What's the best sound compactor?

What is the best way to speed up the tempo of an audio file without raising the pitch or losing sound quality?

I've experimented a bit with using the 'time and pitch machine' in Apple Logic, but the process produces artifacts. Ableton Live does the same thing, in my experience. Has anyone else had success speeding up an audio clip without losing any (or at least much) sound quality?

posted by umbú (12 comments total)

Man, this is a tough one, and the only time I've experienced it was on accident. I've posted songs here that were either too fast or too slow because I used the wrong settings while saving and exporting.

I'm sure this isn't the "right" way of going about it, but when it happened to me I didn't notice any loss or artifacts.
posted by snsranch at 4:13 PM on February 13, 2009

Rubberband is the most talked about time stretcher in the open source world, it is extremely configurable (as you would expect from open source) and has an offline non-realtime operation mode (giving it the ability to look ahead at future samples and use processing algorithms that are more accurate but slower than realtime).

<GEEKSPEEK>
There is no such thing as a time stretcher without artifacts, if you are working with pcm data more complicated than a simple sine tone. There probably won't be any magical new algorithm that totally solves this, ever, because of the time bound nature of periodic signals. The best known way to abstract periodicity from a time series (of which pcm data is a subset) is fourier transform (of which the discrete fourier transform (dft) is a subset, the fast fourier transform (fft) being the most computationally efficient dft and thus the one you see used in audio software). I could be wrong (I tried to look the answer up on wikipedia but the answer just flew over my head, the math on the page beyond me), but I seem to recall that there are signals of finite duration that cannot be described by a fourier transform with a finite number of bins (and, on top of (because of?) this, an fft is inherently lossy on all but the most trivial of signals).
</GEEKSPEEK>

Since there is no way to do time stretch that is guaranteed artifact free, different parameters of a stretching algorithm will be appropriate for different source material, so something at least as tinkerable as rubberband is probably your best bet.
posted by idiopath at 8:13 PM on February 14, 2009

Thanks! That makes sense, idiopath. I mean, not that I understood all of the technical details, but it makes sense that there will always be some artifacts. I'll give Rubberband a try.
posted by umbú at 9:15 PM on February 14, 2009

ableton is probably the most effective tool for this, and you don't even need to own it in order to save time shifted files.

adobe audition also has a stretch/compress function which i've used for relatively small bpm changes.
posted by kimyo at 4:16 AM on February 16, 2009

I've done quite a bit of timestretching / time compressing of songs, samples etc in my time. And used a wide variety of apps that all have slightly different Algorithms.

Ableton live was one of the first one that I was aware of that didn't convert to a fixed percentage of the orignal. - (ie you could easily 'straighten out' the tempo of a song. ) But I don't think it is the best 'sounding'.

I was also under the impression that Most algorithms use a kind of 'Granular Synthesis Approach'.

As said above none are without some Artifacting.

I think you will find that the best sounding results can be had with 'Offline / Non-Realtime' algorithms, I used to find the Sonic Foundfry Timestretch/Time Compress was quite good. but that was year ago.
posted by mary8nne at 5:31 AM on February 16, 2009

kimyo's always on about abletone, he must be shilling for them, no?

(i wish they would sponsor us, they really should! even if it just meant free software/add ons)

anyway, i do agree with mary8nne that 'offline' algorithms are best.

umbu: would you consider posting the track you're working with? then we could all have at it and see what works best.....
posted by kimyo at 3:13 PM on February 16, 2009

we could all have at it and see what works best...

Hey, kimyo, what a cool idea. Nice use of MeFi Music, that'd be.

And, hi, umbú. I don't have enough experience with time stretching or compressing to help answer your question, but... just wanted to say: hi!
posted by flapjax at midnite at 9:09 PM on February 16, 2009

I would recommend that you take a look at REAPER an excellent DAW (and a fully functioning demo available and v cheap for non-commercial use).

It has built in time stretching using the Elastique algorithm, wrapped up in a great interface which focuses on workflow. I have retired ProTools LE in favour of Reaper and have not looked back since.
posted by the_very_hungry_caterpillar at 5:45 AM on February 17, 2009

Hey flapjax! That new video of yours is brilliantly mesmerizing.

kimyo, as an example, what about this song, either the interlocking opening loop, or the whole tune. I'm thinking of changing the tune up to 10-13%. Oh, and please excuse that terrible buzz. chococat's right. It sounded ok in crappy speakers, but now that I have some good headphones, it sounds pretty horrible.
posted by umbú at 6:23 AM on February 17, 2009

I wouldn't recommend actuaually speeding that song up. can't you re-recored it?

BUT if you MUST speed it up I should point out that you will probably get the best results by doing every individual part by itself. I assume you still have the Multi-tracks/Stems?

If you still have the Multi tracks then you could speed up each individual track using an algorithm that works best for that sort of instrument / sound and then put it all back together again.

I think this would sound MUCH better in the end - you would get less noticable artifacting cause it generally works a lot better on individual instruments.
posted by mary8nne at 10:59 AM on February 17, 2009

Oh, and I have done this myself on Remixes where i have been given the Stems and I 'really' just wanted to make it a bit faster or slower.

By individually fine tuning each individual part then I was able to get it to be almost transparent when you actually put it all back together. (cause the individual artifiacting on each track is sort of hidden in the mix).

If youa re working from just the stereo final mix then the main problem is that the sample size for say the vocals - will make the bass sound chunky or the drums to short...
posted by mary8nne at 11:04 AM on February 17, 2009

scan here for 'the mic', 15% faster via adobe audition, surprisingly okay, although the guitars got a little jangly.

but i concur with mary8nne yet again, if you have the separate tracks doing each one individually is going to be much better, or, re-recording, cause the feel will be different too.

i like that tune, at its original tempo, when i first listened i was sad it was over, which isn't something i find myself thinking on a regular enough basis.
posted by kimyo at 1:51 AM on February 19, 2009