完全按照流程进行训练，并且使用了官方数据集，loss值2.x，请问哪里出现问题了 https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/tts_finetune/tts3 - Githubissues

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

10.48k stars 1.81k forks source link

完全按照流程进行训练，并且使用了官方数据集，loss值2.x，请问哪里出现问题了 https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/tts_finetune/tts3 #3760

Open 1477823267 opened 1 month ago

1477823267 commented 1 month ago

https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/tts_finetune/tts3

1477823267 commented 1 month ago

23a12d4b10f17a5ff5c8780f7662597

49e6f2e45609370dcc37e070707b0a2

1477823267 commented 1 month ago

训练流程如下图：

331e9e9f0ca441012ca69ee75176ce8

xu训练完成后，进行推理预测，发现声音模糊不清

1477823267 commented 1 month ago

@ryanrussell @jzhang533 @harisankarh @ZeyuChen

Ray961123 commented 1 month ago

开发者你好，感谢关注 PaddleSpeech 开源项目，抱歉给你带来了不好的开发体验，目前开源项目维护人力有限，你可以尝试通过修改 PaddleSpeech 源码的方式自己解决，或请求开源社区其他开发者的协助。飞桨开源社区交流频道：飞桨AI Studio星河社区-人工智能学习与实训社区