Closed kamo-naoyuki closed 1 year ago
I have a question about the process flow for WPE.
I'm not sure that my understandings are correct, but in the this code, the processed signal by WPE is input for gss, but it's not fed into beamformer.
https://github.com/desh2608/gss/blob/0d2cae4c16d4ffa4410a9be1d5134d3f82ac5ca1/gss/core/enhancer.py#L313
https://github.com/desh2608/gss/blob/0d2cae4c16d4ffa4410a9be1d5134d3f82ac5ca1/gss/core/enhancer.py#L316
https://github.com/desh2608/gss/blob/0d2cae4c16d4ffa4410a9be1d5134d3f82ac5ca1/gss/core/enhancer.py#L344-L349
I checked the original code: https://github.com/fgnt/pb_chime5/blob/master/pb_chime5/core.py#L522-L564 and the paper, but in the code, the WPE signal seems also used for beamformer.
I guess this is a bug caused by implementation of chunk process.
Ahh, good catch! I'll create a PR to fix this. I wonder how much impact this would have on the downstream performance.
Fixed in #31.
I have a question about the process flow for WPE.
I'm not sure that my understandings are correct, but in the this code, the processed signal by WPE is input for gss, but it's not fed into beamformer.
https://github.com/desh2608/gss/blob/0d2cae4c16d4ffa4410a9be1d5134d3f82ac5ca1/gss/core/enhancer.py#L313
https://github.com/desh2608/gss/blob/0d2cae4c16d4ffa4410a9be1d5134d3f82ac5ca1/gss/core/enhancer.py#L316
https://github.com/desh2608/gss/blob/0d2cae4c16d4ffa4410a9be1d5134d3f82ac5ca1/gss/core/enhancer.py#L344-L349
I checked the original code: https://github.com/fgnt/pb_chime5/blob/master/pb_chime5/core.py#L522-L564 and the paper, but in the code, the WPE signal seems also used for beamformer.
I guess this is a bug caused by implementation of chunk process.