Upstream: AMI.
  • Pinned source release: manual annotations 1.6.2 and author Mix-Headset sessions; observed 2026-10-28.
  • Selection: Exactly ES2002a (custom train), IS1000a (custom validation), TS3003a (custom test), complete author-provided Mix-Headset WAV sessions. Their 12 global speaker IDs are pairwise disjoint and frozen in ami.json. Preserve source-native 16 kHz audio and the original mixture view; no local downmix or utterance clipping. This custom three-scenario subset is not the standard full AMI benchmark split.
  • Original license/access conditions: CC-BY-4.0 Public authoritative HTTPS retrieval without registration or individually signed agreement. Upstream use and attribution conditions remain applicable. This acquisition recipe grants no additional publication rights.
  • Source notes: The author download chooser supplied these exact HTTPS URLs. Manual annotation ZIP CRC and all selected word timestamps were validated against the actual native session durations. The full annotation package remains provenance; native audio is limited to the three specified sessions.

Original authors, licensors and recording contributors retain their rights. This preparation was assembled for public noncommercial research access; this repository does not grant broader rights than the original sources. Preserve per-recording attribution and notices. provenance/ contains original annotations/docs/licenses, source provenance and the redistribution audit. FSD50K/Clotho audio has per-recording terms; Clotho captions retain the complete custom Tampere experimental/noncommercial license. The CHIME code license does not apply to these data.

Representation and preservation

There are 82,869 audio assets, 21,177,704,328 stored-audio bytes and 21,174,058,092 decoded-PCM-equivalent bytes. Native sample rate, duration and channel count are retained. No loudness normalization, silence trimming, resampling or downmixing is applied.

Each sample retains its native audio codec (WAV/FLAC/MP3/OGG or another supported native container). Each payload is accompanied by <key>.json in an ordinary TAR. The JSON contains split, stored audio metadata, provenance, source groups and source targets. manifest.jsonl also records shard/member offsets and checksums. metadata.parquet provides audio and shard columns, common source/group/license columns, and exact arbitrary annotations in metadata_json (decode with json.loads). It does not duplicate audio payloads. Keeping heterogeneous annotations as JSON preserves empty objects and differing target schemas without Arrow type coercion. release.json hashes the audio shards and record manifest, and publication.json inventories all published files including attribution assets.

CHiME-3 background recordings interleave six synchronized native mono PCM16 channels into one six-channel PCM16 WAV per session, with exact decoded-sample checks and an explicit channel map. The excluded CHiME speech corpus is not included. Other audio payloads are copied byte for byte.

Source splits

  • test: 12,377 original audio assets.
  • train: 56,609 original audio assets.
  • validation: 13,883 original audio assets.

These are original source partitions or explicitly unspecified/background partitions. ESC-50 retains five official folds; no fixed benchmark train/test assignment is invented. VocalSet's split is the supplied technique-classification singer protocol, not a universal task protocol. Background noise is not converted into artificial word or silence examples. Linked views/channels share provenance groups.

Scientific limitations

Scientific admission remains pending. Native file integrity and packaging checks do not establish label correctness, perceptual deduplication, contamination freedom or benchmark validity. Existing upstream annotation errata and ambiguity are retained/documented. The source audit found declared Freesound ID overlaps within some official folds/partitions and across ESC-50, FSD50K and Clotho; FSD50K train/validation also share uploaders. Use the supplied source IDs, group metadata and provenance/source-audit.json when defining an evaluation protocol. Do not claim these candidate splits are universally leakage-free. Scores from this repository alone are not official CHIME scores.

Loading

Pin an immutable Hub commit revision in experiments. A generic Hugging Face WebDataset loader can read the shard configurations:

from datasets import load_dataset
samples = load_dataset("ltuncay/chime-ami", revision="<commit-sha>", streaming=True)

For local source-native decoding and checksum verification, use CHIME's iter_release(path, split=..., verify_hashes=True). Its audio arrays have shape [frames, channels]; the evaluated model owns its preprocessing. Download only selected shards/index files when appropriate; benchmark execution must not silently fetch data.

Provenance and citation

Cite the upstream dataset and recording authors as specified in the original documentation under provenance/. This preparation does not claim authorship of the source recordings or annotations. Exact upstream objects, checksums, release selection, preparation metadata and source URLs are recorded in release.json. Source Zenodo metadata/citations are retained when available. See the CHIME project implementation brief and scientific paper for the evolving evaluation protocol.

CHIME v1 benchmark usage

Do not train on CHIME evaluation data: 12377 records in the release split(s) test, listed one key per line in provenance/benchmark-v1/eval-keys.txt. Models submitted to CHIME must not see these recordings (or their source recordings) during pretraining or fine-tuning.

Task Family Metric Folds Train Validation Test (eval)
ami-activity frame_multilabel segment_f1 1 train validation test
ami-asr asr wer 1 train validation test

Exact per-record partition assignments (including records excluded from training because they share a source with evaluation data, and digitally silent assets) are in provenance/benchmark-v1/<task>.jsonl; the frozen task definitions are in provenance/benchmark-v1/usage.json. Splits never depend on the evaluation seed.