Audio Repair and Restoration
Recorded on the laptop mic instead of your interface: what can be saved
Published October 11, 2026 · Updated October 11, 2026
Spoken word is usually salvageable to a listenable, publishable standard. A sung vocal for a music track usually is not. The reason is that a laptop does not just use a worse microphone, it also runs the signal through processing you never chose and cannot query from inside REAPER, and that processing is already baked into the file. If you can repeat the take, repeat it. That will beat every repair below, and it takes less time.
The mistake is almost always the same. The interface was plugged in, REAPER was open, the meters moved, and nobody checked which input the meters belonged to. You find out afterwards, when the take sounds thin, roomy, and strangely uneven, as if someone had been riding the fader.
Before any plugin goes on the track, answer one question: can this performance happen again? An interview with a guest who has gone home, a live take, a one-off idea at 2am, these cannot. A verse you can sing again in ten minutes can. Everything in this article is for recordings you genuinely cannot repeat, because re-recording is not a fallback here, it is the better option.
What the laptop actually did to your audio
Three separate things happened, and they need separating because only one of them is a microphone problem.
1. The capsule is small and noisy
A laptop microphone is a MEMS capsule a few millimetres across. Published figures from one widely used part, the Knowles SPH0645LM4H-B, give a signal to noise ratio of 65 dB(A) and an acoustic overload point of 120 dB SPL. For comparison, Audio-Technica publishes a signal to noise ratio of 74 dB and a noise figure of 20 dB SPL for the AT2020, an entry-level studio condenser.
That is roughly 9 dB more noise under everything you recorded, before anything else goes wrong. Two caveats worth stating: the Knowles figure is A-weighted and Audio-Technica does not print a weighting for its noise spec, so the gap is an indication rather than a measured difference, and both numbers are representative of the class rather than a measurement of your particular laptop, because manufacturers rarely publish which part they used.
2. The capsule is in the wrong place
This matters more than the capsule. The microphone port sits in the chassis, usually near the hinge, the keyboard, or a fan. It is typically two to three feet from your mouth instead of six to eight inches. That distance ratio is why the take sounds roomy and distant: you captured far more reflected sound relative to direct sound than you would have with the mic in front of you. It is also why key clicks and fan noise are loud, because they are conducted through the same chassis the capsule is mounted in.
3. The operating system processed the signal on the way in
This is the part that surprises people, and it is the part that limits the repair. Microsoft's driver documentation describes the processing that can be applied to a capture stream, including Acoustic Echo Cancellation, Simple Noise Suppression, which "suppresses noise such as humming and buzzing", Automatic Gain Control, which is "designed to provide a controlled signal amplitude at its output, despite variation of the amplitude in the input signal", and Deep Noise Suppression, which "uses advanced AI / machine learning techniques to suppress noise particularly in voice calls".
Two details from that same documentation decide what you can do next. First, "OEMs define what effects will be used for each mode", so what a given laptop applies is a vendor decision, not a Windows standard. Second, applications "cannot find out what mode is used for each of their streams". REAPER cannot report what was done to the signal, and neither can you. Only a Raw capture stream is guaranteed clean, and Microsoft defines it strictly: raw capture "must not include any time varying or adaptive processing, such as echo control, automatic gain control, or noise suppression".
macOS has its own version of this. Apple documents mic modes reached from the video icon in the menu bar: Voice Isolation to "filter out background sounds", Wide Spectrum to "include the sounds around you", and Standard to return to normal input. Apple notes that the options vary by app, so a mode left on in one application is not something you would necessarily notice in another.
Adaptive processing changes its behaviour moment to moment. Once it is printed to a file, there is no inverse to apply. You cannot un-duck a gain control or put back a band a noise suppressor removed, because the file no longer contains the information needed to know what was taken.
Three checks before you process anything
Each check tells you something different, and the results decide whether the next section is worth your evening.
