NewEricWang
NewEricWang
I also find the reverberation of the generated speech become more obvious along with the train iteration increasing.
@switchzts , It is our owner mandarin corpus.
@switchzts , Its average length is about 4s~5s. It is a single speaker database.
@switchzts ， num_layers=10 in the model. The total time is about 12h. The train loss is about 11.0.
No, use_mu_law=false
What is your input? Do you give model conditional input?
You can try to verify it by using some shorter sentences.
@begeekmyfriend , thanks for your help! I will try your proposed method.
> I had the same issue with 2xA6000 setup. Forcing training on a single GPU fixed it. How do you force training on a single GPU when running "python finetune.py...
> os.environ["CUDA_VISIBLE_DEVICES"] = "0" In my case, it don't work. I add the above code into "finetune.py". When running "python finetune.py", all GPU still are used.