janvainer/speedyspeech

RuntimeError: stack expects a non-empty TensorList

Closed this issue · 5 comments

Moon-sung-woo commented

Hi! First of all, thank you for uploading the source code

When i try Train the duration extraction model, i had error at this code

            spec = self.collate.norm.inverse(out[-1:]) # TODO: this fails if we do not standardize!
            sound, length = self.collate.stft.spec2wav(spec.transpose(1, 2), slen[-1:])

    torch.nn.utils.clip_grad_norm_(X, 1)

image

Please give me solution

janvainer2 commented

Did you train with default settings or did you introduce some changes?

Moon-sung-woo commented

Did you train with default settings or did you introduce some changes?

I trained defalut setting
Just changed git repo because of error
Thank you for answer

janvainer2 commented

What do you mean by changed repo?

Moon-sung-woo commented

What do you mean by changed repo?

image
Like this

Moon-sung-woo commented

Did you train with default settings or did you introduce some changes?

I tried to original code, but i had same error TT