Auto-tune?

jamos

Hey all -

I'm trying to make an autotune patch for vocals - yes, it's poor taste, but so what? I'd think it would be easy: use a pitch tracker to extract pitch, quantize that, take the difference between the quantized and unquantized signals and use that to drive a pitch shifter. But, alas, it doesn't seem to do anything: as near as my ears can tell, the difference between the two is always 0, so no pitch shift occurs.

The sub-patch is below; can anyone tell me what I'm doing wrong?

Autotune.jpg

Description:

Filesize:

30.12 KB

Viewed:

4952 Time(s)

blue hell · Posted: Sat Nov 21, 2009 1:39 pm Post subject:

Nice idea!

I patched it up and some quantization does seem to occur to me, although glides are not fully eliminated ... thinking

BTW, couple of weeks ago I was listing to autotuned singing with gypsy scales .. used in hip hop like music ... I liked that Smile

jksuperstar · Posted: Sat Nov 21, 2009 3:05 pm Post subject:

Interesting! I've had this type of thing on my idea-plate for a while. I wanted to play a didgeridoo, but send in a sequence to pitch shift it, but do it based on the difference between the didge sound and the sequence. Very similar! I'll try to play with it tonight.

Antimon · Posted: Sun Nov 22, 2009 5:30 am Post subject:

Why are you taking the difference between the original signal and the note quantized one? When I've tried my hands at this I've taken a pitch detector, note quantized its output and then plugged that straight into an oscillator pitch input, which then goes into a vocoder. I've had varying results (it's really tricky to get the pitch to "stay on target" for me), but I don't usually have any problem with the pitch itself.

/Stefan

Tim Kleinert · Posted: Sun Nov 22, 2009 7:30 am Post subject:

Auto-Tune, an interesting subject. Smile

Hope nobody minds if I chime in.

AFAIK there is already a patch in the factory preset bank called "pitch correct" (or something like that) which does this kind of thing.

However, keep in mind that the pitch-shifter is based on a delay line, and therefore naturally has an amount of latency. This causes slurs and other artefacts when correcting notes. The patch mentioned above also suffers from this.

Some time ago, I attempted to build a pitch correction patch that is based on a different approach, extracting individual wavecycles of the incoming signal, pitching them according to the correction scheme, and redistributing them along the time-axis, which results in an artefact-free pitch shift (under the condition that the incoming signal is strictly monophonic and periodical).

The basic pitch shifter circuit:
http://www.electro-music.com/forum/topic-27228.html

The concept of the auto-tuner itself works, but the pitch detection isn't bullet-proof. I've since attempted to design the "maximum zero gradient" pitch detection algorithm on the G2, but got fed up with it at a certain point. Laughing

That technique isn't optimal for voice anyway. The best would be AMDF, but that is simply impossible to patch on the G2.

Here's the first version. It has some issues (mentioned above), but anyway...
http://www.electro-music.com/forum/topic-26222.html

I made another one, based on a similar idea. This works better:
http://www.electro-music.com/forum/topic-26247.html

Maybe this is of use to someone.

cheers,
t

jamos · Posted: Sun Nov 22, 2009 1:39 pm Post subject:

jamos · Posted: Sun Nov 22, 2009 2:25 pm Post subject:

Antimon · Posted: Sun Nov 22, 2009 2:48 pm Post subject: