search

PaddlePaddle / PaddleSpeech

Easy-to-use Speech Toolkit including Self-Supervised Learning model, SOTA/Streaming ASR with punctuation, Streaming TTS with text frontend, Speaker Verification System, End-to-End Speech Translation and Keyword Spotting. Won NAACL2022 Best Demo Award.
https://paddlespeech.readthedocs.io
Apache License 2.0
10.99k stars 1.83k forks source link

Speech Server一键部署语音服务时TTS句首句末发音不全 #3251

Closed menfanjia closed 1 year ago

menfanjia commented 1 year ago

General Question

Speech Server一键部署语音服务，使用am为fastspeech2_aishell3，voc为pwgan_aishell3，生成的短句音频句首第一个字和句末最后一个字有吞音，句末吞音现象尤其严重，不加标点的话几乎听不到最后一个字。

Javacr commented 1 year ago

确实，我在本机跑源码，也遇到这种问题，最后一个字发音时长太短，几乎听不到

Javacr commented 1 year ago

我只能在每个短句的最后加一个“了”字，来解决这个问题

Javacr commented 1 year ago

最后发现，合成语音没有问题。不要用win10默认播放器，用音频播放器就没问题。