Technical help on time stretching for a written paper at college

Maybe someone with some inside knowledge about the specific algorithm in LP7 can tell you about it. But the principles of time compression and formant preservation are probably easily found on the web.

In short, time compression or expansion involves deleting samples or adding samples, respectively, to any given piece of audio, in order to conform the audio to a given change of length. Time compressing is usually more successful than time expansion, because in general it ends up being more convincing-sounding to remove samples and smooth out the resulting vacancies in the discontinuous waveform that results than it is to artificially add samples and try to smooth all of that out. (Again, that's a vast generalization.)

There are various algorithms for time comp/exp, including resynthsis and phase vocoding, and these processes may be worth looking up as well.

Logic uses many DSP algorithms, it's own, QuickTime and 3rd party for Time Comp and Expansion. I haven't seen any of those algorithms published.

Is it really Logic specifc?

P.S. What exactly are you looking for and what is the paper/program/major? You could be a Music, Liberal arts, or Computer Science maj. - this info would help you get a more appropriate answer. If you already know more about Time Stretching algos than 90% of the Logic users, let us know that.

If you are looking for non-Logic-specifc info on the DSP required

http://www.dspdimension.com/

is a very good resource. The guy who made the page is a programmer, etc - his algos are used by Steinberg and others.

Another thing that hasn't been mentioned in this topic is realtime calculation, i.e. AppleLoops.

So... specifics.

Cheers,

J

my paper is on the pro's & cons of time stretching
using both hardware & software the hardware i will be
featuring is the Akai S950 sampler as used by fatboy
slim dr dre etc & software i will be featuring Logic
Pro 7 , Pro Tools 7 , Cubase SX3 & also Ableton live
5

All time stretching in this context is software; the hardware difference is only whether it is a dedicated music computer (e.g. Akai S950) or a multipurpose computer (Mac).

For a little extra interest, why not start with a description of time itself, which is not a constant, but relative? Fast, slow, it's all relative based on motion.

discussing time stretching in music with respect to time stretching in nature could be interesting.



G5 Quad + PB G4 + Logic 7.2.3 + RME FireFace 800 Mac OS X (10.4.8)

You're Welcome. Pros and Cons in this case... hmmm.... well, I can tell you that timestretching and sample rate conversion ( related ) between hardware and software is very similar as Matti said - it's also quite different. Mostly because it can advance more quickly in a open environment ( computer vs. hardware ). Many hardwares do have these features, but do not allow you to interface as users are now accustomed to. Also, resources including speed is another issue. Developers try things they wouldn't in a hardware environment because it would be too slow. End user expectations ( affordability ) seems to be what is paramount - a lot of that has to go into R&D since the market is quite small. So the HW is compromised.

Of course, the artefacts may be desirable - but many Logic users don't even rely on Logic capabilities and actually move those tasks to Melodyne, iZotope, Serato... all of which specialize in that area.

Since computers are generally faster, you'll be able use more complex algorithms within reasonable times - thus allowing you to use more sophisticated interpolation. Interpolation... there may not have been much on an S950... it ( S900 for sure ) allowed only 12 bit sampling. Akai did move forward with this in future models - and allowed users means to stretch in a sample editor and fly back to the sampler. Hardware stretching still happens, but it's not progressed as soft has. just my 02

J

You're Welcome.

Not only is time stretching a way to degrade, but the converters, and perhaps most importantly, the realtime resampling for playback purposes. Most hardware samplers use a method where playback rate is altered to adjust pitch and time ( hand in hand ). At least, for how I work, this is the most used form of the family that is retiming. Many software samplers these days have methods to use time and pitch independently - though it requires a bit of extra CPU for realtime purposes. Additionally, some may offer multiple degrees of interpolation for realtime granular playback. Granular Sampling is largely what is used for this, and all it consists of ( in simple terms ) is dicing the audio into small pieces, which are reconstructed dynamically by repeating the pieces at varying speeds and repeats - much more expensive in a computational sense, but allowed samplers to overcome the most basic limitations in realtime use ( the limitation of time, formant and pitch being an entity ).

J

Granular sampling if i am correct this degrades the signal as you use it the more you apply the worse it gets

How do you figure?

one of the things i will be looking into is does the Akai S950 still degrade the sample as you use it

How do you mean, specifically? If it plays back at a pitch other than the original, yes. If you perform a time expansion, yes - and this alters the original file.

also with Ableton you can apply a lot of time stretching or warping without any degradation at all

I doubt that - do you have a reference? There is a difference between altering the original files destructively vs. nondestructive. They must mean it's a nondestructive operation for playback, but when it is played back - it will degrade the signal. just like Apple Loops are stretched and pitched in realtime with Logic - it degrades the sound but does not alter the file.

this is all very much food for thought & i cant wait to start have to wait for my wife to get up if i start messing around with melodyne & logic etc she wont be to amused its 3 in the morning over here

Cans + coffee : ^)

Have a good night K - You're Welcome

Cheers,

J

If you want to do hardware vs. software time compression, then you might want to consider the Lexicon 2400. This little-known hardware unit was a staple in the NY remix scene in the 90's and provided real-time time-compression; in our case we used the 2400 when making transfers between two 2" decks, with the transfer usually being at a faster tempo than the original, but at the same pitch. Occasionally we would use it to slow down the tempo of a track.

One last thing so as not to overwhelm you with stuff (since you only have 3500 words) --- the Eventide Broadcast Delays. These units have been around since the 80's, I believe. Also called "profanity delays", or "7 second delays", they insert delay time into real-time audio, usually during pauses or periods of relative silence, until a certain fixed amount of delay is established (usually 7 seconds). In the event that "something untoward" is spoken, the operator hits the DUMP button on the unit at which time it switches back to real-time audio input and gets rid of (dumps) the audio accumulated in memory. Then the cycle repeats.

So there you have it, two devices which offer real-time time compression/expansion.

Good luck with your paper.

There are two broad classes of algorithms for time streching/compression: the first stays in time domain, the second goes into frequency domain.
Time domain algorithms are variations on granulation: play small chunks of sound. Consider a tape, to time strech, you read a bit of it with a play head, then you rewind half the size of the chunk and play this bit again, etc. This is good for stationary sounds, but bad for attacks, so maybe you add an algorithm to ignore attacks and strech a little more the moments without attacks. I wouldn't be surprised if the Apple loops worked this way.
If you want to go into frequency domain, you get the frequential information with, for instance, a FFT operation. Then, you go back into time domain, but you can change the time scale easily. When transposing, you can apply filters to preserve formants, before coming back to time domain.
For more information, I advise "The computer music tutorial" by Curtis Roads. But I don't know which software uses which algorithm, and the book will not tell you either.
Best,