Automatic Catalan KWS Database for Projecte AINA

Automatically extracted Catalan word database using alignment techniques (Montreal Forced Alignment, MFA) from speech databases with transcriptions. Precisely: Mozilla Common Voice, ParlamentParla, and OpenSLR-69. Usable for training keyword spotting models for home automation.
MFA leverages algorithms to accurately synchronize speech signals with the corresponding text at the phoneme level.

Two versions of the database have been created:

general: This version encompasses all data, providing a comprehensive dataset for various analyses and applications.

split: This version is divided into train, dev, and test to ease the task of training a keyword spotting model. Speaker-wise, It is divided by 80%, 10%, and 10%.