resproj007/pathological_speech

Pathological Speech (TORGO + UA-Speech + LibriSpeech Normal)

Mixed-corpus speech dataset for training and evaluating controllable speech-synthesis and severity-classification models. Three corpora are merged with unified metadata so a single model can learn severity- and gender-conditioned generation without confounds.

Splits

Split	Rows	Bytes (parquet)
train	44737	5,109,103,443
test	1780	210,930,372

Test speakers are held out from training: leave-speaker-out evaluation is supported out of the box.

Corpora

Corpus	Train rows	Test rows	Speaker role
TORGO	10334	431	dysarthric + control
UA-Speech	28407	1069	dysarthric (cerebral palsy)
LibriSpeech	5996	280	Normal supplement (train-clean-100)

Why LibriSpeech?

Without it, every severity == Normal row came from TORGO controls, meaning a model could trivially learn Normal <-> TORGO acoustics instead of true prosodic normality. The LibriSpeech train-clean-100 supplement breaks that corpus×severity confound. All LibriSpeech rows are labelled severity="Normal", condition="Control", diagnosis="None".

Test-set composition

By severity

{
  "Mild": 421,
  "Moderate": 585,
  "Normal": 560,
  "Severe": 214
}

By speaker (corpus-prefixed)

{
  "UA-Speech_M10": 193,
  "UA-Speech_M05": 192,
  "UA-Speech_M09": 186,
  "UA-Speech_M07": 176,
  "UA-Speech_F04": 167,
  "UA-Speech_M01": 155,
  "TORGO_MC03": 82,
  "TORGO_FC03": 74,
  "TORGO_MC04": 68,
  "TORGO_MC02": 47,
  "TORGO_M03": 42,
  "TORGO_F03": 33,
  "TORGO_M01": 24,
  "TORGO_M04": 20,
  "TORGO_M05": 17,
  "TORGO_F01": 15,
  "TORGO_FC01": 9,
  "LibriSpeech_3242": 71,
  "LibriSpeech_6818": 70,
  "LibriSpeech_1263": 70,
  "LibriSpeech_6181": 69
}

LibriSpeech test speakers 3242, 6181, 6818, 1263 are disjoint from the LibriSpeech train pool and gender-balanced (M:140 / F:140).

Metadata canonicalization (applied to all shards)

Two cross-corpus artefacts are fixed at the dataset level so training code doesn't have to remap on the fly:

1. gender is always lowercase ("male" / "female"). Previously UA-Speech emitted "Male" / "Female" while TORGO emitted lowercase; naive grouping produced four bins.

2. speaker_id is prefixed with {corpus}_ (e.g. TORGO_M01, UA-Speech_M01, LibriSpeech_3242). TORGO and UA-Speech both use M01 / M05 for different humans. The prefix prevents speaker-invariance objectives, LSO splits, and per-speaker statistics from silently merging them.

3. severity is recomputed per speaker from the canonical mapping (TORGO: clinical severity from the TORGO speaker sheet; UA-Speech: intelligibility banding → Very-low (<=19%) -> Severe, Low (28-39%) / Mid (58-62%) -> Moderate, High (>=86%) -> Mild). This corrects a prior UA-Speech labelling bug where M01 was Mild and M09 / M10 were Severe.

Other columns (diagnosis, condition, intelligibility) are preserved verbatim from their source corpora.

Schema

Column	Type	Notes
audio	Audio(16 kHz)	WAV (PCM16, 16 kHz mono) across all corpora
text	string	transcript
speaker_id	string	{corpus}_{original_id}
corpus	string	TORGO, UA-Speech, or LibriSpeech
gender	string	male | female
condition	string	Dysarthric, Control, ...
diagnosis	string	free text ("Cerebral palsy", "None", ...)
severity	string	Normal, Mild, Moderate, Severe
intelligibility	string	corpus-specific label (may be empty)
duration	float64 (seconds)

Load

from datasets import load_dataset
ds = load_dataset("resproj007/pathological_speech")
print(ds)
# DatasetDict({
#   train: Dataset({ features: [...], num_rows: 44737 })
#   test : Dataset({ features: [...], num_rows: 1780  })
# })

Intended use

• Controllable pathological-speech synthesis (severity + gender conditioning).
• Severity classification with leave-speaker-out evaluation.
• Cross-corpus robustness analysis (TORGO↔UA-Speech).

Licensing

Derived from corpora released under research-use terms. Redistribution is intended for academic research; verify the TORGO, UA-Speech, and LibriSpeech licenses for your use case.