Audio2MIDI is a free VST3 and AU plugin that transcribes full mixes into one MIDI part per instrument. No credits to buy, no monthly quota, no upload. The only limit is your hardware.
Put Audio2MIDI on an audio track. Drop a wav, aiff, flac, mp3 or ogg, or press Record and play the track.
02
Transcribe on your Mac
The MuScriptor engine runs on 127.0.0.1. It detects the instruments in the full mix and writes one MIDI part per instrument.
03
Drag MIDI into your DAW
Check the result in the piano roll and by ear, then drag All parts or a single instrument straight onto a track.
The whole workflow in one window.
Drop, transcribe, listen, drag. The parts list shows every detected instrument with its note count. Drag a row into your arrangement, or open the MIDI files with Show files.
Local by design
Your studio is the server. The meter never runs.
Cloud transcription bills by the second of audio. Audio2MIDI runs the model on your own computer, so a 3-minute demo and a 3-hour session cost the same: nothing.
Credit-based cloud plugin18,000credits per month
3.6× the free plan's 5,000 monthly credits
Audio2MIDI0credits, ever
Runs on your computer. Bring your own CPU or GPU.
Cloud figures use Mirelo's published rate of 2.5 credits per audio second and its free plan of 5,000 credits per month, checked November 9, 2026. Paid plans not shown. Rates may change.
No credits, no quota
No balance to top up and no monthly cap. Transcribe as long as your machine has power and disk.
Audio stays home
The plugin talks to the engine on 127.0.0.1. Unreleased tracks never leave your computer.
Works offline
After the first download of the engine and weights, transcription needs no internet connection.
No account to keep
No Audio2MIDI login. A free Hugging Face account unlocks the model weights once.
01Your DAW
audio
Audio2MIDI pluginJUCE · C++
http
MuScriptor engine127.0.0.1:8722
all on your computer
small
Fastest
About 0.4 GB extra download. Good for quick sketches.
medium · default
Balanced
About 2 GB with PyTorch on the first run.
large
Most accurate
About 5.5 GB. Slower, and needs more memory.
Built for the DAW
Everything you need between the bounce and the edit.
Per instrument
Full mixes in, parts out
Drop a finished song. Guitar, bass, drums, voice and more come back as separate MIDI files, plus one file with every part.
Tempo + meter
Lands on your grid
Every MIDI file carries the detected tempo and meter, so parts snap into the arrangement instead of drifting.
Quantize
Align to beats
Snap every note to the detected beat grid, or leave it off to keep the performance exactly as played.
Preview
Check it by ear
Preview through the track: one slider mixes the original audio with the MIDI. Click the piano roll to jump.
Recall
Saved with your project
The transcription is stored in the plugin state. Reopen the session and your parts are still there.
Record
Record straight from the track
No bounce needed. Press Record, play the track in your DAW, and transcribe the take.
Compare
Audio2MIDI vs. Mirelo.
Mirelo makes a great cloud plugin, and co-released the open MuScriptor model that Audio2MIDI runs. The difference is where the work happens and who pays for it.
Aspect
Audio2MIDI
Mirelo Audio-to-MIDI plugin
Cost
Free download and local transcription. No subscription, no credits.
No Audio2MIDI account. Free Hugging Face account for model access.
Mirelo account required
Plugin source
Open source, AGPL-3.0
Commercial product
DAW workflow
File drop or recording, full mixes, per-instrument and combined MIDI
File drop or recording, full mixes, per-instrument and combined MIDI
Model
Open MuScriptor research model
Audio-to-MIDI Pro production model
Checked November 9, 2026 against Mirelo's plugin announcement and FAQ, model overview and the MuScriptor model card. Free does not mean unrestricted: the MuScriptor weights are CC BY-NC 4.0 (non-commercial), and we don't claim the accuracy of Mirelo's production model.
Install
Up and running in a few minutes.
Install uv. The plugin uses it to set up an isolated Python 3.12 environment. uv install guide
Install the plugin. Open the .pkg. It is not notarized, so allow it once under System Settings → Privacy & Security → Open Anyway. No admin password needed.
Rescan your DAW and insert Audio2MIDI on an audio track. The first transcription downloads the engine and weights.
FAQ
Questions, answered.
Is it really free?
Yes. The plugin is free and open source under AGPL-3.0, and transcription runs on your own computer, so there are no credits, plans or per-song fees. You supply the hardware, disk space and electricity.
Can I use the results commercially?
The plugin code is AGPL-3.0. The MuScriptor model weights are licensed CC BY-NC 4.0, which is a non-commercial licence. Read the licence on Hugging Face and decide for your use case before you rely on it commercially.
Does my audio leave my computer?
No. The plugin talks to the engine on 127.0.0.1:8722, on your machine. The internet is only needed once, to download the Python environment and the model weights.
Is it as accurate as Mirelo's Audio-to-MIDI Pro?
We don't claim that. Audio2MIDI uses the open MuScriptor research model. Mirelo's cloud product uses its own production model. Try it on your material: it costs nothing.
Why do I need a Hugging Face account?
The model weights are gated. Accept the licence on the MuScriptor model page once, run uvx hf auth login, and the plugin downloads the weights on the first transcription.
Which DAWs does it work in?
Any macOS host that loads VST3 or Audio Units, such as Ableton Live, Logic Pro or FL Studio. There is also a standalone app. There is no AAX version, so Pro Tools is not supported.
What about Windows?
It builds from source but is untested. On Windows the engine runs PyTorch on the CPU, the MIDI preview is silent, and the engine can outlive a crashed DAW.
Why does macOS block the installer?
The release is not signed or notarized. Open System Settings, Privacy & Security, and click Open Anyway once. The installer needs no administrator password and installs into your user Library.
Stop paying per second. Start transcribing.
Free, local and open source. Your songs, your machine, your MIDI.