Watch the level during a long pause
Find a gap of two seconds or more where nobody speaks. If the background noise visibly swells during the pause and ducks again when speech returns, automatic gain control was active. That is the single clearest sign that the capture path was processing, not just recording.
Look at where the top end stops
Put ReaFIR on the track and leave it in EQ mode. Cockos describes it as an FFT based dynamics processor, and its display shows the live spectrum of what is coming in. Play a loud, bright word and watch the right-hand side. If the energy falls off a cliff at a fixed frequency rather than tapering, the capture path band-limited the signal and nothing above that line exists in the file.
Listen to the noise, not the voice
Solo the track and listen to a pause at raised monitoring level. Steady hiss is repairable. Noise that warbles, swirls, or sounds like water is already a processing artefact, and reducing it further normally makes it worse rather than better.
The repair order in REAPER, with stock plugins
Work in this order, and stop as soon as it sounds acceptable for its purpose. Every one of these is a starting range to be adjusted by ear on your own material, not a preset.
1. High-pass the chassis noise
Add ReaEQ, which Cockos describes as an unlimited band equalizer, and set the first band to a High pass shape somewhere around 90–130 Hz for speech. You can usually go higher here than you would on a proper vocal recording, because a laptop capsule at arm's length captured very little useful low end to protect, and plenty of desk thump and fan rumble to remove.
2. Reduce the boxiness
Recordings made at this distance in an untreated room usually pile up somewhere between 300 Hz and 700 Hz. Take a second ReaEQ band, set it to Band, boost it narrowly and sweep until one spot sounds worst, then turn that boost into a cut of roughly 2–5 dB. Judge it on a sentence, not a word.
3. Add presence only if the content exists
If check 2 showed real content above 5 kHz, a gentle High shelf of about 1–4 dB somewhere in 3–6 kHz buys back intelligibility. If check 2 showed a cliff, skip this step. Boosting a band that contains only noise raises the noise and nothing else.
4. Treat steady noise, once, carefully
If the noise floor is steady, ReaFIR's Subtract mode is the stock tool for it. The Cockos effects guide explains the method: enable Automatically build noise profile "while auditioning a section which contains only the noise", then "un-check Automatically build noise profile, stop playback and save the preset". The same guide warns that raising the captured profile costs you quality, noting that "the higher you raise it, the more artifacts you will hear". On a laptop recording that is already carrying vendor noise suppression, keep the reduction light. We cover the method in detail in how to reduce background noise in a podcast in REAPER, and if the noise changes through the recording, the non-steady case needs a different approach.
5. Even out the level last
Gain control leaves the loud parts and quiet parts closer together than a human would have mixed them, and often leaves the whole file quiet. Fix the level after the tone is right, not before, because EQ changes how loud things read. For spoken word, aim at a published delivery target rather than at a meter position, which is covered in setting a podcast to -16 LUFS in REAPER. If the whole file is simply too quiet, rescuing a vocal recorded too quietly covers the order to do it in.
What this will not fix
These limits are the honest part of the answer, and they are worth reading before you spend three hours on the track.
- It will not restore missing high frequencies. If the content was never captured or was filtered out on the way in, no plugin can reconstruct it. Anything that claims to is generating new material and guessing.
- It will not undo automatic gain control. You can compress or automate on top of what it did, but the original dynamic shape of the performance is gone.
- It will not remove the room. A distant capsule in a live room prints reflections into the signal. De-reverb helps a little and costs clarity, as covered in reducing reverb and room echo from a vocal in REAPER.
- It will not separate keyboard and fan noise from speech when they overlap. Clicks in gaps can be edited out. Clicks underneath a word stay.
