Experiment - Voices as Music
I'm working to figure out how to analyze the dominant frequency of an audio signal. This takes me into the world of fft~. In an effort to avoid further reading, I search the MAX/MSP forums for audio to midi conversion.
This yields a patch called fiddle~ which is supposed to follow pitch. The fiddle patch appears to only run on Macintosh... it requires externals.
I return to the MAX/MSP documentation and try to puzzle over fft~. I read the tutorials in chapters 25 and 26. The documentation seems to present a progression. The fft~ object performs basic Fast Fourier Transforms, but doesn't take into account the overlap (?) of sample bins. I was skimming, so this is all a little unclear. All I'm looking to do is take frequency readings every "n" seconds in order to construct a "melody" out of some recorded speech. The tutorials seem to indicate I'll need to "window" my samples and that the pfft~ object takes this problem into account. There's even a diagram of a patch that shows how to read off the current frequency. The problem is, I don't know how to create an embedded subpatch for pfft~ to use. I try, instead, to create a named subpatch.
This works, but I found that the output I was sending out of the subpatch I create didn't change at all. Reading the tutorial more closely:
Note that in the above example the number~ object is used for the purposes of demonstration only in this tutorial. When DSP is turned on, the number displayed in the signal number box will not appear to change because the signal number box by default displays the first sample in the signal vector, which in this case will always be 0. To see the center frequency values, you will need to use the capture~ object or record this signal into a buffer~.
I added a capture~ object to the output of my test program, but this didn't work.
