BuaaAlban

Results 10 comments of BuaaAlban

How can I synthesize my own text to speech?

> @Wenqikry Have you tried https://github.com/Rayhane-mamah/Tacotron-2 or https://github.com/NVIDIA/tacotron2 to train log-mels? Combine with the Melgan? > Do you have any experiences? > I used https://github.com/Rayhane-mamah/Tacotron-2 , change feat to log-mel...

Getting ValueError: Attempt to convert a value (PerReplica ..) when starting training

same error

Getting ValueError: Attempt to convert a value (PerReplica ..) when starting training

> @BuaaAlban @prajwaljpj Were you able to run it on multiple GPUs yet? No, and it does't converge

TypeError: iteration over a 0-d tensor

have you solved the problem?

"File contains data in an unknown format." when loading ".wav" file with soundfile.read()

same issue

Model 2012 reproduce

Ps， the features in 2012 is not the same as 2013CTC

How to train using main.py using multiple GPUs?

Another question, it seems no speed up using GPU compared with CPU, have you met the same problem? Both take 280-290 s each epoch

backbone weight names?

Yes, pretrained model on imagenet. I want to replace vgg by resnet34 as backbone. Do you pretrain the vgg on imagenet by yourself?

backbone weight names?

Yes, but it's in pytorch version, though I can do the mapping of names

how to use peoples speech dataset?

Hi, I found there were 260988flac files in the released tar dataset（part-00000-07a8f0d3-6d27-4299-887a-dc12a6d72f8d-c000.tar）which only has about 1000 hours, but the metainfo of the json file （part-00000-4e132642-c01c-4db6-9db0-a1e19193f6f8-c000.json）has 4321002 files in it, which...