Audio2GMGS MIDI PS6
A multitrack audio to GM/GS MIDI converter that combines source separation, tone and rhythm recognition, instrument mapping and export of individual MIDI tracks and a common multitrack mix.
What the project solves
A multitrack audio to GM/GS MIDI converter that combines source separation, tone and rhythm recognition, instrument mapping and export of individual MIDI tracks and a common multitrack mix.
Typical use
- Transfer WAV / MP3 / FLAC and other audio formats to the editing MIDI
- Separation of mixed audio on drums, bass, vocals, guitar, piano and other ingredients
- Export of one MIDI track for each 100 and common MIDI multitrack
- Preparation of MIDI for DAW, GM / GS synthesizers and other arrangements
Main characteristics
- Up to 16 input straws
- Per- st quantization OFF / ADAPTIVE / MANUAL
- GM / GS Program + Bank MSB / LSB mapping
- Separate treatment of percussion on MIDI channel 10
- CPU and accelerated backend inference according to available runtime
- batch mode / batch workflow next to graphical interface
How the application works and what it is for
From audio to editorial arrangements
Audio 2GMGS MIDI PS6 is not a simple "audio to MIDI" converter of one melodic line. The project aims to decompose the recording or already prepared stems on a musically meaningful part, analyze them according to their nature and create a MIDI that can be further modified in DAW or played over GM / GS compatible synthesis. Workflow therefore combines resource separation, pitch / onset analysis, rhythmic processing, heuristic tool recognition, choice of MIDI channel and program and subsequent export.
The project counts on both mixed entry and manual tracking. Mixed audio can be separated by Demucks into basic stems drums, bass, vocals, guitar, piano and other; the user can also work with his own input points files and dynamically add tracks up to the practical limit of 16 inputs.
Analytical backends and specialization by track type
Melodic and polyphonic tracks can be used by Basic Pitch and ONNX runtime according to the available profile. Drums have different requirements than vocals, bass or piano, therefore one universal heuristics is not used in the project for all sources. Drum workflow uses its own percussion mapping and the resulting MIDI percussion track is directed to standard GM channel 10. Melodic tracks are solved by pitch tracking, length of notes, polyphony, dynamics and selection of suitable GM / GS program.
Runtime contains NumPy / SciPy / Librosa, soundfile, mido, pretty-midi, mir _ eval, remampy, ONNX Runtime, Basic Pitch and Demucs. The computer backend is designed to run on CPU and use accelerated inference variants according to available runtime.
graphical interface, batch mode and batch workflow
The graphical interface desktop is used for interactive project preparation, selection of inputs, parameters of the stems, tool mapping and export. Next to it is calculated batch mode / batch mode to convert multiple files without manual clicking and the same settings to be used again in automated workflow. The format layer is built to receive common audio formats such as WAV, MP3, FLAC, OGG, OPUS, M4A / AAC, AIFF / AIF or WMA according to the available decoding backend.