README.md

metadata

license: other
language:
  - en
task_categories:
  - automatic-speech-recognition
tags:
  - dysarthria
  - speech-accessibility
pretty_name: SAPC2

SAPC2

Speech Accessibility Project corpus, second release, repackaged for training.

Audio is 16 kHz mono WAV. Speakers have a range of etiologies (Parkinson's disease, ALS, cerebral palsy, Down syndrome, stroke).

Access is restricted and inherited from the upstream Speech Accessibility Project data use agreement. Do not redistribute.

Layout

One tar per contributor, rather than one archive per split, so a run can fetch a subset and an interrupted download resumes cheaply.

Train/<contributor-id>.tar    1,157 archives
Dev/<contributor-id>.tar        155 archives
manifest/                       upstream CSV / TXT / reference transcripts
sapc2_train.jsonl               336,075 utterances
sapc2_dev.jsonl                  47,929 utterances

Each archive expands to <contributor-id>/, holding that speaker's WAV files plus a <contributor-id>.json metadata file:

<contributor-id>/
├── <contributor-id>.json
├── <contributor-id>_101_4308.wav
└── ...

Download and extract

pip install "huggingface_hub[hf_transfer]"
export HF_HUB_ENABLE_HF_TRANSFER=1

CORPUS=/path/to/corpus/sapc2
hf download dys-asr/sapc2 --repo-type dataset --local-dir "$CORPUS"

for split in Train Dev; do
  find "$CORPUS/$split" -maxdepth 1 -name '*.tar' -print0 |
    xargs -0 -P 16 -I {} bash -c 'tar -xf "$1" -C "$(dirname "$1")" && rm -f "$1"' _ {}
done

To fetch a single speaker instead of the whole corpus:

hf download dys-asr/sapc2 Dev/0a71bb2c-5579-468f-3729-08dc13d38337.tar \
  --repo-type dataset --local-dir "$CORPUS"

Manifests

Each JSONL line is one utterance:

{"audio": "/abs/path/sapc2/Train/<id>/<id>_1131_4218.wav", "text": "...", "duration": 8.59, "metadata": {...}}

audio holds absolute paths from the machine that generated the manifest. Rewrite the prefix to your own corpus root before training:

OLD=/scratch/u6sn/yangw.u6sn/corpus/audio/sapc2
python - "$OLD" "$CORPUS" <<'PY'
import json, sys
old, new = sys.argv[1], sys.argv[2]
for name in ("sapc2_train.jsonl", "sapc2_dev.jsonl"):
    with open(name) as src:
        rows = [json.loads(line) for line in src]
    for row in rows:
        row["audio"] = row["audio"].replace(old, new, 1)
    with open(name, "w") as dst:
        for row in rows:
            dst.write(json.dumps(row) + "\n")
PY

Then confirm every file resolved before starting a long job:

python -m ataxia.cli validate-manifest sapc2_train.jsonl sapc2_dev.jsonl \
  --check-audio --audio-check-workers 96