Tacotron synthesis returned void mels and zero seconds wav on custom sentences
Stanley80 opened this issue · 1 comments
Dear Rayhane-mamah,
thank you very much for your amazing work.
I trained Tacotron after preprocessing of 500 audio files on custom language.
python train.py --model='Tacotron'
Than I tried to synthesize on new sentences and synthesis ended
python synthesize.py --model='Tacotron' --text_list=sentences.txt
However, looking at the mels I see this
and wav is zero seconds.
Can you help me understand what happens here?
These are the latter messages in the output
/Users/XXX/Desktop/workspace/venvSynt/Tacotron-2/tacotron/utils/plot.py:27: UserWarning: Attempting to set identical left == right == -0.5 results in singular transformations; automatically expanding.
interpolation='none')
/Users/XXX/Desktop/workspace/venvSynt/Tacotron-2/tacotron/utils/plot.py:72: UserWarning: Attempting to set identical left == right == -0.5 results in singular transformations; automatically expanding.
im = ax2.imshow(np.rot90(pred_spectrogram), interpolation='none')
/Users/XXX/Desktop/workspace/venvSynt/Tacotron-2/tacotron/utils/plot.py:70: UserWarning: Attempting to set identical left == right == -0.5 results in singular transformations; automatically expanding.
im = ax2.imshow(np.rot90(pred_spectrogram), aspect='auto', interpolation='none')
100%|███████████████████████████████████████████| 13/13 [00:23<00:00, 1.20s/it]
synthesized mel spectrograms at tacotron_output/eval
I have fixed unistalling Tensorflow 1.15 adn installing Tensorflow 1.13.1.