New TTS framework synthesizes Tibetan dialects from limited data

Researchers have developed FMSD-TTS, a novel few-shot text-to-speech system designed to generate speech for the low-resource Tibetan language across its three main dialects: Ü-Tsang, Amdo, and Kham. The system utilizes a speaker-dialect fusion module and a Dialect-Specialized Dynamic Routing Network to accurately capture variations while maintaining speaker identity. Evaluations show FMSD-TTS outperforms existing methods in dialectal expressiveness and speaker similarity, with the synthesized speech validated on a speech-to-speech dialect conversion task. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Enables creation of synthetic speech for low-resource languages, potentially aiding in dialect preservation and accessibility.

RANK_REASON This is a research paper describing a new text-to-speech system for a low-resource language.

Read on arXiv cs.CL →

• FMSD-TTS
• Tibetan
• Ü-Tsang
• Amdo
• Kham
• Yutong Liu
• arXiv

• paper
• other