james34602
about the win function applied in stft
Hello, again. I write down analysisWnd and synthesisWnd and drawed their curves. I cannot put a picture here. but it seems that these two win function looks strange. AnalysisWnd is not symmetric and synthesisWnd has two peaks.
I read the original spleeter code. It seems it uses hanning win as its stft win function.
Hello, It doesn't matter much for using different window function to operate, empirically, CNN look for patterns not much on actual value, so that no observable real-world accuracy reduced. Neither the spectral leakage nor the time-frequency resolution break the structure of the spectrogram, you are free to use it. The popular window function breaking this is rectangular window, which is the window function create impulse-liked image on spectrogram.
The window function uses in the project allows lower latency(Asymmetric) and follow as much as Hann window behave, also perhaps most importantly, it doesn't prevent perfect reconstruction, which mean we can do computation tricks to run only 1 stem neural network to separate 2 stems signal.
