Fixing a stereo recording — channel swaps, width and gating

You get a stereo file back and something is off. The rhythm guitar is on the right when you recorded it on the left. A field recording that sounded wide in the park sounds like it came out of one speaker. The take is clean, but between the phrases you can hear the room's air conditioning. None of these needs a remix. All three are fixed by understanding what the two channels are doing relative to each other.

What a stereo file is

Two independent tracks, played simultaneously, one per speaker. Everything you hear as position comes from the difference between them. A sound equally loud in both channels appears in the middle — not because anything is in the middle, but because your ears infer a source between two speakers producing the same thing. A sound only in the left channel appears hard left. Everything else is somewhere in between.

That gives a useful way to describe any stereo file as two components:

Mid = (L + R) ÷ 2 — what the channels share, heard in the centre Sum of differences = (L − R) ÷ 2 — what they disagree about, heard as width

Bass, lead vocals and kick drums are usually almost pure mid. Room reverb, ambience and wide instruments live largely in the difference. Nearly every stereo repair is a matter of adjusting one of these two without disturbing the other.

A channel on the wrong side

The most common cause is physical: two mics into inputs 1 and 2, and the interface's channel 1 is not the side you assumed. It happens with interfaces whose inputs are labelled on the back and used from the front, with reversed cable pairs, and with cameras where the external mic's left is the internal mic's right.

The fix is literally to exchange the two channels, and the important part is that it is lossless. No sample is altered, resampled or re-encoded in the process — the same two streams are written out in the other order. If the only thing wrong is the orientation, you lose nothing by correcting it.

Two ways to know you have the problem rather than an unusual mix: check against something with an unambiguous position, like a hi-hat or a voice you watched being recorded on a known side; or look at the waveform, since a directional mic pointed at a source gives that channel a visibly higher level. If you fix a swap and the file now sounds wrong in a new way, the swap was not the fault — put it back rather than stacking a second fix on top.

A related but different fault is a polarity inversion, where one channel's waveform is upside down relative to the other. This does not sound like a position error; it sounds hollow, and the bass disappears when the file is played in mono, because the two channels cancel. Swapping channels does not fix that. Inverting one channel's polarity does.

A recording that is too narrow, or too wide

Widening works on the mid/side split above: keep the mid, turn the difference up, and the image spreads outward. Narrowing does the reverse. That is the whole mechanism, and it explains both what it is good for and how it fails.

Where it helps: a spaced-pair recording that ended up narrower than the space felt, a synth pad that should surround rather than sit in front, or a mix where two elements are fighting for the centre and one can move out.

Where it hurts, and this is worth knowing before you reach for it:

Mono compatibility. Widening turns the difference up, and the difference is exactly what cancels when the two channels are summed to mono. Push the width hard and a phone speaker, a club PA on one side, or a smart speaker may play something noticeably thinner than what you approved. Always check the result in mono, not just as a sanity test but because that is how a good share of listening happens.

The hole in the middle. Beyond a modest amount, the centre thins out and the mix reads as two separate speakers with nothing between them rather than a wide field. On a vocal-led recording this is obvious immediately.

Noise gets wider too. Mic self-noise and room hiss are largely uncorrelated between channels, so they are mostly difference, which means widening raises them faster than it raises the music.

Small amounts do a lot. If you have gone far enough that you are worrying about whether it still sounds right, you have gone too far.

Hiss between the phrases

A noise gate mutes the signal when it falls below a threshold, so the quiet gaps become actually quiet. It is a level test, not noise removal — it cannot touch hiss underneath the music, only hiss that is alone.

Four controls, and they matter in this order:

Threshold — the level below which the gate closes. Set it just under the quietest thing you want to keep. Too high and quiet notes, breaths and reverb tails get chopped; too low and nothing happens.

Attack — how fast it opens. Fast for percussion, or you shave the transient off the front of every hit. Slower for sustained sources, where an instant open can click.

Hold — how long it stays open after the signal drops. This is the control that stops the gate chattering open and shut on a wavering sustain, and it is the one people skip.

Release — how gradually it closes. Too short and reverb tails are truncated dead, which sounds far more artificial than the noise you were removing. Let the tail through.

The thing a gate does badly is a source recorded at a similar level to its noise floor. If the hiss is nearly as loud as the quiet passages, there is no threshold that separates them, and the honest answer is to re-record or to use spectral noise reduction rather than to gate harder.

Order matters if you are doing more than one of these. Fix the swap first, since it changes nothing but orientation. Gate next, while the noise is still narrow and easier to separate. Widen last, so you are not amplifying noise you were about to remove.

The audio channel swapper exchanges the two channels without re-encoding, the stereo widener adjusts the mid/side balance with a mono check to hand, and the noise gate exposes all four controls above rather than a single "strength" slider — all three in the browser, so nothing is uploaded.