A multitrack audio editor that shows your recording as a spectrogram, so a cough under a voice or a chair scrape becomes a shape you can select and remove — leaving everything around it untouched. No account, no upload, no subscription.
Version 1.2.0 · macOS · Windows · Linux
This release is about not losing work. Everything below came from asking what happens when a session ends badly — the power goes, the app is closed by accident, a plug-in takes it down mid-edit.
A recovery copy is written every few seconds. If the app stops for any reason, the next launch offers your tracks, edits and audio back.
Closing with unsaved changes offers to save, to keep a recovery copy, or to go back to editing. A running recording refuses to be closed at all.
Recordings repair their own file once a second while they run, so losing the app costs you the last second rather than the whole take.
Anything recorded, generated or edited is written alongside the project as lossless audio, so reopening it gives back exactly what you heard.
Loop a selection or the whole arrangement, and work with time selection, zoom, hand scrolling and dragging several clips at once as one move.
The spectrogram is drawn at full resolution, lines up exactly with the waveform above it, and refreshes the moment you repair something.
Recording, editing, restoration and separation all run locally. Nothing is uploaded, which means client material never leaves the room it was recorded in.
Arrange a session across tracks, or open one file and work on it directly — trim, gain, normalise, fade, reverse, with undo that stores only what changed.
Sound drawn as time against frequency. Select a region and attenuate it, or heal it from the moments either side so no hole is left behind.
Adaptive noise reduction, DeReverb, Voice Enhance and de-click, each working from a noise floor estimated from the recording itself.
A real FFT spectrum analyser with selectable size and window, and a phase correlation meter and goniometer reading the master as it plays — so what is drawn is what is playing, not an approximation of it.
A phase vocoder changes length without changing pitch, or pitch without changing length, on any clip. Lock the pitch for a straight time stretch, or let it move with the speed for a varispeed effect.
Line a studio take up against the production dialogue it replaces. The two takes are measured against each other and the clip is moved onto the reference, as tightly as you ask for.
EQ and graphic EQ, compressor, brickwall limiter, gate, de-esser, a four-band multiband compressor, reverb, delay, chorus, flanger, phaser, stereo width, pitch correction and filters — each band and control reaching the engine.
Your own plug-ins load into any chain, each hosted in a separate process so a plug-in that crashes takes itself down and not your session.
Draw volume, pan or any effect parameter over time, with linear, hold and smooth curves between breakpoints.
ITU-R BS.1770-4 and EBU R128 metering with true peak from a 4×-oversampled signal, measured live off the master as it plays, and targets for the platforms you deliver to. Every channel strip meters its own contribution.
Import a video and its audio arrives as an ordinary clip, so every tool works on it while the picture follows the transport.
WAV, AIFF, FLAC and MP3 with real dither, batch jobs across a session, markers,
regions and a project archive format. Your title, artist and ISRC are written
into the file — ID3 for MP3, Vorbis comments for FLAC, and a broadcast WAV
bext chunk with originator and coding history.
Say what the audio is — dialogue, music, sound effects or ambience — and get processing suited to it, with a loudness target that matches where it is going.
Source separation runs offline on your own machine. The model is fetched once, from inside the app, with a button — no hunting for files and no dragging anything into a folder.
Separation uses Spleeter, whose weights are MIT-licensed and published separately from the app. The first time you open the separation dialog there is a Download button beside the model; press it once and the weights are verified against a checksum and kept for next time.
Spleeter models frequencies up to 11 kHz, so its stems have less sparkle at the top than the original. That is how the model was built, and the app says so before you spend the download.
Four steps from an untouched recording to a delivered file. The order matters: repair before you process, and set loudness last, once nothing downstream can change it again. Every part of the app, button by button, is in the full guide.
Press Import and pick a file — WAV, AIFF, FLAC, MP3, M4A — or Record straight from a microphone. Import a video and its sound arrives as an ordinary clip with the picture alongside it. If you are working through an interface, Hardware lists what the machine actually has and lets you set its clock rate and buffer size.
Switch the clip to the spectral view and the recording becomes a picture of itself. Drag a box around a cough, a door, a phone buzzing on the table, and choose Heal — the app rebuilds that patch from the moments either side. For steady noise across the whole file, use Adaptive Noise Reduction; for a room that rings, DeReverb.
Open Essential Sound, say what the audio is — dialogue, music, effects, ambience — and it applies treatment suited to it. Or build a chain by hand from the seventeen built-in effects, and drop in your own VST3 plug-ins beside them. Draw automation if something needs to move over time.
Watch the LUFS and true-peak meters against the target for wherever the file is going — −16 for Apple and Spotify, −23 for broadcast. Then Export to WAV, AIFF, FLAC or MP3, or queue a batch across the whole session.
Select the clip, open Demix AI, and press Download beside the model the first time only. When it finishes, press Separate Stems: the vocal and the accompaniment arrive as two new tracks, lined up with the original, editable like anything else.
If yours is not here, write to us and we will answer it — and add it.
Yes. Editing, restoration, effects, metering and export all run on your own machine and never contact a server. The one exception is the first source separation, which downloads its model; after that separation is offline too.
No. Nothing you open, record or export leaves your computer. There is no account, no telemetry and no cloud processing, which is the point — client material never has to leave the room it was recorded in.
Nothing at the moment. Audio Editor is free during its first release. Pricing will be announced before it changes, and anything you have made stays yours and stays openable.
macOS 12 or later on Apple Silicon, Windows 10 or 11, and 64-bit Linux. Intel Macs are not supported: the machine-learning runtime that source separation depends on publishes no build for them.
No. The app has not been through Apple's notarisation yet, and that is the
message macOS shows for any application it cannot check. Right-click the app,
choose Open, then Open again. If it still refuses, run
xattr -dr com.apple.quarantine "/Applications/Audio Editor.app"
in Terminal once.
The Spleeter model that separates vocals from backing. Its weights are MIT-licensed and published apart from the application, so the app fetches them once, checks them against a known checksum, and keeps them. Everything else works without it.
It splits a mix into a vocal and an accompaniment. The model only covers frequencies up to 11 kHz, so both stems come back with less air at the top than the original — audible on cymbals and sibilance. It is a two-stem split, not four; drums and bass are not separated from each other.
VST3 plug-ins, yes, on any chain, and each runs in its own process so one that crashes does not take your session with it. Audio Units are not supported. Plug-in editor windows currently open on macOS only.
It opens WAV, AIFF, FLAC, MP3, M4A and the audio inside common video files, and exports WAV, AIFF, FLAC and MP3 with real dither. Projects save as a single file, and a whole session can be packed into one archive.
It covers the same ground for spectral repair, restoration, effects and loudness, with no subscription and nothing uploaded. Audition has a deeper toolset, Audio Units support and a direct link into Premiere Pro, which this does not. There is no round-trip with Premiere.
In your own Application Support folder, not inside the application, so it survives updates and needs no administrator rights to install.
macOS on Apple Silicon, Windows 10 and 11, and 64-bit Linux.