Evan

I've been asked to create a Max patch/device that controls a mono synth according to vocal input. The development is moving along, but I figured I'd ask around to see if ay users have any valuable advice on as to getting the pitch tracking to be awesome.

I've got a basic signal flow concept of input -> gate -> Eq -> Compression -> Pitch tracking. Does this seem about right? Im curious as to the difference if I put the Compression before the EQ. I just figured I'd only want to compress the data that I will actually be using (frequencies within the range of the human voice).

I've implemented it with retune~, but I am also thinking about investigating fzero~. My experiments with retune have been encouraging, but it's hard to accurately track the octave of the fundamental due to the complexity of the voice. On a synth it works incredibly well. I've had trouble getting a stable reliable pitch and 'note' events from fzero~. It seems to want to spit output data constantly, including the onset indication. I was thinking of cooking something up in Gen to tame the data and get it solid.

Any advice, anecdotes, or warnings would be greatly appreciated.

I've been asked to create a Max patch/device that controls a mono synth according to vocal input. The development is moving along, but I figured I'd ask around to see if ay users have any valuable advice on as to getting the pitch tracking to be awesome. 

I've got a basic signal flow concept of input -> gate -> Eq -> Compression -> Pitch tracking. Does this seem about right? Im curious as to the difference if I put the Compression before the EQ. I just figured I'd only want to compress the data that I will actually be using (frequencies within the range of the human voice). 

Any advice, anecdotes, or warnings would be greatly appreciated.


vocal-pitch-tracking-best-practices

try using the sigmund~ external. you'll no doubt get better results with it!

No 32 bit. Building this for compatibility going forward. Sigmund~ is not 64 bit yet (tis a shame, because I would've used it).

yikes. you might be restricted to fzero or retune (i'm not familiar with retune) then

I'm confident I'll be able to get usable, stable tracking with a bit of work and some careful treatment of the data.

Can you post your patch? Are you using the default fzero~ settings?

I think that the onset detector in fzero~ might be a little too nervous. I'm not sure if it can be tweaked to work bettor. The help patch for fzero~ uses the amplitude output of fzero, rather than the onset detection, for this reason. That might be a good approach to explore.

Can you post your patch?  Are you using the default fzero~ settings?  

I think that the onset detector in fzero~ might be a little too nervous.  I'm not sure if it can be tweaked to work bettor. The help patch for fzero~ uses the amplitude output of fzero, rather than the onset detection, for this reason. That might be a good approach to explore.


Yeah I'll post the pitch detection thing later today, not at my machine currently.

I am not using the onset detection of fzero, it is indeed much too nervous. I am doing some amplitude tracking and driving the envelope of the synth from that. I'm doing some compression and filtering prior to the pitch tracking, so I don't think that using the amplitude output of fzero~ will accurately reflect what's coming into the patch, but it might be worth a look. One of the reasons I initially avoided that is because the data coming out of fzero~ isn't at signal rate, which is kind of a bummer.

Yeah I'll post the pitch detection thing later today, not at my machine currently. 

I am not using the onset detection of fzero, it is indeed much too nervous. I am doing some amplitude tracking and driving the envelope of the synth from that. I'm doing some compression and filtering prior to the pitch tracking, so I don't think that using the amplitude output of fzero~ will accurately reflect what's coming into the patch, but it might be worth a look. One of the reasons I initially avoided that is because the data coming out of fzero~ isn't at signal rate, which is kind of a bummer.


Here's a small part of the patch taht deals strictly with the pitch tracking. If any one is still interested in taking a look, I'd love to hear any suggestions. There's some controls for various parameters, and some switches to play with different sources and objects.

A 64-bit version of sigmund~ is now available:

https://github.com/v7b1/sigmund_64bit-version/releases

BTW, filtering out all the high frequencies is not necessarily a good idea, since some tracking algorithms use the harmonics to figure out the fundamental.

A 64-bit version of sigmund~ is now available:
https://github.com/v7b1/sigmund_64bit-version/releases

You might also try vb.pitch~
http://vboehm.net/downloads/

Vocal pitch tracking best practices