A Pytorch implementation of WaveNet ASR (Automatic Speech Recognition)
Reference:
https://github.com/kingstarcraft/speech-to-text-wavenet2
https://github.com/buriburisuri/speech-to-text-wavenet
conda install pytorch=1.4.0
python train.py --exp example
-
ctcloss in pytorch and tensorflow have the different default blanks. " " is not the blank, however, "" is the blank.
-
Xaiver init is better.