abnerh/TORGO-database

The TORGO Database: Acoustic and articulatory speech from speakers with dysarthria

Dataset Summary

• This database only includes the short words and restricted sentence portion of the TORGO dataset.
• For the full dataset which also includes non-words and unrestricted sentences please see: https://www.cs.toronto.edu/~complingweb/data/TORGO/torgo.html.
• Transcripts have been normalized to remove punctuation but casing has been left. Few transcripts only had 'xxx' as text, these were removed.
• About 5.5 hours of dysarthric speech data and 8 hours of healthy speech.

Short words
These are useful for studying speech acoustics without the need for word boundary detection. This category includes the following:

• Repetitions of the English digits, 'yes', 'no', 'up', 'down', 'left', 'right', 'forward', 'back', 'select', 'menu', and the international radio alphabet (e.g., 'alpha', 'bravo', 'charlie'). These words are useful for hypothetical command software for accessibility.
• 50 words from the the word intelligibility section of the Frenchay Dysarthria Assessment (Enderby, 1983).
• 360 words from the word intelligibility section of the Yorkston-Beukelman Assessment of Intelligibility of Dysarthric Speech (Yorkston and Beukelman, 1981).
• The 10 most common words in the British National Corpus.

Restricted sentences
In order to utilize lexical, syntactic, and semantic processing in ASR, full and syntactically correct sentences are recorded. These include the following:

• Preselected phoneme-rich sentences such as "The quick brown fox jumps over the lazy dog", "She had your dark suit in greasy wash water all year", and "Don't ask me to carry an oily rag like that."
• The Grandfather passage.
• 162 sentences from the sentence intelligibility section of the Yorkston-Beukelman Assessment of Intelligibility of Dysarthric Speech (Yorkston and Beukelman, 1981).
• The 460 TIMIT-derived sentences used as prompts in the MOCHA-TIMIT database (Wrench, 1999; Zue et al, 1989).

Dataset Structure

• Data points comprise the path to the audio file and its transcription.
• Additional fields include gender, speech status (dysarthria or healthy), duration, and speaker-level severity.
• No dev/test split is provided as there is no standard split for this dataset.
• Filenames are as follows:
  • speakerNumber_sessionNumber_micType_utteranceNumber.wav
    • Speaker number has the format of gender-speechStatus-speakerNumber (e.g. FC01 = Female control #1, M04 = Male dysarthric #4)

Severity

severity is a speaker-level clinical category from the Frenchay Dysarthria Assessment, not an utterance-level intelligibility score. Every row from a speaker has the same value. Dysarthric speakers follow Mengistu and Rudzicz (ICASSP 2011) and Joy and Umesh (Interspeech 2017). Control speakers were not assessed and are labeled typical.

Speakers	severity
F01, M01, M02, M04	severe
M05	moderate-severe
F03	moderate
F04, M03	mild
FC01, FC02, FC03, MC01, MC02, MC03, MC04	typical

from datasets import load_dataset

dataset = load_dataset("abnerh/TORGO-database")
print(dataset) 
DatasetDict({
    train: Dataset({
        features: ['audio', 'transcription', 'speech_status', 'gender', 'duration', 'severity'],
        num_rows: 16552
    })
})

dataset = load_dataset("abnerh/TORGO-database")
print(dataset['train'][0]) 
{'audio': {'path': 'FC01_1_arrayMic_0066.wav',
  'array': array([ 0.00125122,  0.00387573,  0.00115967, ...,  0.00149536,
         -0.00326538,  0.00027466]),
  'sampling_rate': 16000},
 'transcription': 'alpha',
 'speech_status': 'healthy',
 'gender': 'female',
 'duration': 3.3,
 'severity': 'typical'}

print(dataset['train'][12200]) 
{'audio': {'path': 'M02_1_headMic_0066.wav',
  'array': array([ 0.00115967,  0.00106812,  0.00091553, ..., -0.00073242,
         -0.00082397, -0.00054932]),
  'sampling_rate': 16000},
 'transcription': 'yet he still thinks as swiftly as ever',
 'speech_status': 'dysarthria',
 'gender': 'male',
 'duration': 7.605,
 'severity': 'severe'}

Use of this database is free for academic (non-profit) purposes. If you use these data in any publication, you must reference at least one of the following papers:

• Rudzicz, F., Hirst, G., Van Lieshout, P. (2012) Vocal tract representation in the recognition of cerebral palsied speech. The Journal of Speech, Language, and Hearing Research, 55(4):1190-1207, August.
• Rudzicz, F., Namasivayam, A.K., Wolff, T. (2012) The TORGO database of acoustic and articulatory speech from speakers with dysarthria. Language Resources and Evaluation, 46(4), pages 523--541. This may be the most informative of the database itself.
• Rudzicz, F.(2012) Using articulatory likelihoods in the recognition of dysarthric speech. Speech Communication, 54(3), March, pages 430--444.

The severity labels are taken from:

• Mengistu, K. T. and Rudzicz, F. (2011) Adapting acoustic and lexical models to dysarthric speech. Proc. ICASSP 2011.
• Joy, N. M. and Umesh, S. (2017) On improving acoustic models for TORGO dysarthric speech database. Proc. Interspeech 2017.