Tibetan-TTS:Low-Resource Tibetan Speech Synthesis with Large Model Adaptation

2026-05-04

AI Summary

This work addresses the challenges of Tibetan text-to-speech (TTS) synthesis, which include scarce resources, significant dialectal variation, and complex grapheme-to-phoneme mappings. The study presents the first industrial-scale Tibetan TTS system built upon a large speech synthesis model, achieving high-quality, unified multi-dialect synthesis under low-resource conditions. Key innovations include enhanced data quality, Tibetan-specific text normalization, joint optimization of syllable-level and byte-pair encoding (BPE) tokenizers, and a cross-lingual adaptive training strategy. In subjective evaluations, the system achieves mean opinion scores (MOS) of 4.28–4.35 and pronunciation accuracy rates of 96.6%–97.6%, substantially outperforming existing commercial Tibetan TTS APIs.

Abstract

Tibetan text-to-speech (TTS) has long been challenged by scarce speech resources, significant dialectal variation, and the complex mapping between written text and spoken pronunciation. To address these issues, this work presents, to the best of our knowledge, the first large-model-based Tibetan TTS system in the industry, built upon a large speech synthesis model developed by Xingchen AGI Lab. The proposed system integrates data quality enhancement, Tibetan-oriented text representation and tokenizer adaptation, and cross-lingual adaptive training for low-resource Tibetan speech synthesis. Experimental results show that the system can generate stable, natural, and intelligible Tibetan speech under low-resource conditions. In subjective evaluation, the MOS scores of the syllable-level and BPE-based systems reach 4.28 and 4.35, while their pronunciation accuracies reach 97.6% and 96.6%, respectively, outperforming an external commercial Tibetan TTS interface. These results demonstrate that combining a large-model backbone with Tibetan-oriented text representation adaptation and cross-lingual adaptive training enables highly usable low-resource Tibetan speech synthesis, and also provides a technical foundation for future unified multi-dialect Tibetan speech synthesis.

Problem

• Tibetan TTS
• low-resource speech synthesis
• dialectal variation
• text-pronunciation mapping

Innovation

• large-model adaptation
• low-resource TTS
• Tibetan text representation
• cross-lingual adaptive training
• tokenizer adaptation