Tempo Changer
Slow the music down to learn it. Pitch stays put.
or paste one from your clipboard
How to use it
- Drop the track you want to work on onto the page. It is decoded locally, so a rehearsal recording or an unreleased mix never leaves your machine.
- Set the tempo as a percentage of the original. 100% is untouched, 75% is three quarters of the original speed, 120% is a fifth faster.
- If you know the tempo of the recording, type it into the BPM box. The tool then tells you the beats per minute you will actually be playing at.
- Click Preview the tempo to render it and listen. Time-stretching is too expensive to redo on every drag, so it runs when you ask for it.
- Download the result, or loop a section first with the trimmer if you only need one difficult bar.
Why practising along with a recording needs this
Learning a part by ear from a record is a good way to learn, right up until the passage is faster than your hands. The obvious fix, slow the recording down. Used to mean slowing the tape, which drops the pitch with it. A solo slowed to three quarters speed on tape arrives roughly five semitones flat. Now your fingering no longer matches what you are hearing, your ear is being trained on the wrong intervals, and you cannot play along on a tuned instrument at all.
Pitch-preserving stretching removes that problem completely. The recording stays in its original key at any tempo, so you can put the metronome away, play along at 65%, and step up as the passage settles. Everything you learn at the slow tempo transfers directly, because the notes were never in a different place.
The same property matters outside practice rooms. A cue that has to fit a picture edit can be nudged a few percent without anyone hearing that the key moved. Two takes recorded on different days at slightly different tempos can be pulled into agreement. A sample can be made to sit on a grid without being transposed off the chord it belongs to.
How the stretch is done
The technique is overlap-add with a similarity search, usually called WSOLA. The audio is cut into windows of about 46 milliseconds, each faded in and out with a raised-cosine shape. To slow the recording down, those windows are written to the output further apart than they were read; to speed it up, closer together. Because each window still contains real waveform at its real frequency, nothing is transposed.
The difficulty is the joins. Two windows dropped next to each other at arbitrary phase will cancel or reinforce in ways that sound like a comb filter sweeping past. So before writing each window the algorithm looks at a small range of nearby read positions and picks the one whose waveform best continues what has already been written. That search is why a stretch takes a second or two rather than being instant, and it is the difference between a usable practice tool and something that sounds like a broken CD.
Where the limits are
Stretching is interpolation, and interpolation invents nothing. At moderate ratios (say 75% to 130%) there is very little to invent and the output is close to transparent. Push further and the algorithm has to repeat or drop an increasing share of the material. On sustained sounds that is harmless. On transients it is not: a snare hit stretched across two windows arrives twice, faintly, and reads as softness rather than as a doubled hit.
If you need a very large change and the pitch does not matter, use the speed changer instead. A plain resample is artefact-free at any ratio. If you need the pitch moved and the timing held, that is the pitch shifter. This page is for the case in the middle, where the timing is the thing that has to change and the key must not.