Hello! Thank you for your great work. But I found that the iSTFT_module_1_8 is a wrong function, which does not work well when the frame shift is 8ms and the frame len is 32ms (which means 3/4 overlap). I carefully checked the value of the result of the process:
print(torch.abs(test_sample-new_sample).mean())
After I do these, I found that the mean value of abs is very big, which means the calcucation of this function is wrong.
Did I miss something? Could you give me some advise? My wechat is yezhangyinge, could you add me if you are convenient?
Thank you very much!
Hello! Thank you for your great work. But I found that the iSTFT_module_1_8 is a wrong function, which does not work well when the frame shift is 8ms and the frame len is 32ms (which means 3/4 overlap). I carefully checked the value of the result of the process:
test_sample= torch.randn((512*100), dtype=torch.float32)*100iSTFT_module_1_8to get new_sampleprint(torch.abs(test_sample-new_sample).mean())After I do these, I found that the mean value of abs is very big, which means the calcucation of this function is wrong. Did I miss something? Could you give me some advise? My wechat is yezhangyinge, could you add me if you are convenient? Thank you very much!