How to Hear What a Plug-in Removed: Delta Listening and Null Tests

By IntrectNovember 15, 2026Last reviewed MixingBleedTechniqueGuide

Every repair plug-in takes something out: bleed, noise, sibilance, resonance or residue. The question that decides whether it helped is simple to ask and easy to skip: what exactly did it take? A quick before-and-after does not answer it, because the processed version is usually quieter and our ears judge loudness before anything else. This guide shows two ways to hear only the removed part, in any DAW, and how to tell when a cut has gone too far.

We make de-leak, a bleed remover with a built-in removed-signal monitor called Residual. The technique on this page works with any tool, free or paid.

The short answer

Listen to the difference, not the result. Use the plug-in’s delta or residual button if it has one; otherwise run a null test: play the original against the processed copy with one polarity-inverted, both lined up to the sample at the same level. Turn the monitoring up, and back off as soon as you hear the performance in what was removed.

Why before-and-after listening is not enough

Removing anything lowers the level a little, and in a quick A/B the louder version tends to win whether or not it sounds better. More importantly, the processed version hides what went missing: you hear a cleaner vocal and do not notice that the consonants got softer. Hearing the removed signal on its own turns that into a direct question. Is there anything musical in there?

Method 1: the plug-in’s own monitor

Many repair plug-ins have a button that plays only what they remove. It goes by different names: delta, residual, difference, or “output noise only”. Check the manual of the tool you use. When it exists, it is the most reliable route, because the plug-in already lines the two signals up for its own latency.

In de-leak, the button is Residual. It plays dry minus wet, exactly what is being removed and nothing else, and the header switches to RESIDUAL so you know which you are hearing. Its Bypass plays the input delayed by the plug-in’s latency, so switching between Bypass and the processed signal stays in time.

Method 2: a null test in any DAW

If a tool has no such button, build the check yourself.

  1. Duplicate the track. Same region, same start, same fader level.
  2. Put the plug-in on one copy only. Leave everything else identical.
  3. Invert the polarity of the other copy. Most DAWs have a polarity (often labelled Ø) switch on the channel or in a utility plug-in.
  4. Play both. What they share cancels; what remains is the difference, which is what the plug-in removed (or added).
  5. Check the alignment. If you hear the whole mix, thin and phasey, the tracks are not lined up. A plug-in with latency delays its output; most DAWs compensate automatically, but only if delay compensation is on and both tracks go through the same path. A manual offset or a gain change breaks the cancellation too.

Without the plug-in on either copy, the test should give silence. Try that first, so you know your routing nulls before you judge the processing. Sound On Sound describes the same check: bypass the plug-in under test and the meters should read minus infinity.

A null test is not exact for every kind of processing. Sound On Sound notes that oversampling changes phase, which can make the test invalid, and the same applies to other processing that shifts phase across the spectrum. In those cases trust the plug-in’s own delta monitor over a manual null.

How to listen to the difference

What you should hear, tool by tool

ToolGood: the removed signal containsToo far: you start hearing
Bleed remover on a separated vocalDrums, instruments, roomWords, melody, the body of the voice
DenoiserSteady hiss or humBreaths, reverb tails, the top of the voice
De-esserOnly the s, sh and t soundsVowels or the edge of every word
Resonance suppressorA ring at one or a few pitchesWhole notes or the instrument’s tone
AI artifact removerGlassy fizz, hash, shimmerConsonants, breath, sung pitch

Some of what goes is a judgement call. A bleed remover also takes part of the reverb tail, because a tail looks like energy that does not belong to the dry voice. On a dry pop vocal that may be welcome. On an ambient track it may be exactly what you wanted to keep. The removed signal lets you hear that trade before you print it.

Using it on a separated vocal with de-leak

  1. Insert de-leak on the vocal stem. It loads the Medium preset with the Gate engine.
  2. Switch on Residual and turn your monitoring up.
  3. If you hear the singer, step to a gentler preset (Default on a vocal) or raise the protection on the band where the voice is leaking into the removed signal.
  4. If you hear only bleed and room and some bleed is still in the vocal, try Leak Clean or Gate → Clean on the same passage.
  5. Switch Residual off, match levels, and A/B against Bypass in the mix.

de-leak runs as VST3, CLAP and Audio Unit on macOS, AAX for native Pro Tools on Apple Silicon, and VST3 and CLAP on Windows. The manual covers every control; the product page has an A/B player whose Removed button plays what it took out of a real htdemucs vocal.

Found bleed in a stem?

de-leak has a 14-day trial for macOS and Windows hosts that load VST3 or CLAP. Comparing tools first? Read the bleed comparison.

Related guides

Sources

FAQ

What is a null test in audio?

A null test plays the original and the processed version together with one of them polarity-inverted. Whatever the two have in common cancels, so you hear only the difference: what the processing added or removed. If the two are identical, the result is silence.

Why does my null test sound like the whole song, flanged?

The two tracks are not lined up in time or in level. A plug-in with latency delays its output; most DAWs compensate automatically, but a manual offset, a different start point or a gain change breaks the cancellation. Line them up to the sample and match the level, then test again.

How loud should I listen to the removed signal?

Louder than normal. The difference is usually quiet, and at normal level it sounds empty when it is not. Turning your monitoring up by 20 to 40 dB is a common starting point; turn it back down before you bypass the check.

What should I hear in the removed signal?

Only the problem you meant to remove: drums and instruments for a bleed remover, hiss for a denoiser, s and t sounds for a de-esser. If you can follow the words, the melody or the body of the voice, the processing is taking part of the performance and should be backed off.