For a school's project, I am programming a standard phase vocoder for time compression/expansion of digital audio. The audiodata is broken up in frames, each frame is multiplied by a window function (in my case a Hann-window) and the result is transformed into frequency-domain by FFT. Now because the frames overlap each other, I was wondering if it's possible to optimize this computation. For example, I am using an FFT/window size of 4096 and a frame hopsize of 64. So strictly speaking there are only 64 "new" samples per new frame. Also I noticed a standard implementation gives incorrect beginning and endings of the processed data, with a fade-in respectively fade-out due to the window function. Thanks in advance
Phase Vocoder
Started by ●April 7, 2009