Automatic speech recognition (ASR) system for Aragonese:

Automatic transcription and subtitling

Overview

We have developed the first ASR system for the Aragonese language. We moved from the previous modular system to an End-to-End (E2E) system and experimented with monolingual and bilingual models.

Challenge

The main challenge was to develop an effective system using limited resources. We also had to create a set of linguistic resources, such as corpora of transcribed audio and text standardisation and phonetic transcription systems.

Collaboration

The project was developed in collaboration with the Government of Aragon, within the framework of the European project Linguatec-IA (Interreg POCTEFA).

Result

The new E2E system has significantly improved speech recognition results compared to the previous modular system. The first speech recognition system has been developed for Aragonese, contributing to improving documentation, accessibility and online presence.