Reproduce “Tibetan-TTS:Low-Resource Tibetan Speech Synthesis with Large Model Adaptation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jiaxu He, Chao Wang, Jie Lian, Yuqing Cai, Yongxiang Li, Renzeg Duojie, Jie Li

This paper adapts a pretrained speech generator for Tibetan text-to-speech amid scarce resources. The authors clean and align speech-text data, use syllable-based or Tibetan-trained BPE units, and fine-tune the model. In preliminary Ü-Tsang evaluation, BPE scores 4.35 MOS versus X-API’s 3.74, while syllable modeling leads in pronunciation accuracy at 97.6%; this does not establish multi-dialect coverage.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.