错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Sentences vs Phrases in Neural Speech Synthesis

  • Daniel Tihelka,
  • Jindřich Matoušek,
  • Zdeněk Hanzlíček,
  • Lukáš Vladař

摘要

The neural network-based TTS models are usually trained and inferred on the whole sentences, or, in general, on longer chunks of speech. However, these may negatively affect the responsiveness of the TTS system in cases when latency should be kept as small as possible. We present experiments using smaller chunk lengths, namely phrases, and their impact on speech quality when various chunk length combinations are used for training and inference in the VITS synthesizer.