Conv1DResNet + Diff-based vibrato (flagged test version)
Pre-release
Pre-release
This is a test release of the multisinger model. The only difference is that the hidden_dim value is increased to 512 to test the automatic pitch of the model. This model is only trained for 60 epochs instead of the normal 90 I do for the other multisinger models.