"old school" pitch shifting

Roxas · Post by **Roxas** » Thu Jan 06, 2011 11:43 am

hello everybody, first, I'm a begginer, so sorry for this question:

I was checking the website of a famous VST plugin maker, and found this:

"Modern pitch-shifters use sophisticated mathematical algorithms to transform music and vocals as naturally as possible. The original old school devices (like Eventide's classic H910 Harmonizer) used a resample and cross-fade technique that introduced audible 'glitches' in the pitch-shifted audio."

is this (the old school one) the same approach you use to take to get pitch shifting in the FV1, so the same described on the knowledge base page, or is something different and more "rude"?

thanks everyone!

frank · Post by **frank** » Thu Jan 06, 2011 12:06 pm

FV-1 basically does the old school technique because it is designed to do it real time. If you are pitch shifting a file on a computer then using advanced techniques will give better results but the techniques are much slower and more complex so not realistic to do real time.

Roxas · Post by **Roxas** » Thu Jan 06, 2011 12:58 pm

many thanks
I believe every real time effect we use for guitar or other use the same technique then?

I didn't understand very well the explaination of Pitch Transp. vs Pitch Shift on the website (that's my fault, I'm not english), but is Pitch shift not more than a pitch trasp. like the one described here http://www.spinsemi.com/knowledge_base/ ... nsposition, but using an increment instead of a factor? that's the only difference?

frank · Post by **frank** » Thu Jan 06, 2011 2:12 pm

Yes, we are really doing a pitch transpose. We are basically multiplying by a factor rather than adding an increment. For example, in transposing we may want to transpose up by an octave, so all frequencies are multiplied by 2. Therefor 1Khz becomes 2KHz (1KHz * 2), 3KHz becomes 6KHz (3KHz * 2), etc. This is nice and musical because the ratio between signals is maintained.

In shifting, a constant value is added to the frequency. If we want to shift by 1KHz we add 1KHz to all frequencies so 1KHz becomes 2KHz, 3KHz becomes 4KHz, etc. This is not musical because the ratio between signals is not maintained but it can have desirable effects like slightly shifting a persons voice by a small increment.

So while we tend to say "pitch shifting" we typical mean "pitch transposing".

Roxas · Post by **Roxas** » Fri Jan 07, 2011 5:26 am

many thanks again Frank

that helped a lot!
so I will study better "pitch transposing" on the knowledge base page and then try to get something
cheers

Roxas · Post by **Roxas** » Fri Jan 07, 2011 9:04 am

another question

some shift effects claim to be polyphonic, for example the POG. others, like the whammy, glitch a lot if you play a chord or more than 1 note.
the difference is in the software-algorythm or in hardware limits?

livingston · Post by **livingston** » Fri Jan 07, 2011 12:15 pm

Frank can respond with a deeper answer probably, but it has to do with the speed of the LFOs which do the pitch bending and crossfading. If the LFOs oscillate at a frequency related to the pitch you're playing, you can sometimes get a periodic cancellation which sounds like wobbling or chirping.

But it's not really a hard line between "polyphonic" and "monophonic" - The Whammy gets warbly, and so does the FV1 pitch transpose program, but in fact so do the HOG and POG - I've used both and gotten these warbly sounds in both. So you can minimize the phase cancellations to an extent, and get less "glitchy" but there's no technique, except for maybe fourier transform which is not possible in the FV1, which would completely remove warbling.

frank · Post by **frank** » Fri Jan 07, 2011 1:31 pm

Actually, nothing I can really add to Taylor's answer.

Roxas · Post by **Roxas** » Fri Jan 07, 2011 1:47 pm

and what about the autocorrelation trick? is it usually used?

frank · Post by **frank** » Fri Jan 07, 2011 5:09 pm

Not sure of that trick, link?

Roxas · Post by **Roxas** » Sat Jan 08, 2011 9:28 am

yes sure,
I found it here
http://valhalladsp.wordpress.com/2010/0 ... glitching/
and I read that in other parts too that I don't remember right now

frank · Post by **frank** » Sat Jan 08, 2011 10:48 am

OK, I'm familiar with that process. Never tried to do it in the FV-1, it is a complex thing to do to so may be too complex for the FV-1.

Roxas · Post by **Roxas** » Sun Jan 09, 2011 6:01 am

many thanks again to both of you
I've run out of questions for now

seancostello · Post by **seancostello** » Thu Jan 20, 2011 10:56 pm

I tried doing some simple auto-correlation on the FV-1. I couldn't get it to work. Theoretically, you could use the JAM instruction to get things to reset, but the problem is figuring out the proper reset point. The early de-glitching pitch shifters used dedicated auto-correlation boards for computing the correct splicing points.

Today, you can use FFTs (accelerated by vector libraries) to compute the autocorrelation for splicing in pitch shifters. Or, you can use the same FFTs for phase vocoding pitch shifting, which seems to produce "better" results. Of course, plenty of people like the artifacts of the old school "glitchy" pitch shifters.

Sean Costello

seancostello · Post by **seancostello** » Thu Feb 17, 2011 11:56 am

livingston wrote: But it's not really a hard line between "polyphonic" and "monophonic" - The Whammy gets warbly, and so does the FV1 pitch transpose program, but in fact so do the HOG and POG - I've used both and gotten these warbly sounds in both. So you can minimize the phase cancellations to an extent, and get less "glitchy" but there's no technique, except for maybe fourier transform which is not possible in the FV1, which would completely remove warbling.

I am fairly certain that the HOG and POG use FFTs for their pitch shifting. Probably phase vocoder, maybe phase locked vocoder as described by Puckette. My MicroPOG sounds very smooth, but it isn't perfect. I've never heard perfect polyphonic pitch shifting, and it may not be possible.