Tacotron2 TTS demo: Adapt to the latest tensorflow

[r9y9]

It seems there's no clean way to downgrade tensorflow on google colab, so we might need to update https://github.com/r9y9/Tacotron-2 to make it run on the latest tensorflow.

[sandeepnedeng]

Up vote!

[m-toman]

Actually for me adding
! pip install "tensorflow<=1.9.0"
worked.

But I agree about the "clean" thing ;)

[yhgon]

If you want to use GPU in mel generation in COLAB, eliminate current tensorflow 1.12r0x and re-install tensorflow-gpu==1.9.0 However, it requires to install exactly matching CUDA libraries. so below script you could some CUDA libraries which COLAB didn't included such as cublas-9.0

before git clone the files run below script. it take 10 minutes to prepare jupyter each time. It's temporal solution before current tf tacotron2 implementation support tensorflow 1.12.x,

%%bash
nvidia-smi
pip3 uninstall -y tensorflow tensorflow-gpu pytorch torch
rm -rf cuda-repo*
wget https://developer.nvidia.com/compute/cuda/9.0/Prod/local_installers/cuda-repo-ubuntu1704-9-0-local_9.0.176-1_amd64-deb
apt-get install dirmngr
dpkg -i cuda-repo-ubuntu1704-9-0-local_9.0.176-1_amd64-deb
apt-key add /var/cuda-repo-9-0-local/7fa2af80.pub
apt-get update
apt-get install  -y --no-install-recommends  \
 cuda-core-9-0 \
 cuda-cublas-9-0 cuda-cublas-dev-9-0 cuda-cudart-9-0 cuda-cudart-dev-9-0 \
 cuda-cufft-9-0 cuda-cufft-dev-9-0 cuda-curand-9-0 cuda-curand-dev-9-0 \
 cuda-cusolver-9-0 cuda-cusolver-dev-9-0 cuda-cusparse-9-0 \
 cuda-cusparse-dev-9-0 \
 cuda-libraries-9-0 cuda-libraries-dev-9-0 \
 cuda-misc-headers-9-0 cuda-npp-9-0 cuda-npp-dev-9-0 \
 cuda-nvgraph-9-0 cuda-nvgraph-dev-9-0 cuda-nvml-dev-9-0 cuda-nvrtc-9-0 \
 cuda-nvrtc-dev-9-0
rm -rf cuda-repo*
rm -rf wget-log*
pip3 install -q tensorflow-gpu==1.9.0  
pip3 install -q torch

it takes about 17 min x 2 to generate 'good job wavenet'
0%| | 3/26624 [00:00<15:42, 28.26it/s]
0 good job wavenet
100%|██████████| 26624/26624 [16:01<00:00, 27.70it/s]

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[japita-se]

somebody get working the google colab notebook?

[r9y9]

Didn't it work?

[r9y9]

btw I'm unlikely to have time to do this for a while... pretty busy recently.

[MohammadJavadD]

! pip install tensorflow==1.6
! pip install keras==2.2.0

@japita-se I've tried this and it works.

[MohammadJavadD]

Hi @r9y9,
I almost have done this. How do you want to update that? I mean a new branch or what?

[r9y9]

@MohammadJavadD HI, thank you for working on this. If the changes are only for the notebook, I would appreciate it if you make a PR to https://github.com/r9y9/Colaboratory.

Did you make changes to Tacotron2 code as well?

[MohammadJavadD]

@MohammadJavadD HI, thank you for working on this. If the changes are only for the notebook, I would appreciate it if you make a PR to https://github.com/r9y9/Colaboratory.

Did you make changes to Tacotron2 code as well?

@r9y9
The problem is tf.compat which have used on both Tacotron2 and WaveNet_vocoder so I followed what was done on WaveNet_vocoder newest version (Copying module) so all three repos need to change a bit. Also, the params.py file has changed in the different versions of WaveNet_vocoder (builder parameter).

[do-web]

The current codelab demo is not working, any updates here?

[qwertyuu]

https://colab.research.google.com/drive/1EPoNkTHIyYeLq_fZZY5HS1ecjCUQEGvg?usp=sharing

That one works on my end, but it uses TF 1.x instead of TF 2.x. I just removed/adjusted the code where it needed to.