StemMIDI

AUDIO → MIDI

MP3 to
MIDI

Turn a piano take, a hummed tune, or a solo recording into editable MIDI. Your audio stays on your device.

01 / SOURCE

Add your audio

MP3, WAV, M4A · Up to 60 sec / 25 MB

Audio stays on your device.

02 / RESULT

Piano roll

ILLUSTRATION
Illustration · Convert audio to see your notes

From recording to MIDI

01

Pick a melody

Choose a short, clear recording of one instrument or voice.

02

Find the notes

Convert the recording and check the notes in the piano roll.

03

Make it yours

Download the MIDI and edit it in your music software.

Good to know

What kind of audio works best?

A clear recording of one instrument or voice, including solo piano with chords. Full mixes and drums can produce stray or missing notes.

Is my recording uploaded?

No. The browser decodes the audio, detects notes and creates the MIDI file on your device.

What will the MIDI file contain?

It contains the detected pitches, note timings and approximate velocities. The MIDI stores performance data, so you can choose a new instrument after importing it.

THE WORKFLOW

From a recording to an editable performance

An MP3 captures how a performance sounded. A MIDI file describes what notes were played and when. StemMIDI estimates that second layer from the first, so you can take a piano phrase, guitar idea, or hummed melody into a music project and change it. You might keep the shape of the phrase while moving it to a different key, try the same notes on a synth, or repair one awkward note without recording the entire phrase again. Think of the output as a first draft for editing. Your ears and the original recording are still the best references.

Choose a source that gives the notes room

Start with one foreground instrument or one clear voice. A solo piano recording can contain multiple notes at once; the converter looks for overlapping pitches and can place them together in the same MIDI track. A dry recording from a phone or keyboard can work well when the instrument is close to the microphone and the room is quiet. A heavily reverberant recording can leave one note ringing into the next, so the end of a note may be less clear. Background music, conversation, and the sound of objects hitting the desk can also produce misleading events.

Trim to the passage you intend to use. If you are trying to recover a four bar idea, leave a little space before the first attack and let the last note decay naturally. Long stretches of silence make the roll harder to read. Cuts through the middle of a sustained note can make its first attack ambiguous. For a longer take, split it at musical boundaries and leave a short overlap between excerpts, then use the original recording to place each imported clip on your timeline. Remove any duplicate notes where the excerpts meet.

Which files work and how long can they be?

Choose a local MP3, WAV, or M4A file that your browser can decode. Each clip can be up to 60 seconds and 25 MiB. WAV is a useful first choice when you can export it directly from your recorder or DAW. MP3 is convenient for small files, and M4A is common for phone recordings. Changing an extension does not convert the audio inside the file. If an M4A or WAV cannot be read, open the original in an editor and export a fresh PCM WAV. For a large WAV, export just the needed section before changing the file format.

File size and duration are checked separately. A small MP3 can still exceed one minute, and a short high resolution WAV can still exceed the size limit. Reducing the bitrate only helps with size. If you need to shorten a file, keep a separate copy of the original and export a new clip. Avoid applying aggressive noise reduction, pitch correction, or normalization until you have compared the result with the untouched audio. Those effects can sometimes erase quiet attacks or make bleed between instruments harder to distinguish.

Convert the recording and inspect the piano roll

Choose the file or drop it on the source panel, then select Convert to MIDI. The browser decodes the audio and runs the note detection model on your device. The first conversion needs to load the model files from the site; your recording stays in the browser. Processing time depends on the length of the recording and the device. The sample button is a quick way to see how a successful conversion looks. It creates a MIDI from a built in synthetic melody, so it does not require you to find an audio file first.

Read the roll from left to right. Horizontal position shows timing, and higher rows represent higher notes. A longer block means the note was estimated to last longer. Stacked blocks indicate notes estimated to overlap. Check the first few attacks against the original: does the phrase start where you expect, and does the rise and fall match what you hear? An occasional very short note far above the melody may be a harmonic mistaken for a note. If the shape looks right, download the .mid and do the fine edits in your DAW.

