Loading datasets...
Task: ASR
Release Date: 6/25/2026
Format: WAV, TSV
Size: 105.22 MB
Share
RAPNIC (Reconeixement Automatic de la Parla No Intel-ligible en Catala) is the first Catalan speech corpus collected from individuals with speech disorders, primarily cerebral palsy and Down syndrome. It was built to develop and evaluate Automatic Speech Recognition (ASR) systems accessible to Catalan speakers with impaired speech. This package is an EXAMPLE SET ONLY: it contains 1,000 clean recordings (10 per speaker) from 100 speakers, totalling about 1.33 hours of 16 kHz mono WAV audio in a single train split, with an accompanying metadata TSV (transcriptions, durations, task IDs, etc.). To protect speaker privacy, no per-recording demographic metadata is included. The full RAPNIC corpus is NOT yet publicly released and has no planned release date. It currently comprises around 100 participants and over 22,000 recordings, collected under controlled conditions via a read-speech protocol following participatory and ethical protocols. Recordings were collected with a web platform adapted from Project Euphonia, with participants supported by speech therapists and caregivers. (Complete-set demographics are given under Other information below.) If you are interested in the full dataset, please get in touch (see Contact information). This example set was produced by iSocial in collaboration with the CLiC research group at the Universitat de Barcelona (UB).
Licensing
Creative Commons Attribution Non Commercial Share Alike 4.0 International (CC-BY-NC-SA-4.0)
https://spdx.org/licenses/CC-BY-NC-SA-4.0.htmlRestrictions/Special Constraints
Non-commercial use only. Intended for research and educational use aimed at improving accessibility technology and ASR for people with speech disorders.
Forbidden Usage
No commercial use without permission. No attempts to re-identify speakers. No redistribution without attribution and the same CC-BY-NC-SA-4.0 license. No use that violates the license terms or the privacy/dignity of the participants.
Ethical Review
All participants (or their legal guardians) provided informed consent covering use of voice data and processing of personal data. Data is anonymized (speaker IDs contain no personally identifiable information) and no demographic metadata is shipped with this example set to prevent re-identification. The dataset complies with GDPR.
Intended Use
Research and development of inclusive ASR and accessibility technologies for Catalan speakers with speech disorders; benchmarking and fine-tuning of speech recognition models on impaired speech.
This is an EXAMPLE SET of the RAPNIC corpus, the first impaired-speech corpus for Catalan, covering two of the most frequent developmental speech disorders: Down syndrome and cerebral palsy. The full corpus is not yet published (no planned date); please contact iSocial for access or more information.
Contents of this package: 1,000 clean recordings (10 per speaker) from 100 speakers, ~1.33 hours of 16 kHz mono WAV audio, single train split, plus a metadata TSV. No per-recording demographic metadata is shipped, to prevent re-identification.
Full corpus (unreleased): ~100 participants and over 22,000 recordings collected via a read-speech protocol. Demographics: Down syndrome 51%, cerebral palsy 35%, no response 10%, other 4%; female 54% / male 44%; age mostly 31-45; dialect predominantly Central Catalan (Barcelona-Tarragona), with Girona, Septentrional and Nord-Occidental varieties represented.
Data collection: Recordings were made with a web-based platform, a fork of Google's Project Euphonia Audio Tool. Participants read prompts on screen, supported as needed by speech therapists and caregivers. Each recording has 2 s trimmed from the end; only clean (non-duplicate, within-threshold) recordings are included here.
Example set on Hugging Face: https://huggingface.co/datasets/CLiC-UB/rapnic-example
A formal citation is not yet available. A methodology paper with experimental results on ASR is in preparation.
Please check README.md for more information.
