Time-stretching and pitch-shifting explained: how the algorithms work and which one to pick

Drag a loop into a project at a different tempo and your DAW quietly does something that would have needed a tape machine, a razor blade and an afternoon in 1975. It stretches the audio in time without moving its pitch. Pull a vocal up two semitones and it does the reverse. Because this happens with one click, most producers never choose the algorithm doing it, and then wonder why their drums smear or their vocal sounds like it was recorded down a drainpipe.

This guide explains what is actually happening, in plain terms, and gives you a way to pick the right mode for the material in front of you. If you are chopping samples rather than stretching them, our sampling and chopping workflow covers that side; if you are tightening a performance, see vocal editing, comping and timing.

The problem in one sentence

Playing audio faster raises its pitch, because speed and pitch are the same thing to a tape head or a sample player. Every time-stretch algorithm is a trick to pull those two apart, and every trick leaves a different kind of fingerprint.

Family 1: re-pitch (no trick at all)

The simplest mode just changes the playback speed, like a turntable. Tempo and pitch move together. Ableton calls this Re-Pitch in its Live 12 manual; Logic's nearest equivalent is the Speed effect mode in Flex Time.

It is the only mode with zero artefacts, because nothing is being reconstructed. That is why a lot of jungle, house and hip-hop sounds the way it does: a break sped up on a sampler gets brighter and tighter at the same time. If the pitch change suits the track, re-pitch is not a compromise. It is the cleanest option you have.

Family 2: time-domain slicing and overlap-add

The next family works on the waveform directly. It chops the audio into short pieces and either spaces them apart (to slow down) or overlaps them (to speed up), crossfading between pieces so the joins do not click.

Two variations matter in practice:

• Transient slicing. The algorithm finds the hits (kick, snare, pluck) and moves each one to its new position, leaving the audio between them alone. Live's Beats mode and Logic's Slicing and Rhythmic modes work this way. Drums stay punchy because the attack is never touched. The cost is what happens in the gaps: when slowing down, there is a hole after each hit, which the software fills by looping or fading the tail. Live's Beats mode lets you choose how that tail is handled.
• Grain-based overlap-add. Rather than following transients, the algorithm cuts small grains at a regular rate and lays them down at a new spacing. Live's Tones and Texture modes are of this kind, with a grain size control. Small grains suit clear, monophonic sources; larger grains suit pads and noise, where a little smear sounds like part of the texture. If this sounds like a synthesis technique, it is: our granular synthesis guide uses the same idea on purpose.

Time-domain methods are cheap on the CPU and keep the character of single sources well. They struggle with dense, full mixes, because no single slice length suits a kick, a bass line and a reverb tail at once.

Family 3: frequency-domain (the phase vocoder)

The third family converts audio into its frequency content, frame by frame, changes the spacing of those frames, and resynthesises. The core idea, the phase vocoder, was described by James Flanagan and Roger Golden at Bell Labs in a 1966 paper, originally for speech coding rather than music.

The difficult part is phase. When you move frames apart, each frequency's phase has to be advanced consistently, or the result sounds hollow and smeared, the "phasiness" that cheap stretchers are known for. Jean Laroche and Mark Dolson showed in 1999 that locking the phases of bins around each spectral peak reduces that problem considerably. Most modern "complex" or "polyphonic" modes descend from this work, with added transient detection so drums are not blurred.

In Live, Complex and Complex Pro sit here; Logic's Polyphonic mode does too. The open-source Rubber Band Library, a library built for developers to integrate into their own apps, is another example; its integration notes offer a choice between a "Finer" engine and a "Faster" one for when CPU power is limited.

These modes handle full mixes and chords best. Their typical faults are softened transients, a slight "underwater" quality on large stretches, and formant shift on voices when pitching.

Pitch-shifting is time-stretching plus resampling

Most pitch-shifters are a stretch followed by a speed change. To raise pitch by an octave, stretch the audio to twice its length, then play it back twice as fast: the duration returns to normal and the pitch doubles. So the same artefacts apply, and the algorithm choice still matters.

The extra issue with pitch is formants, the resonances of the voice or instrument body that make a voice sound like a particular person. Shift a vocal up and its formants move up too, which is the "chipmunk" effect. Live's Complex Pro mode has a Formants control; preserving formants keeps a voice sounding like the same singer at a new pitch.

A step-by-step way to choose

1. Ask whether the pitch change is acceptable. If yes, use re-pitch. It is clean and often sounds better.
2. Is it drums or anything percussive? Use a transient mode (Beats in Live, Slicing or Rhythmic in Logic). Check the transient markers sit on the hits before you judge the sound.
3. Is it one voice or one instrument line? Try a grain or monophonic mode (Tones in Live, Monophonic in Logic). Adjust grain size by ear.
4. Is it a full mix, chords or a stem with lots going on? Use the frequency-domain mode (Complex or Complex Pro, Polyphonic). Expect higher CPU use.
5. Is it a pad, texture or noise? Texture mode, or a large grain, often sounds more musical than a "clean" mode.
6. Pitching a voice? Turn on formant preservation unless you want the character change.
7. Keep stretches small where you can. Every algorithm sounds better at a few percent than at 50%. If a loop needs a big change, consider re-recording or re-pitching first and stretching less.
8. Render, then judge. Real-time warping in a busy project can fall back to lower quality settings in some DAWs. Bounce the clip and listen in solo and in context.

What we think

Artefacts are not failures. They are the sound of a decision, and the producers who sound most like themselves are often the ones who chose the "wrong" mode deliberately. Re-pitched breaks, smeared Texture-mode vocals, the stutter of a Beats-mode loop slowed too far: each became a signature because someone heard it as material rather than an error.

The real risk for working artists is the opposite: never choosing at all. Default modes are tuned to sound acceptable on average material, which means they sound acceptable on yours and distinctive on nobody's. Spend ten minutes with one loop, run it through every mode at three tempos, and write down what you hear. You will learn more about your own taste from that than from any preset.

We would also push back on the idea that newer, "smarter" stretchers make the choice obsolete. Better phase handling removes the obvious faults, but it also removes the character. Keep the clean mode for when the audio has to disappear into the track, and use the others when you want the listener to hear that something was done.

For clips that came out of separation or generation tools, which often carry their own smearing before you stretch them, our guide to fixing generated stems in your DAW is a useful next step.