Time Stretching And Pitch Shifting of Audio Signals

[seraph]

http://www.dspdimension.com/admin/time-pitch-overview/

[Antimon]

I often forget to check this part of the forum - I've skimmed a bit through these tutorials you've posted, Seraph. They look marvellous! All those bored evenings are saved - thanks!

/Stefan

[seraph]

Stefan
whenever I log in I click on View unread posts to make sure to check out all new ones. Of course they are too many to check them all out but...

[Acoustic Interloper]

I did a bit of this last year inside Live for *Ordinary Machinery* -- Live only lets you do this for samples, not LIVE audio streams (maybe the upgrade does?), and time stretching has to be multiples or divisors of powers of 2 -- so I used it only on my daughter's voice samples.

This January I tried pitch shifting on a mic'd banjo inside of Max/MSP (with the help of a couple folks here for this Max/MSP newbie), using FFT to get it into the frequency domain and using MSP's gizmo~ for the pitch shift. I had to make the sample window size fairly big in order not to get modulation artifacts (still get a few at high pitches), and I found an old piece that seems to work, but it didn't sound like what I expected:

1. Shifting up an octave makes it sound somewhat like I am playing bedsprings, and down tends to be a bit muddy. Some of this is probably the software, but I also suspect that some of those banjo timbres just don't resonate with the ear after electronic pitch shifting.

For example, if I "pitch shift" up an octave the old fashioned way by playing up the neck (no computer or mics or anything), both the pitch AND the timbre of the instrument change. But since true electronic pitch shifting preserves the harmonic relationships of the partials, it transports timbre from a lower register to a higher register that is *not* preserved by just playing up the neck on the acoustic-only instrument. I think this accounts for some of the "bed springs" on shifting up, and the "mud" on shifting down -- the ear hears those 'preserved timbres' differently after pitch shifting.

2. The window sample size for the FFT introduces some delay. The delay seems a little sensitive to the direction and magnitude of the pitch shift -- although it shouldn't, and maybe this is just my hearing -- but the (unwanted) delay makes it tough to play with headphones on. I might just perform this piece with my ear stuck to the banjo head and ignore the electronic output, although working on this last night, I started to get a feel for playing against the delays without stumbling.

The final thing I noticed is that "less is more" -- layering too much of this crap makes it unlistenable -- I need to use it tastefully as seasoning. But, of course, that's what an acoustic hillbilly would do in any case.

[Acoustic Interloper]

Well, I guess if I had read one of your other posts on this topic, I would have had my answer: