Improving sequence-to-sequence Tibetan speech synthesis with prosodic information

There are about 6,000 languages worldwide, most of which low-resource languages. Although the current speech synthesis (or text-to-speech, TTS) for major (e.g., Mandarin, English, French) has achieved good results, voice quality TTS Tibetan) still needs to be further improved. Because prosody plays a significant role in natural speech, article proposes two sequence-to-sequence (seq2seq) Tibetan models with prosodic information fusion improve synthesized speech. We first constructed large-scale corpus seq2seq TTS. Then we designed generator extract from sentences. Finally, trained by fusing information, including feature-level and model-level fusion. The experimental results showed that proposed models, fuse could effectively Furthermore, only 60% ~ 70% training data synthesize similar baseline Therefore, methods can