ILG2021

I am trying to port openvpi nsf-hifigan working with Amphion, but I found that nsf-hifigan needs f0 to inference, which gan_vocoder_inference.py not supply.

您好，sovits 4.1中会用到您项目中的diffusion底模，可以补充下协议吗？ https://huggingface.co/datasets/ms903/Diff-SVC-refactor-pre-trained-model

[Feature] How to support new languages

1

你好，我想训练一个法语的tts，不知道是否需要修改代码？如何修改可以支持。另外想咨询下大概需要多少小时的干声可以训练出来一个比较好的tts？这个tts是专有领域的（科技），不需要那么强的泛化。

不知道pytorch-lighting使用tpu会不会降低使用门槛

1

最近huggingface也在开发optiumum-tpu

very slow on cpu

1

I have tried four models on desktop cpu Intel(R) Core(TM) i7-10700 CPU moondream-0_5b-int4.mf.gz moondream-0_5b-int8.mf.gz moondream-2b-int4.mf.gz moondream-2b-int8.mf.gz it will cost much time to inference, at least 7s, int4 is faster than...

This is the weight link: https://github.com/bshall/hubert/releases/tag/v0.1

I wanna to know what the license of these pretrain weights: https://huggingface.co/lj1995/VoiceConversionWebUI/blob/main/hubert_base.pt https://huggingface.co/lj1995/VoiceConversionWebUI/blob/main/rmvpe.pt

Need help for training a 44k model

2

I have trained a 44k model recently, but the inference result is a little strange, the audio seems raised the pitch. I have change the mel parameters in ljspeech.yaml: n_fft:...

[Speaker-Encoder by @mueller91](https://drive.google.com/drive/folders/15oeBYf6Qn1edONkVLXe82MzdIi3O_9m3)