- It will not make the result sound like your interface. The goal here is clear and listenable, not transparent. Aiming for the latter is how people over-process a take until it sounds worse than it started.
| What you hear | What it usually means | What to do | Realistic outcome |
|---|---|---|---|
| Noise swells in the pauses | Automatic gain control was on | Level the file last, gently | Consistent, but the original dynamics are gone |
| Dull, with a hard ceiling on the top end | Band-limited capture path | Clean the low end, skip the high shelf | Clearer, still dull. This is the ceiling |
| Watery or swirling background | Vendor noise suppression artefacts | Leave it alone or reduce very lightly | Usually no improvement available |
| Steady hiss, otherwise fine | Capsule and preamp noise floor | Light ReaFIR Subtract | Good. This is the most fixable case |
| Boxy, distant, roomy | Mic distance and room reflections | High pass, low-mid cut, accept the rest | Noticeably better, never close |
A speech recording from a laptop mic can usually be brought to a point where listeners follow it without effort and stop noticing the audio. That is a real result and it is often all that is needed. A sung vocal going into a music mix is a different standard, and a laptop take will not meet it. If the song matters, the repair is a scratch reference and the real answer is another take.
Stop it happening again
The underlying fault is that REAPER will happily record from whatever input is assigned, and nothing warns you that it is not the one you meant. Two habits close that gap. Before every session, arm the track and tap the actual microphone capsule you intend to use, then confirm that the meter moves on that tap and not on your voice alone. And set the input deliberately rather than leaving whatever was assigned last, which is covered in REAPER not detecting your microphone or interface.
If every recording you make has a hiss under it whether or not you used the laptop, that is a separate question with a separate answer, covered in why your recording hisses, and whether it is the mic or the interface.
A laptop take is not simply a worse microphone. It is a small noisy capsule, in the wrong place, behind processing you did not choose and cannot inspect from the DAW. Check for gain control, check where the top end stops, and check whether the noise is steady. Then high pass, cut the low mids, add presence only if there is content to add it to, treat steady noise lightly, and level it last. Speech comes back. A music vocal does not, and the honest move there is to record it again.
Sources
- Microsoft, “Audio Signal Processing Modes”, Windows drivers documentation - definitions of Raw, Default, Speech and Communications capture modes, the effect descriptions quoted for Noise Suppression, Automatic Gain Control and Deep Noise Suppression, the statement that OEMs define which effects are used per mode, the statement that applications cannot find out which mode is used for their streams, and the restriction that raw capture must not include adaptive processing. learn.microsoft.com
- Apple, “Use Mic Modes on your Mac”, macOS User Guide - the Standard, Voice Isolation and Wide Spectrum mic modes, the menu bar control, and the note that available options vary by app. support.apple.com
- Knowles, SPH0645LM4H-B datasheet - MEMS microphone signal to noise ratio of 65 dB(A) at 94 dB SPL and acoustic overload point of 120 dB SPL at 10% THD. Cited as representative of the class of part used in laptops, not as a measurement of any specific machine. Knowles datasheet
- Audio-Technica, AT2020 specifications - frequency response 20 to 20,000 Hz, signal to noise ratio 74 dB at 1 kHz at 1 Pa, noise 20 dB SPL printed without a stated weighting, maximum input SPL 144 dB SPL at 1 kHz at 1% T.H.D. Used as the entry-level studio condenser comparison. audio-technica.com
- Cockos, “The REAPER Cockos Effects Summary Guide” - ReaFir modes including Subtract, the instruction to enable
Automatically build noise profilewhile auditioning a noise-only section and to uncheck it once satisfied, the warning that raising the profile produces more audible artifacts, and the ReaEQ band shapes including High pass, Band and High shelf. reaper.fm/guides/ReaEffectsGuide.pdf - Cockos, “ReaPlugs” - ReaEQ described as an unlimited band IIR based equalizer and ReaFIR as an FFT based dynamics processor, both bundled with REAPER. reaper.fm/reaplugs
Every source above was retrieved and checked on October 11, 2026. No listening tests are claimed for this article, and no specific laptop model is identified as applying any specific processing, because that information is not published by the vendors.