On Thu, Jun 12, 2014, at 09:41 AM, Fons Adriaensen wrote: > There's a paper by Dolson and Laroche (1999) which is a 'must read' > for anyone dabbling with phase vocoders. One warning, the method described in it is patented -- I had to do a hasty rewrite in some early Rubber Band code. I don't know whether anyone enforces the patent though. > For code, maybe have a look at Rubberband which may contain > interesting things (I don't know, never dared to look Probably wise -- I expect it would horrify you. And of course this application is an incredible time-sink simply because there's no right way to do it. It's a subject that can surely drive you mad. You gave a low-level example of the problem earlier (with neighbouring frequency bins). Looking at it at a high level, you're basically trying to synthesise a signal that corresponds to "what the same instruments would have sounded like if they were playing slower" (or higher, or whatever). You don't have anything like enough knowledge frame-by-frame to actually do that. You can get closer for many signals with a sinusoidal modelling decomposition (in which you track the frequencies that appear to be consistent frame-to-frame, adjust their phases, and treat the rest as noise whose phase you don't change) but those methods are still fairly expensive to do and of course even the best method can never actually be technically correct. Rubber Band does basically the same sums as you just gave. By default it fudges the time/frequency question for neighbouring bins by taking groups of bins that appear to be moving in the same direction (in frequency) and giving each one a phase advance somewhere between its single-bin predicted value and the value that would be expected if the group were all following the same path. That's mathematically... barely supportable at all, but in many cases it sounds OK. The other thing it does by default is reset the stretch factor and revert to the input phases when a sufficiently noisy transient is found, which is why you get quite poppy transients in e.g. drum loops that are either satisfying or unrealistic depending on your point of view. Chris _______________________________________________ Linux-audio-user mailing list Linux-audio-user@xxxxxxxxxxxxxxxxxxxx http://lists.linuxaudio.org/listinfo/linux-audio-user
