Despite instructions to install ffmpeg in README.md (#414), people still skip it and get audioread's NoBackendError whenever trying to load mp3s. The exception is confusing to the end user and we get a lot of questions about it.

• The error message looks like this: CorentinJ/Real-Time-Voice-Cloning#386 (comment)
• What people need to do to resolve it is this: ht

Hello, if i enable bidirectional decoder training script drops eror.
There is in train.py line 185
keep_avg.update_values({'avg_decoder_b_loss': loss_dict['decoder_backward_loss'].item(),
need to change on
keep_avg.update_values({'avg_decoder_b_loss': loss_dict['decoder_b_loss'].item(),

I'd like to train this model on 8 V100 GPUs - does it support multi GPU training?

We have to create contributing files for two repositories:

1. cboard-api
2. ccboard

Also, please update cboard's contributing file, in order to:

• Add a list of ways to Contribute .
• Include and describe testing as a way to contribute.
• Include discord channel .
• Include Coding conventions.