z.q.mao

Breakpoint problem

22

hi @kan-bayashi , When using this project to train multi-band melgan vocoder, there will always be breakpoints in the generate audio! Trying to modify the kernal size of the first...

reference_output is not used for vae! right?

1

https://github.com/yanggeng1995/vae_tacotron/blob/11a062ffc5534c44e6963b2ab62fa5b503b2835b/models/modules.py#L30 ReferenceEncoder is only use one rnn layers, is different form paper https://arxiv.org/pdf/1812.04342.pdf, which say: "We use the same architecture and hyperparameters for reference encoder as Wang et al. [4]...

C++ infer

23

hi @geneing ! I used the C++ code for infer! but the speed is so slowly mels shape: (80, 500) take times:53.759968996047974 Seven seconds of audio takes about 53 seconds！...

hop_length questions

4

hi, Thank you for your hard work, why set hop_length to 128 in you config file, How does this parameter affect audio quality?

can refer parallel branch

5

The following branch implements a part of the parrel that can be referred to as a reference https://github.com/vincentherrmann/pytorch-wavenet/blob/parallel/wavenet_training.py