RAPNIC (Reconeixement Automàtic de la Parla No Intel·ligible en Català) is a Catalan speech corpus collected from individuals with speech disorders, specifically cerebral palsy and Down syndrome.

This dataset was collected to develop and improve automatic speech recognition (ASR) systems that are accessible to people with speech disorders who speak Catalan.

Dataset Statistics

Speakers: 100
Recordings: 1000
Total Duration: 1.33 hours
Sampling Rate: 16 kHz
Audio Format: WAV
Language: Catalan (multiple dialects)

Disorder Distribution

Síndrome de Down: 510 recordings
Paràlisi cerebral: 340 recordings
Sense resposta: 100 recordings
Altres trastorns de la parla: 40 recordings
NA: 10 recordings

Gender Distribution

Dona: 550 recordings
Home: 420 recordings
Sense resposta: 30 recordings

Dialect Distribution

Central (Barcelona, Tarragona): 650 recordings
Girona: 150 recordings
Nord-Occidental (Lleida, Tortosa): 180 recordings
NA: 20 recordings

Data Fields

audio: Audio file (WAV format, 16 kHz)
'audio_id': Identifier for each audio
speaker_id: Unique identifier for each speaker (anonymized)
filename: Original filename of the recording
task_id: Task/prompt identifier
transcription: Text that was read/spoken
clean_transcription: Text that was read/spoken in lowercase and without punctuation (? are kept)
original_duration: Duration in seconds before preprocessing
trimmed_duration: Duration in seconds after preprocessing (2s cut from end)
category: Recording category (clean, duplicate, over_threshold)
reason: Additional category information

Data Collection

The data was collected using a web-based recording platform adapted from Google's Project Euphonia. Participants recorded themselves reading prompts displayed on the screen.

Preprocessing

Each recording has 2 seconds trimmed from the end to remove silence
Duplicate recordings (same speaker, same task) were identified and marked
Recordings over 10 seconds were flagged for review

Data Splits

This is a test upload with 10 samples per speaker. Only clean recordings (no duplicates or over-threshold recordings) are included.

Ethical Considerations

All participants provided informed consent
Data is anonymized (speaker IDs do not contain personally identifiable information)
The dataset complies with GDPR regulations
This dataset should be used to improve accessibility technology for people with speech disorders
Currently, we do not provide the speakers metadata to prevent re-identification

Citation

If you use this dataset, please cite:

[Citation information to be added]

Contact

For questions or access requests, please contact gr.clic@ub.edu

License

This dataset is released under the Creative Commons Attribution 4.0 International License (CC-BY-NC-SA-4.0).

Downloads last month: 20

Total file size:

139 MB