Someone searching for Mirelo audio to MIDI is usually choosing between two different kinds of transcription tool. Mirelo is built to take a finished mix apart into labeled instrument tracks. Sound to MIDI runs a smaller model inside the browser and is better suited to one clear instrument, voice, or stem. Both produce editable notes, but the files they want and the results they emphasize are not the same.
Try Mirelo when you need separate tracks from a full song, MusicXML, score tools, or a hosted editor. Use a private browser converter when you have a clean stem or melody, do not want to upload the recording, and mainly need a standard MIDI draft to edit in a DAW.
What Mirelo audio to MIDI does
Mirelo’s official audio-to-MIDI page describes a model developed with Kyutai for full mixes. It detects instruments, labels the parts, and returns a separate MIDI track for each detected instrument. The workbench includes a piano roll, and Mirelo says users can solo, mute, edit, and export tracks. It also offers MusicXML output and musical context such as chords, key, and tempo.
That full-mix focus is the main reason to use it. A song containing voice, drums, bass, and keys asks a transcription system to solve several problems at once. Mirelo is designed for that case instead of asking you to separate a stem first. Its public model page also links to the open-source MuScriptor project, while the hosted Studio uses Mirelo’s own version trained on a larger dataset.
The Studio is an online service. You upload a recording and work with the result in a hosted project. Its current pricing page lists a free monthly credit allowance, project limits, and a watermark on free exports. Audio-to-MIDI usage in Studio consumes credits by input duration. Anyone planning to transcribe long songs should check that page before starting because allowances and rates can change.
Mirelo has been changing quickly. Its product changelog records support for formats including WAV, MP3, FLAC, OGG, AIFF, M4A, and WebM. It also documents score view, tempo and meter handling, per-instrument MIDI channels, quantized or performance timing, note editing, and MusicXML fixes. Those are meaningful features if the transcription is going into notation or a detailed multi-track arrangement.
What Sound to MIDI does differently
Sound to MIDI runs Spotify’s Basic Pitch model on the user’s device. The MP3 or WAV is decoded in the browser, the model predicts note events locally, and the page creates a standard .mid file. The recording is not sent to Sound to MIDI for transcription. There is no project account or upload queue for the free Raw MIDI path.
Spotify’s Basic Pitch repository describes the model as instrument-agnostic and polyphonic, with pitch-bend support. It can hear more than one note at once, which makes it useful for piano and chordal instruments. Spotify also says it works best on one instrument at a time. That last sentence matters more than the word “polyphonic.” A clean piano stem is a realistic input. A mastered song with drums, bass, vocals, and effects is much less predictable.
The browser tool includes a piano-roll preview and controls for onset confidence, note confidence, minimum note length, tempo, and pitch bends. Its optional cleanup modes can keep the melody, bass line, or piano chords and apply different timing strengths. This is a smaller workflow than Mirelo’s multi-instrument workbench, but it can be faster when the user already has the right stem.
Mirelo and Sound to MIDI compared
The Mirelo audio to MIDI workflow and the browser-local workflow solve different source problems, so the comparison starts with the recording rather than a feature count.
| Question | Mirelo | Sound to MIDI |
|---|---|---|
| Best starting material | A full mix or complete song | One instrument, vocal, or isolated stem |
| Processing location | Hosted Studio after upload | Current browser on the user’s device |
| Track separation | Labeled MIDI tracks for detected instruments | Melody, bass, or chord-preserving cleanup from one transcription |
| Main exports | MIDI tracks and MusicXML, plus Studio sharing options | Standard Raw MIDI, piano-roll image, and optional Cleaned MIDI |
| Editing environment | Hosted piano roll and score-oriented tools | Preview and basic cleanup before DAW editing |
| Account model | Studio account with plan and credit rules | No account for local conversion and Raw MIDI download |
| Strong reason to choose it | Full-song, multi-instrument transcription | Privacy and quick work on a clean source |
This is not a quality ranking. Each tool has an easier problem it is prepared to solve. Sending a dense full mix to a one-instrument model and calling the output bad is not a useful test. Nor is sending a confidential solo stem to a hosted service when local processing is the main requirement.
Choose Mirelo audio to MIDI for full songs and notation
Mirelo is the better match when the input cannot be separated before transcription. A producer may have only a stereo master but want independent drum, bass, vocal, and keyboard parts. A musician may want a MusicXML score rather than a MIDI sketch. A hosted editor also makes sense when the user wants to inspect instrument labels, correct notes, follow a score view, and export several tracks from one job.
Start with a short passage even then. Check whether the detected instruments match the arrangement and whether the timing mode suits the performance. A useful four-bar result is better evidence than processing a five-minute song and discovering at the end that the source or settings were wrong.
Choose browser-local conversion for stems and private work
A local converter fits a different routine. Export a vocal, piano, bass, or guitar stem from the DAW, open the WAV to MIDI converter, and test the notes without uploading the recording. This is useful for unreleased material, client stems, voice memos, and quick melodic ideas where privacy matters.
It also keeps the handoff simple. Download the MIDI, place it beside the original audio in the DAW, and correct the notes there. You do not need a browser editor to become the permanent home of the project. The transcription is a draft that moves into the software already used for arranging.
The local model does not label every instrument in a full mix. If you need that, separate the source before conversion or use a system designed for multi-instrument transcription. A stem splitter followed by local conversion can work well when you need only one or two parts and want the later transcription to stay on the device.
A fair way to test both tools
Use the same musical passage, but give each tool the input it expects. Keep the original full mix for Mirelo. For Sound to MIDI, export or separate the specific stem you want. Then compare the part you actually need rather than counting every note across unrelated tracks.
Listen for four things: missed note starts, extra short notes, sustained notes cut into pieces, and timing drift. Import both MIDI results into the same DAW at the same tempo. Mute the generated instrument and switch between it and the source audio. The result that needs less repair for your task is the useful one.
Automatic transcription is still transcription. Neither workflow can recover the original MIDI session from rendered audio. Instrument labels, tempo detection, and cleanup controls reduce manual work, but a final pass by ear remains part of the job.
Which one should you use?
Use Mirelo audio to MIDI when the full mix is the only source and separate instrument tracks or MusicXML are the goal. Use Sound to MIDI when the source is already isolated, a standard MIDI draft is enough, and keeping the recording in the browser matters.
If you are unsure, test one short section first. The source material will answer the question faster than a long feature list.
