Spanish atypical speech corpus
Recordings of people with dysarthria and aphasia, transcribed and annotated, for training and evaluating recognisers.
In a nutshell
Voices that are almost never in the data machines are trained on, recorded with consent and ready for research.
The corpus gathers recordings of people with dysarthria and aphasia reading sentences and in free conversation. Each recording is transcribed and annotated by speech therapists. Sample text.
What it contains
- of audio
- 42 h
- speakers
- 120
- WAV, mono
- 16 kHz
- with phonetic annotation
- ES
Terms of use
CC BY-NC 4.0 licence: it may be used, shared and adapted for non-commercial purposes, citing the source. For commercial use, write to us. The recordings are anonymised and their use requires accepting a data agreement.
How to cite
Apellido, N., Apellido, N. y Apellido, N. (2025). Corpus de habla atípica en español (v2.0) [Dataset]. DELTA, Universidad de Granada. https://doi.org/10.5281/zenodo.0000000
@misc{apellido2025spanish, author = {Nombre Apellido Apellido and Nombre Apellido}, title = {Spanish atypical speech corpus}, year = {2025}, version = {v2.0}, publisher = {DELTA, Universidad de Granada}, doi = {10.5281/zenodo.0000000}, url = {https://doi.org/10.5281/zenodo.0000000} }
Nombre Apellido Apellido and Nombre Apellido. 2025. “Spanish atypical speech corpus.” Version v2.0. Dataset. DELTA, Universidad de Granada. https://doi.org/10.5281/zenodo.0000000.
Related
Code
Recogniser adaptation toolkit
Scripts to adapt a speech recogniser to a specific speaker with few samples.
MIT · v1.4 · Related publication: Título completo de la publicación en una o dos líneas (2024)
Model
Recognition model for atypical speech
Model trained on this corpus. Ready to use with the toolkit.
Apache 2.0 · v1.0