Automatic speech recognition (ASR) system for Occitant:

Automatic transcription and subtitling

Overview

Two automatic transcription systems were developed for the Gascon and Languedocian dialects of Occitan for the association Lo Congrès Permanent de la lenga occitana (The Permanent Congress of the Occitan language). The modular system has been replaced with the End-to-End (E2E) system.

Challenge

There were two challenges: to improve the ASR systems previously developed for each dialect and to update the linguistic resources. The corpora of transcribed audio and the text standardisation and phonetic transcription system were updated.

Collaboration

The project was developed in collaboration with the Lo Congrès permanent de la lenga occitana, within the framework of the European project Linguatec-IA (Interreg POCTEFA).

Result

The E2E model was adapted for both dialects using updated Occitant corpora and, based on this adaptation, two more robust systems than the previous modular system were developed for both dialects (Gascon and Languedocian). This contributed to improving the documentation, reach and online presence of Occitant.