RVC-Boss

Results 538 comments of RVC-Boss

You should fine tune the model using all languages you need to cross when cross-lingual inference. Freeze earlier layers: It may work, but I haven't do the experiment, and therefore...

因为底模的标点没那么丰富,可能得你用很多带标点的数据,微调很多轮

你用的是哪个api?然后api的话 @ChasonJiang 一起看看?

楼主可以把bad case发上来我们测测吗

可以试试推完长文本后在https://github.com/RVC-Boss/GPT-SoVITS/blob/main/GPT_SoVITS/inference_webui.py#L515后面加上 del pred_semantic torch.cuda.empty_cache()

> 不追求推理速度可以加 `with torch.no_grad():`试试 这个早加过了

> > 可以试试推完长文本后在[https://github.com/RVC-Boss/GPT-SoVITS/blob/main/GPT_SoVITS/inference_webui.py#L515后面加上](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/GPT_SoVITS/inference_webui.py#L515%E5%90%8E%E9%9D%A2%E5%8A%A0%E4%B8%8A) del pred_semantic torch.cuda.empty_cache() > > del释放不掉显存。 > > `Current GPU memory usage-t2s: 10.50 GB > > Max GPU memory usage-t2s: 10.51 G Current GPU memory usage-vits: 10.56...