Forums

Pitch calculation (FFT or autocorrelation)

Started by gjunge June 5, 2006
robert bristow-johnson wrote:
> anyway, any application i can think of for a PDA desires the pitch > returned to be the same as the perceived pitch. if the note sounds > like middle C, the PDA should say it's middle C, not the C an octave > below nor the F that is 7 semitones lower than that.
Robert, I agree.
> the problem > is that with many instruments and the human voice, both spoken and > singing, sometimes the nature of the vibration of whatever it is (reed, > lips, strings, vocal cords, or a column of air) is that one vibration > is nearly identical to the next but slightly different but the > following vibration is very much like the first. every odd cycle is > more identical to each other than they are to the even cycles.
Moreover in some signals the odd and even pulses are different significantly and then one can hear a pitch somewhere between F0 and F1. The good examples are here: http://www.mrc-cbu.cam.ac.uk/cnbh/web2005 /teaching/sounds_movies/Pitchchroma&height.htm My interest is to research why and how does the human's pitch perception mechanism work this way. A problem of waveform repetition period becomes difficult when two subsequent pulses are similar but not identical. The result of two pulses (or parts of signal) comparison depends on the definition (or concept) of similarity. Dmitry Terez has his own concept and his own method. Just one of others. I don't see a breakthrough. Nothing to argue about. The only thing that could be interested for someone (not for me) is how much the concept of Dmitry differs from the auto-correlation or from AMDF. I can say only that the human's pitch perception mechanism differs from these waveform comparison methods. The spectral analysis also doesn't work well. Sorry for my probably poor English. Vladimir.