I cant even use Wavenet pre-trained

[kosumosuSpb]

I clone repo, then:

git checkout 2092a64
python preprocess.py ljspeech ~/data/LJSpeech-1.1 ./data/ljspeech --preset=20180510_mixture_lj_checkpoint_step000320000_ema.json

But i have:

AttributeError: module 'tensorflow' has no attribute 'contrib'

(of cause i have ~/data/LJSpeech-1.1 and 20180510_mixture_lj_checkpoint_step000320000_ema.json)

I found that its cause i have new version of tensorflow (2.10) and i have to downgrade tensorflow to 1.15 or update the code of this repo with update script from Tensorflow.

But:

• pip there are no versions of tensorflow older that 2.x
• if i try to update code with scrip, got StopIteration and stuck on this step with no update code.
• when i try to build old tensorflow from source (1.15.5 or 1.13), bazel, which i have to use to build tensorflow packet (according to tensorflow docs) cant download dependensies.

what am I doing wrong?

(actually, later i wanted to use wavenet as decoder to codec2)

[kosumosuSpb]

i got it!

I had installed tensor 1.15 over anaconda, added other dependencies mostly with pip, and stacked with this:

  File "/home/kosumosu/projects/WaveNet_vocoder/wavenet_vocoder/wavenet_vocoder/__init__.py", line 4, in <module>
    from .version import __version__
ModuleNotFoundError: No module named 'wavenet_vocoder.version'

it turned out that in commit 2092a64 there is no file version.py in wavenet_vocoder dir. I added it with this:

__version__ = '0.2.0'

And its worked.