Hello,

I'm trying to implement a Phase Vocoder by using Short-Time-Fourier-Transform (STFT). I have some questions about windowing and the overlap-add (OLA) in the Resynthesis - part. I hope someone has experiences with that.

In all that I read about it so far, I just found that if windowing is used, the sum of the windows has to be constant to avoid amplitude modulation in the output sound.

But in a Phase Vocoder there is windowing twice : once before the FFT and once after the IFFT. So the sum of the window^2 has to be constant to avoid AM, is that right ?

In my implementation of the phase vocoder I use a Hanning-Window. The double-windowing leads to an amplitude modulation, when the hopsize is half the frame size (overlap = 50%). If the hopsize is framesize/4 (overlap = 75%), there is no AM and the signal passes fine.
To test it, I calculated the sum of hanning-window^2 with overlap 75% and it's really constant. The sum of hanning-window^2 (overlap 50%) is modulating.

Does anyone know how to avoid amplitude modulation with greater hopsizes or other window types ?


Thanks in advance.

Bernd

What was the outcome of the above? I am also getting amplitude modulation when using 50% overlap (for obvious reasons),but assume there is a work around?

Thanks

I never worked with phase vocoders, but the same issue occurs in MDCT-based audio coders.

So try one of the power complementary windows used in Ogg Vorbis, MP3, or AAC.

Chris

Will check it out, thanks

Yes, power-complementary windows are the way to go.
Another way is to compensate for the resulting amplitude modulation after overlap-add operation.

Consider taking the square root of your window function before applying it to your frames.
Simply: win = sqrt(hanning)
With a constant 50% overlap add you would want more of a sinusoidal envelope, giving you constant power as frames are overlapped.