Conv1DResNet + Diff-based vibrato (flagged test version)

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.

Conv1DResNet + Diff-based vibrato (flagged version)

Conv1DResNet_Diff_vib_Flagged.zip contains the MDNv2 model type for timelag and duration, while Conv1DResNet_Diff_vib_Flagged_Pretrain.zip contains the FFN and LSTMRNN model type for timelag and duration

Conv1DResNet (flagged version)

Initial release.