Hi, if you have short echoes that annoy you, you should decrease the window length. For your relatively simple spectra and relatively fast dynamics (piano attacks) I would guess that 1024 or perhaps 2048 should be good, and similar for fft length. *Not* as high as possible! Window type doesn't matter much - I usually use Hanning (not Hamming). I don't know what program you are using, but I assume that "windows per second" has to do with overlap. If you do very long stretches, you might like to increase this value, but otherwise it shouldn't matter much. Dr. Oyvind Hammer, senior researcher PGP - Physics of Geological Processes University of Oslo > I've been trying to stretch a soundfile. It is a piano track with mostly two > voices only. I did it with an fft plain phase vocoder. Now there are a lot of > parameters and I got some undesireable delay. In general the stretched sound > was good, but the delay is disturbing. So here's the question, how should I > set the following parameters to get the best result? > fft-length (as high as possible?), windows per second, window length, > windowtype (hamming, triangular, kasier with several coeffecients). > Any ideas or practical experience with that? - I'm NOT doing it in realtime, > so CPU-time is no problem. I've got time! :-)
