search

CODEJIN / GST_Tacotron

Implementation of Global Style Token Tacotron in TensorFlow2
MIT License
25 stars 10 forks source link

Pretrained model #2

Closed ShaharDagan closed 4 years ago

ShaharDagan commented 4 years ago

Hi, Is there any chance for a pretrained model(created from the checkpoint) that outputs the Mel Spectrogram from text(without wav)? Because when I tried to load the checkpoint I get the following output:

================================================================================================== Total params: 506,720 Trainable params: 505,824 Non-trainable params: 896

Checkpoint 'Checkpoint/S_36000.CHECKPOINT.H5' is loaded. Inference running... 2020-04-21 10:41:16.278946: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudnn64_7.dll 2020-04-21 10:41:17.310366: W tensorflow/stream_executor/gpu/redzone_allocator.cc:312] Internal: Invoking GPU asm compilation is supported on Cuda non-Windows platforms only Relying on driver to perform ptx compilation. This message will be only logged once. 2020-04-21 10:41:17.599566: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_10.dll

And then the program just stops without any warning. Hope for your help

CODEJIN commented 4 years ago

Hi, do you want to get mel spectrogram numpy array by inference? I just modified that. Now Inference function returns mels, stops, spectrograms, and alignments numpy arrays. Please check the Readme.md

ShaharDagan commented 4 years ago

Hi Thanks for your quick respond, The problem I facing is that the algorithm doesn't output the WAV file in the folder and stop without any reason and message. I set up the project in Google Colab: https://colab.research.google.com/drive/1hworMHo8dRgh8NbroeZpFXJ9wCU-qoOU Is there any chance you can fix that the model will output the wav file from the text? I am using your project as reference for my school project so I will be really glad for your help ( :

CODEJIN commented 4 years ago

Hi, I checked your Colab and wrote several code lines. However, when I tried to load the checkpoint, I got 'OutOfRangeError: Read less bytes than requested [Op:RestoreV2]' error message. This might mean the checkpoint files are broken. I re-uploaded the checkpoint to the Colab directly and tested, and it worked. Please see the code I wrote in your colab and replace the checkpoint files in your github.

ShaharDagan commented 4 years ago

Thank you so much! really appreciate it! it works!