Top-notch Automatic speech Recognition System (TARS)
Follow the instructions as the original wavenet implementation.
However, instead of installing
scikits.audiolab, use soundfile instead, which supports Python 3.x
Also, when compiling Libsndfile from source / installing from brew, make sure to enable Flac support
MacOS: (Courtesy of this person)
brew install libsndfile --with-lame --with-flac --with-libvorbis
brew link --overwrite libsndfile
Also, installed ffmpeg
preprocess.py - written by someone else,used to create lmfcc from the sound files and store those the labels in asset/data/preprocess folder Conf.py - Parameters and alphabet mapping
train.py - Main file for training, run this to get the log files which you can plot in tensorboard model.py - Creates the graph for the model dataloder.py - The dataloading pipeline, loads data efficiently. Does the batching too! test_ctc.py - Some experiments, can be ignored