Why 16000hz sampling rate has been choosed for this research ?

[chazo1994]

I have tried to train a lot of model with sampling rate 22050 but, it can not reproduce quality of 16000 hz model. Can you explain why you use 16000 in your research ?

[chazo1994]

@KevinMIN95 help me please.

[KevinMIN95]

I chose a 16kHz of sampling rate to match with other baselines. I haven't tried with 22050 but I think it may be helpful to increase the capacity of the model.

[chazo1994]

to match with other baselines. I haven't tried with 22050 but I think it may be helpful to increase the capacity of the model.

Thanks, I will try. Maybe I will increase encoder dim from 256 to 384.

[MrGolden1]

to match with other baselines. I haven't tried with 22050 but I think it may be helpful to increase the capacity of the model.

Thanks, I will try. Maybe I will increase encoder dim from 256 to 384.

Did you try this?

[chazo1994]

to match with other baselines. I haven't tried with 22050 but I think it may be helpful to increase the capacity of the model.

Thanks, I will try. Maybe I will increase encoder dim from 256 to 384.

Did you try this?

Yep, I try with 384 and 6 fftblock and it work.

[MrGolden1]

@chazo1994 Great! Was the result better than 16kHz? Could you share the network and pretrained weights if possible?

[chazo1994]

@chazo1994 Great! Was the result better than 16kHz? Could you share the network and pretrained weights if possible? @MrGolden1 The results is slightly better than 16khz. I'm sorry, I cannot share the pretrained weights but the network is same as this repository, only hidden dimesion and number of fftblock are differents.

[chazo1994]

@chazo1994 Great! Was the result better than 16kHz? Could you share the network and pretrained weights if possible?

@MrGolden1 The results is slightly better than 16khz. I'm sorry, I cannot share the pretrained weights but the network is same as this repository, only hidden dimesion and number of fftblock are differents.