- prepare a balanced dataloader
- prepare a utt2wav and utt2label. [utt] [wav] [start]
- write in multithread style
- run the VAD model
- prepare a customized dataset
- make sure label2int's order is the same as initializing the sampler.
- the object that the sampler is working on is merely a feature, using a window to sample. There must be an encoding&decoding mechanism to sample from the same file.
- remove unnecessary part in
train.pyand config files.