search

Rayhane-mamah / Tacotron-2

DeepMind's Tacotron-2 Tensorflow implementation
MIT License
2.28k stars 905 forks source link

Noisy Unclear wavs generateed with LJspeech dataset #406

Open shubhishukla10 opened 5 years ago

shubhishukla10 commented 5 years ago

I am getting a really noisy and unclear wavenet_output for the LJspeech dataset. I only changed one parameter in hparams.py . wavenet_synthesis_batch_size = 2*1 (because of OOM) Tacotron training stepps : 50k Wavenet training steps :50k RAM:128GB GPU:12GB

Steps followed : python preprocess.py python train.py --model='Tacotron' --tacotron_train_steps=50000 python train.py --model='WaveNet' --wavenet_train_steps = 50000 python synthesize.py --model='Tacotron' --mode='synthesis' --GTA=True python synthesize.py --model='WaveNet'

Here are my wavenet output files: wavenet_output.zip I'm new to this and I can't understand what exactly is going wrong. Please help !

N0vel commented 4 years ago

Same problem

chalathecfc commented 3 years ago

same problem have you fixed this problem @shubhishukla10 ? when i try to synthesize Tacotron-2 alone it works well but when i run synthesis with wavenet only outputs in Tacotron_output directory contain expected output but wavenet_output directory contains incorrect output.