Phonetic Intermediate Latents for Fast and Accurate Swiss German Text-to-Speech Synthesis

• Unité de recherche: INNOSUISSE
• Numéro de projet: 127.910 INNO-ICT
• Titre du projet anglais: Phonetic Intermediate Latents for Fast and Accurate Swiss German Text-to-Speech Synthesis

Mots-clé (Anglais)

ICT, Deep learning

Résumé des résultats (Abstract) (Anglais)

We propose a Swiss German TTS system that uses phonetic representations to synthesize speech from High German input. It avoids reliance on non-standard dialect writing by directly mapping to dialectal phonemes. Benchmarks will assess speed and quality, supporting Agigo?s AI speech platform.