What the MIDI file contains

The file stores note numbers, onset times, note lengths, and approximate velocities in one standard MIDI track. Velocity is an estimate of how hard a note should be played; it may need adjustment, especially if the original recording was compressed or clipped. The export uses a fixed 120 BPM tempo to encode the estimated real time positions. It does not infer a tempo map, snap a performance to the grid, separate instruments into tracks, or preserve a sustain pedal performance as controller messages. Choose a sound in your DAW before judging how the exported notes play.

Keep the audio beside the MIDI during cleanup. Play the two tracks separately to compare the outline, then together at a low volume to spot obvious timing differences. If the MIDI region starts in the wrong place, move the whole region before editing individual notes. Once the first attack lines up, compare the last one. If the two recordings gradually drift apart, check the project's tempo and how the DAW imported the MIDI. Changing a project tempo after import can also change the playback speed of MIDI regions relative to audio.

Open the result in Logic Pro

Download the file and create or select a software instrument track in your Logic project. You can use File → Import → MIDI File and choose the download, or drag it from Finder into the Tracks area where the part should begin. Logic places an import relative to the playhead or drop location. If your goal is to replace the original sound while preserving the performance, load a simple piano or neutral synth first. Open the Piano Roll and compare a short stretch with the source before you choose a more elaborate patch.

Use the piano roll to remove stray notes, correct pitches, and adjust note ends. For example, if a left hand note in a piano passage ends too early, extend just that note until it feels right with the chord. If the right hand has natural push and pull, start by correcting only the notes that feel misplaced. Quantizing the entire region can remove a deliberate pause or make a rolled chord land as one block. See Apple’s guide to Standard MIDI files for import and placement options.

Open the result in Ableton Live

In Live, drag the downloaded MIDI file into a MIDI track, or use Create → Import MIDI File. In Arrangement View, check where the insert marker is before using the command. In Session View, select an empty clip slot on the intended track. Add an instrument to hear the clip; a track with MIDI notes and no instrument can look correct and still be silent. Put the original recording on a separate audio track. If Live asks to use the tempo embedded in the file, decide whether to keep the existing set tempo before comparing the two tracks.

Double click the clip to edit its notes. Try a short loop first: fix the downbeat, remove a wrong note, and see whether the line works in context before editing the rest. To reuse the part in another key, transpose the clip after cleaning it up. To use it as an arrangement sketch, duplicate the clip and vary one note or chord at a time. That is where a MIDI draft saves time: you can audition several arrangements without repeatedly opening the original recorder. The Live manual’s section on MIDI files covers both import views.

Why drums and full mixes need different tools

This converter looks for musical pitch. Many drum sounds contain a short burst of broad frequency energy rather than a sustained pitch. A snare may be crucial to the groove even when the model does not emit a stable note for it. If the model does emit an event around a kick or tom, that is still a pitch estimate; it is not a reliable mapping to a drum rack pad or General MIDI percussion key. Use a drum transcription or transient detection workflow when you need kick, snare, and hi hat events.

A mixed song brings another problem: more than one instrument can play the same pitch, and their sounds overlap. The exported file has one note track and cannot tell you which instrument played each event. If you already have stems from your own project, select the piano, lead, or vocal stem and convert it individually. If all you have is a full mix, try a short exposed passage where the part is clear, then use the result as a sketch. Compare it to the recording before relying on it for an arrangement or performance.

What is free, and is there a paid tier?

The converter and MIDI download are free to use without an account. There is no paid tier at present, so there is no upgrade for longer files, batch processing, or separate instrument tracks. The same file and duration limits apply to every conversion. If you want to process a longer session, cut the recording into smaller musical sections and align the results manually in your music project. If you need collaborative editing or a larger arrangement, bring the resulting MIDI files into the DAW you already use.

For a quick first pass, try a clean 15 to 30 second excerpt, check the piano roll, and import it into a project with a plain piano sound. Save both the untouched MIDI download and your edited region. Keeping those two versions makes it easy to undo an overcorrection or compare a later conversion using a cleaner recording. Once a short example works for your material, repeat the process on the rest of the take.