End-to-end deep learning systems for automatic speech translation

Objectives

MINTZAI proposes a research project aimed at addressing the challenges described above and contributing to consolidating the Basque Country's position as a benchmark in language technologies. The project's main objective is the research, development, and validation of machine speech translation systems based on end-to-end (E2E) deep learning methods.

The project's main objective is broken down into the following scientific and technological goals:

- Research and development of end-to-end (E2E) speech-to-speech and speech-to-text machine translation methods and systems.

Chained speech-to-speech and speech-to-text machine translation.

- Collection, generation, and preparation of data for Basque-Spanish speech translation.

- Automatic and human evaluation of developed systems.

The following scope and impact objectives have also been defined:

1. Definition of a pilot case of speech-to-speech systems for the Basque-Spanish pair.

2. Definition of a pilot case of speech-to-text systems for the Basque-Spanish pair.

3. Validation of the prototypes in relation to the state of the art regarding the quality of final E2E speech machine translation systems.

4. Academic dissemination of the project results at international conferences.

5. Transfer of results to industry and analysis of the possibilities for commercial exploitation.

To achieve these objectives, MINTZAI has a strong and highly experienced consortium in the field of language technologies, which aims to establish itself as a leading authority in the area. This is a strategic project for all participating entities, including MondragonLingua, EITB, and Argia, which have expressed their support through letters of endorsement.