Music to MIDI

Music to MIDI

Converts an audio file into editable MIDI or separate WAV tracks, with seven modes including multi-instrument and piano transcription, for musicians and producers.

Made by mason369 on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Music to MIDI turns an MP3, WAV, FLAC, OGG or M4A file into editable MIDI, or splits it into separate WAV tracks. The Space, a desktop version and a Colab notebook share seven modes: SMART, VOCAL_SPLIT, SIX_STEM_SPLIT and four piano modes. The models behind it include MuScriptor, YourMT3+, MIROS and several piano transcribers.

Who it suits: Musicians, arrangers and producers who want a MIDI file from a recording so they can edit the notes.

Industry | Who uses it | How they use it | Example
Music production | Producers and arrangers | Turn a recording into MIDI they can edit | Pulling a melody out of a short clip
Piano teaching | Teachers and students | Transcribe a piano performance into notes | Getting a MIDI file from a short piano take
Audio editing | Sound editors | Split a track into vocals, accompaniment or six stems | Separating vocals from a backing track

If you need to convert full-length songs on the hosted page, you can skip it, because ZeroGPU accepts only clips a few seconds long.

Where to use it:

• A web browser, on the Space's page at Hugging Face, after you sign in
• Python 3.12.12 on Windows, macOS or Linux, for a copy you run yourself
• The desktop version or the Colab notebook, which the source says use the same modes

Use case | Who | What they get back
Direct transcription | Musicians | A MIDI file from one click
Stem separation | Sound editors | WAV tracks to work on one by one
Per-track transcription | Arrangers | A MIDI file for each separated track, with its own model choice

How to set it up: On the hosted page you only sign in. A copy on your own computer needs more.

• A Hugging Face account, because conversion on the hosted page needs a login
• Python 3.12.12 and the pip tool
• The packages in requirements.txt and the system packages in packages.txt
• For the MuScriptor routes, accepted terms on three gated repositories and a private secret named HF_TOKEN

1. Install the Hugging Face hub package.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio at the pinned version.
5. Install the packages from requirements.txt.
6. Start app.py.

pip install -U huggingface_hub
hf download mason369/music-to-midi --repo-type space --local-dir music-to-midi
cd music-to-midi
pip install gradio==6.17.3
pip install -r requirements.txt
python app.py

On a Space copy, save the token as a private secret in Settings, under Variables and secrets. A public variable, the repository or the logs are not supported places for it.

How to use it: Sign in, upload an audio file and pick a mode. The five direct transcription modes make a MIDI file with one click. The two separation modes make WAV tracks and open a multi-track workbench, where each track has its own models and its own start button.

python app.py

Tool | What it does
SMART | Transcribes several instruments into MIDI, with an automatic or chosen instrument list
VOCAL_SPLIT | Separates a vocals WAV and an accompaniment WAV
SIX_STEM_SPLIT | Separates the audio into six WAV tracks
Piano modes | Transcribe piano audio straight to MIDI with one of four piano models
Multi-track workbench | Offers thirteen transcription routes per track and a separate start button

Example: You sign in, upload a short MP3 of a piano melody, pick a piano mode and click once. You get a MIDI file whose notes you can edit. Clips longer than 2.50 seconds are refused on ZeroGPU, so use a copy with its own GPU or Colab for more.

For newcomers and experts: MIDI is a file of notes rather than sound, so you can change a note, a tempo or an instrument after the conversion. To start, upload a few seconds of solo piano and use one of the piano modes.

SMART uses MuScriptor with 5 second windows and prelude forcing. A Beat This grid corrects onsets and sets tempo and bar phase. An empty instrument choice means automatic detection, and a filled one becomes a generation constraint. The default route is YourMT3+, and MIROS with MusicFM is another route. The ZeroGPU limits are admission estimates, not measured run times.

Sign in and upload audio
Pick a direct or a separation mode
The app estimates GPU use and checks the limit
Model runs, then you edit the MIDI or WAV

Good to know:

• Limits: ZeroGPU suits short clips, and the app refuses a job whose estimate passes the 300 GPU second window of a free signed-in account
• License: MIT covers only the Space's own app code, while third-party code, checkpoints, SoundFonts and datasets keep their own licenses
• Cost to run: a free Hugging Face account gets 5 minutes of ZeroGPU a day, and a copy needs an HF_TOKEN secret with access to the gated MuScriptor repositories
• Data it reaches: the audio you upload, models and datasets pulled from the Hugging Face Hub, and the HF_TOKEN secret on a copy
• Not for: long audio on ZeroGPU, or commercial use of the MuScriptor weights, which are gated and licensed CC BY-NC 4.0

Title: Longest audio the ZeroGPU admission check accepts
SMART default: 2.00 s
SMART with MIROS: 1.00 s
SMART with MuScriptor Large: 0.833 s
SMART with MuScriptor Medium: 2.083 s
SMART with MuScriptor Small: 3.571 s
VOCAL_SPLIT: 0.666 s
SIX_STEM_SPLIT: 0.277 s
Piano direct modes: 2.50 s

On Hugging Face, 13 November 2026

• Likes: 9
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/mason369/music-to-midi
• Files: https://huggingface.co/spaces/mason369/music-to-midi/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.