Why slowed audio sounds underwater

“Slow it down” describes two different operations. They sound different, and most questions about pitch shifting come from expecting one and getting the other.

Varispeed: the honest one

Play a record at half speed and every frequency in it halves. Pitch is how often a waveform repeats per second, so halving the playback rate halves the repetition rate: an octave lower.

Tape does this. So does a turntable, and so does changing the sample rate. It sounds natural because it is what physically happens — and it is no use for practising along, because the recording is now in a different key from your instrument.

Time-stretching: the useful one

Holding the pitch steady means rebuilding the audio. The engine cuts it into grains a few tens of milliseconds long, then re-schedules them: spaced further apart to slow down, overlapped to speed up. Each grain still plays at its original rate, so the pitch stays put. Only the timing changes.

The grain boundaries are where it goes wrong. At every join, two pieces of waveform meet that were never adjacent; if they meet out of phase the result is a click, a smear or a faint ring. That happens thousands of times a second, and the accumulation is what people hear as watery, phasey or underwater.

Why it gets worse at extremes

At 0.9× the grains barely have to move, and the artefacts are inaudible. At 0.25×, each grain has to be stretched or repeated four times over, so there are four times as many joins and each one is under more strain. That is why the floor is 0.25× and not lower: past that point the errors stop being artefacts and start being the dominant sound.

Which is why there is a quality setting

Better grain placement costs processing. Capo ships three time-stretch engines and the Audio quality setting picks between them: the lightest is cheap and smears sooner, the heaviest holds together further down. Below about 0.5× the difference is obvious; above 0.8× you will struggle to hear it at all.

Settings · audio & playback

1Audio & playback

1aAudio quality1bSkip interval1cCount-in duration

The reasoning behind having three rather than one is in why there are three engines and not one.

Voices are the hard case

Voices suffer more than instruments. A voice carries resonances set by the size of the throat and mouth — formants — and listeners are sensitive to them because they encode how big the speaker is. Smear those and the change registers immediately, even when the listener cannot name it.

That sensitivity is also why transposing a voice needs a separate correction — covered in singing a song that sits four semitones too high.

Two things this does not settle

It does not explain why some recordings survive slowing far better than others. Dense, loud, heavily limited masters have less room between transients and tend to smear earlier than sparse acoustic recordings, but that is a rule of thumb rather than a rule.

It also does not cover pitch shifting, which uses the same grain machinery in the opposite direction and has its own failure modes.