lingxue

The fifth alchemy failed as expected... The advantage compared to before is that it no longer makes noise when there is no sound, but singing comes with noise and distortion. The dataset has already removed all the accompaniments with noise that were not cleaned properly. I don't understand where the problem is. I have avoided all the pitfalls and switched back to v1, using the same method as the third training but with a simplified dataset, yet the results have unexpectedly worsened. Repeatedly modifying parameters and debugging only leads to worse results than the third training... I can't understand it.