search

barronalex / Tacotron

Implementation of Google's Tacotron in TensorFlow
236 stars 80 forks source link

don't get anything by running test.py, how to generate the wav? #14

Closed azraelkuan closed 7 years ago

azraelkuan commented 7 years ago

hi, i am new to tts. when i run the test.py, i only get a dir, weightes/arctic/tacotron, but in the dir , there is nothing in it. so i want to know how can i generate the wav from the text???

mrelich commented 7 years ago

Hello,

You have to first train a model and then you can test it. At the moment, I don't believe there is a default model given with this package. Right now, it stores the audio file as a tensor, which you can see in the log directory unter log/<datasetname>/tacotron/test/. I'm not super familiar with how to view them, but I've been using tensorboard and pointing the logdir for tensorboard to log directory.

Alternatively, you can edit the code to dump the wav files. Edit this line: https://github.com/barronalex/Tacotron/blob/master/test.py#L60 You add at the end of the method a filename. See the function call here: https://github.com/barronalex/Tacotron/blob/master/audio.py#L56

One thing I am struggling with is understanding why the output during training is somewhat audible, but when I run the same text through the testing script (test.py) it is total garbage... If I can't figure that out soon I will post another issue.

barronalex commented 7 years ago

You can now run 'download_weights.sh' then 'test.py' and generate audio from pre-trained weights without downloading the data.