Why you use pseudo_vocals to Calculate the loss of the validation set？

[Niuguangshuo]

No description provided.

[tsurumeso]

Because I don't have vocals dataset.

[Niuguangshuo]

因为我没有声音数据集。

I have a large amount of noise and speech data, so can I not use pseudo anymore? Is it better to calculate directly using real speech as the validation set?

[Niuguangshuo]

You have two datasets, Mix and Inst. Isn't speech calculated as Mix minus Inst?

[tsurumeso]

The reason I use pseudo_vocal is that my own mix source and inst source have different volume, so if I simply calculate the difference, the inst source will remain in the vocal.

[Niuguangshuo]

我使用 pseudo_vocal 的原因是我自己的混音源和 inst 源具有不同的音量，因此如果我简单地计算差异，inst 源将保留在人声中。

I know .Thank you very much，I obtained the mixture from the original noise and speech, so I can calculate the difference. I will modify the code to adapt to my data.