Aydous

我在测试说话人相似度的时候发现训练集和在LibriTTS的train-clean-100上测得的平均相似度很接近，是因为提供的pt文件是已经在LibriTTS上已经fine-tune好的吗？还是我测试说话人相似度的方法不太合适？我用的是该项目自带的pretrained speaker encoder提的emb vector计算转换后的语音和参考音频之间的余弦相似度。

Could you please upload the checkpoint of Experiment 1, which is trained only with VCTK, again? The original link is no longer valid.

hello, I followed your training code, but it causes the discriminator to easily distinguish between fake and real samples, and the generator ultimately fails to generate realistic samples. Could your...