Analysis updated 2026-05-18
Split a song into separate vocal and instrumental tracks locally.
Transcribe a sung or hummed melody into standard or Chinese numbered notation.
Correct an off-key vocal recording to match a reference melody in your own voice.
Translate sheet music between Chinese numbered notation and Western notation.
| cohenjikan/musicmaster | 410979729/scope-recall | arahim3/mlx-dspark | |
|---|---|---|---|
| Stars | 33 | 33 | 33 |
| Language | Python | Python | Python |
| Setup difficulty | hard | moderate | easy |
| Complexity | 4/5 | 3/5 | 3/5 |
| Audience | general | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Vocal separation and pitch reshaping need a GPU with at least 8 GB VRAM, transcription alone can run on CPU.
MusicMaster (声谱坊) is a local music processing tool that runs as a web server on your own machine, accessible from a browser at localhost. It bundles four distinct capabilities into one interface: splitting a mixed recording into separate vocal and instrumental tracks, transcribing a vocal melody into sheet music notation, translating between Chinese numbered musical notation and standard Western sheet music formats, and correcting the pitch of an off-key vocal recording while preserving the original voice's character. The vocal separation pipeline works in three stages: it first splits the full mix into vocals and accompaniment, then removes background harmonies to isolate the lead vocal, and finally runs a noise-reduction pass to produce a clean result. These three models run in sequence automatically. The transcription feature offers several selectable engines: a deep audio pitch tracker called CREPE works best for sung or hummed single-note melody and outputs both standard notation and Chinese numbered notation with a per-note confidence score, a general-purpose detector handles chords and instruments, and a specialized model covers clean piano recordings. The pitch correction and timbre reshaping feature takes three inputs: your original vocal recording (off-key is fine), a clean reference recording of the melody you want to match, and a short sample of your own voice as a timbre anchor. The tool first corrects the pitch to align with the reference, then maps the result back to your vocal timbre using a model called Seed-VC, so the output sounds like you singing in tune rather than like someone else. Most features run on your own hardware with no external API calls. The separation and reshaping tasks require a GPU with at least 8 GB of video memory, transcription and notation translation run on CPU only. Several of the integrated model weights carry non-commercial licenses, so commercial use requires checking each component's terms separately. The tool itself is released under Apache-2.0.
A local web app that separates vocals from music, transcribes melodies to sheet music, and corrects off-key vocal pitch.
Mainly Python. The stack also includes Python, CREPE, Seed-VC.
The tool code is Apache-2.0, free to use and modify, but several bundled model weights carry non-commercial licenses that must be checked separately before commercial use.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly general.
This repo across BitVibe Labs
Verify against the repo before relying on details.