andabi / parallel-wavenet-vocoder Goto Github PK

Hi,

excellent work in this repo. I have one question, can you report how much time does it take to generate the sounds? Is it real-time or at least near to real-time.

Originally posted by @HallidayReadyOne in #1 (comment)

Is pretrained model for TTS availible?

Hi, could you add a guide to the ReadMe file on how to run your code? Thank you very much!

Hi,
Firstly, thank you for this great work.
I'm running a training on arctic database and after a couple days of training I would like to get some synthesis sample.
I tried to run generate.py but could not get any synthesis records. I got the following console output:

parallel-wavenet-vocoder-master$ python generate.py
...utils.py:165: RuntimeWarning: Couldn't find ffmpeg or avconv - defaulting to ffmpeg, but may not work
warn("Couldn't find ffmpeg or avconv - defaulting to ffmpeg, but may not work", RuntimeWarning)
dataset size is 114
WARNING:tensorflow:From parallel-wavenet-vocoder-master/models.py:32: init (from tensorflow.contrib.distributions.python.ops.logistic) is deprecated and will be removed after 2018-10-01.
Instructions for updating:
The TensorFlow Distributions library has moved to TensorFlow Probability (https://github.com/tensorflow/probability). You should update all references to use tfp.distributions instead of tf.contrib.distributions.
2019-01-24 13:39:12.324703: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2019-01-24 13:39:12.668828: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1432] Found device 0 with properties:
name: Tesla P100-PCIE-16GB major: 6 minor: 0 memoryClockRate(GHz): 1.3285
pciBusID: 0000:87:00.0
totalMemory: 15.90GiB freeMemory: 15.61GiB
2019-01-24 13:39:12.668918: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1511] Adding visible gpu devices: 0
2019-01-24 13:39:13.133942: I tensorflow/core/common_runtime/gpu/gpu_device.cc:982] Device interconnect StreamExecutor with strength 1 edge matrix:
2019-01-24 13:39:13.134166: I tensorflow/core/common_runtime/gpu/gpu_device.cc:988] 0
2019-01-24 13:39:13.134271: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1001] 0: N
2019-01-24 13:39:13.134819: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 15129 MB memory) -> physical GPU (device: 0, name: Tesla P100-PCIE-16GB, pci bus id: 0000:87:00.0, compute capability: 6.0)
Successfully loaded checkpoint parallel-wavenet-vocoder-master/logdir/default/model-83800
Done.

I wonder if the generate.py script is the correct way for synthesizing or there is some other tool for test synthesis from custom text?
Best Regards.

hi @andabi
Is there any samples generated by Parallel WaveNet ？

In generate.py, it seems that for every input, the generate() function will create a new session and do session.run():

I'm just wondering, will this approach incur session creation overhead every time? Why not create a session once and re-use the session as most of the other models do?

hi, in my training process, the loss do not decrease. How about your super parms setting?

Brilliant jobs! Have you considered integration with tacotron porting? Like https://github.com/r9y9/Tacotron-2