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.【安全加固,暂停交互,请耐心等待】
https://paddlespeech.readthedocs.io
Apache License 2.0
10.88k stars 1.83k forks source link

ECAPA-TDNN voice clone 问题 #3094

Open panxin801 opened 1 year ago

panxin801 commented 1 year ago

您好感谢你们公开的工作,不过有个问题想请教下。我在尝试examples/aishell3/vc2 TCAPA-TDNN的voice clone时,我直接跑了run.sh中voice_clone这部。我使用的测试src是我自己的原声(男声),但是tts出来的结果竟然是女声。我使用的都是预训练模型。请问这是不是spk_embedding有问题也就是tcapa-tdnn提取的有问题。

ben-8878 commented 1 year ago

您好感谢你们公开的工作,不过有个问题想请教下。我在尝试examples/aishell3/vc2 TCAPA-TDNN的voice clone时,我直接跑了run.sh中voice_clone这部。我使用的测试src是我自己的原声(男声),但是tts出来的结果竟然是女声。我使用的都是预训练模型。请问这是不是spk_embedding有问题也就是tcapa-tdnn提取的有问题。

@panxin801 遇到同样的问题,请问你解决了吗 @yt605155624

gooloosk commented 3 months ago

如果用aishell-3里面的男声作为参考音频效果是可以的