search

JeremyCCHsu / vqvae-speech

Tensorflow implementation of the speech model described in Neural Discrete Representation Learning (a.k.a. VQ-VAE)
MIT License
128 stars 31 forks source link

tf.contrib.ffmpeg is already deleted from TensorFlow #4

Open Ice9Coffee opened 5 years ago

Ice9Coffee commented 5 years ago

I met a lot of problems with tf.contrib.ffmpeg !!! wav2tfr.py cannot work on my server. It eats all my GPU memory and outputs NOTHING! Then I tried to run it on my Windows 10 PC, and another fxcking problem came!

No module named 'tensorflow.contrib.ffmpeg.ops'

Holy shxt! I thought I was using the same TF until I saw THIS ISSUE.

FFmpeg is already deleted in TensorFlow! tf.train.string_input_producer is also DEPRECATED. It will be removed in a future version of TensorFlow.

This fxcking problem wasted me almost 3 days, and I just start to train the model after I rewrite wav2tfr.py! Please consider about updating the code.

No matter how, still thank you for your code.

JeremyCCHsu commented 5 years ago

Hi @Ice-Cirno,

Thank you for pointing this out and I feel sorry about what you have been through. In fact, I specified tensorflow-gpu==1.5.0 in requirements.txt (and stated in README.md) because I knew that Tensorflow was fast evolving and was likely that it stopped supporting some of the functionalities that I used; similar reason for the OS specification.

Per your suggestion, I'm afraid that I won't be l to keep this repo updated due to a tight schedule. Feel free to send a push request of your updated version of wac2tfr. Reviewing it is what I can do at least.

Thank you again for your feedback.