Antwort: [sdiy] Has anyone built a voice pitch tracker?

[Richard Wentk]

At 13:24 13/11/2002 +0100, Christian.Bergmiller at daimlerchrysler.com wrote:
>Hello,
>
>how do this cheap guitar(or instrumen)t tuners work?
>they extract the pitch of complex waveforms really good.
>(the most of them cheap ones are slow but in some applications this may be ok)

That's why they're cheap and easy to build. You'll find most of them don't 
get a good lock till the spray of harmonics from the initial attack has 
died down. Something that takes maybe half a second to a second to lock 
isn't much use for real-time tracking.

For real-time, digital really is the way to go if you want high quality 
results. With an analogue design you're always going to run into problem 
situations.

I suppose you could do something like this: divide your estimated frequency 
by two, use a tracking filter followed by an envelope follower to see if 
there's any significant power in the octave below your estimate. If there 
is you can be fairly sure you've locked onto the 2nd harmonic instead of 
the fundamental. If there isn't, you know you're already on the fundamental.

You'd also need some kind of voiced/unvoiced detector too, so you don't try 
to lock onto sssssssibiliant noise. And a S&H to keep the last pitch if the 
sound is unvoiced, so you don't get that droopy blippy mistracking effect 
that bad guitar synths are so good at. And some hysterisis and filtering in 
all of these so they don't leap around all over the place - although the 
latter maybe isn't so critical, as you don't get many octave jumps on 
vocals outside of opera. (And Kate Bush)

All of this might work, but it would be a lot of work to get the response 
times right. I wouldn't want to be the one trying to design it. ;)

Richard