lixuyuan102

CosineAnnealingLR

5

The params of CosineAnnealingLR scheduler in valle_trainer.py seem different with pytorch Docs. code: ``` scheduler = torch.optim.lr_scheduler.CosineAnnealingLR( self.cfg.train.warmup_steps, self.optimizer, eta_min=self.cfg.train.base_lr, ) ``` pytorch 2.0 Docs: `torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max, eta_min=0, last_epoch=- 1,...

soundfile.info does not work on mp3

2

soundfile.read can read ".mp3". However sf.info get : Error opening '.....mp3' File contains data in an unknown format.

Mel model

10

May I ask if this implementation of the model has been experimented on the MEL spectrum.? I used Transformer model with only convolutional positional coding added at the beginning to...

How many steps would be enough if i train this model from start?

8

Hi! Nice work! Could you share how many steps would be sufficient to train a new model? I'm trying to train a 16k FAcodec. The results reconstructed by ckpt 130,000...

Drop the unmasked tokens

1

Nice work! May i ask the 0.9 probability of dropping unmasked tokens to condition on audio only is important? Could you share the detail of AB study?