How many epochs to run on VBD dataset to reproduce the results in the sgmse+ paper?

[Yangliusha]

Thank you for the open source of this awesome work. I'm trying to reproduce the results in the paper. But after 870 epochs of training (far more than the 160 epochs written in the paper), I only got PESQ: 1.95+-0.69. Is there anything I'm missing to run the codes? I run the codes with all the default settings.

[libai2345]

Hi, have you solved this problem now? Hope to hear from you soon.

[Huanianss]

Thank you for the open source of this awesome work. I'm trying to reproduce the results in the paper. But after 870 epochs of training (far more than the 160 epochs written in the paper), I only got PESQ: 1.95+-0.69. Is there anything I'm missing to run the codes? I run the codes with all the default settings.

I trained the model with a batch size of 8 and a gradient accumulate step of 4 to approximate the batch of 32 on the 4 GPUs described in the paper. Other settings are the default and the PESQ: 2.76 ± 0.66. May I ask if have you tested with the pre-trained weight? I tested the pretrained weight on the VB-DMD dataset and the PESQ is 2.91 ± 0.63.