Skip to content
SPECIFICTION
Get listed

BACH

github.com

Self-hosted
Visit github.com
Spec score 6.8 #121 of 123 in AI music generators

Spec sheet

Product
BACH is described as a human-editable, bar-level symbolic song generator.Read 12 Oct 2026 · github.com
Intended users
The README presents BACH for human-controllable long-song generation using bar-level symbolic notation.Read 12 Oct 2026 · github.com
Architecture
The architecture uses a BACH-1B decoder-only Transformer to produce ABC scores, with Dual-NTP and Chain-of-Score components.Read 5 Oct 2026 · github.com
Control
Its Dual-NTP component jointly predicts vocal and accompaniment patches at each step.Read 12 Oct 2026 · github.com
Evaluation dependency
The README says pitch extraction requires librosa, RMVPE, and its model checkpoint.Read 5 Oct 2026 · github.com
Evaluation requirements
Pitch extraction requires librosa, RMVPE, and its model checkpoint.Read 12 Oct 2026 · github.com
Generation flow
The README says an LLM writes lyrics, a Transformer emits an ABC score, and off-the-shelf renderers produce music.Read 12 Oct 2026 · github.com
Maker affiliation
The repository owner's profile identifies Tongxi Wang as affiliated with the School of Future Technology at Southeast University.Read 12 Oct 2026 · github.com
Model
The project describes BACH-1B as a decoder-only Transformer with 1 billion parameters.Read 12 Oct 2026 · github.com
Model size
The README describes BACH as a 1B-parameter model.Read 5 Oct 2026 · github.com
Music representation
The model uses non-overlapping 16-character patches per bar and outputs an ABC score.Read 12 Oct 2026 · github.com
Notable limit
The preview is not a complete CPU-ready generation pipeline, and required codec modules and assets are absent from the repository.Read 12 Oct 2026 · github.com
Paper
The repository says the related paper was accepted to ICASSP 2026.Read 12 Oct 2026 · github.com
Preview contents
The preview repository contains inference code and an example song, along with architecture figures.Read 5 Oct 2026 · github.com
Release limitation
The README says the preview is incomplete and not a CPU-ready generation pipeline; required codec modules and assets are not included.Read 5 Oct 2026 · github.com
Release status
The repository calls itself a preview release containing partial inference, preprocessing, and evaluation utilities.Read 12 Oct 2026 · github.com
Rendering tools
The architecture description names MIDI, FluidSynth, and VOCALOID in the rendering and mixing flow.Read 12 Oct 2026 · github.com
Score structure
The README says the model predicts vocal and accompaniment patches jointly and uses section tags for long-form coherence.Read 5 Oct 2026 · github.com
Security guidance
The README advises using trusted model assets and says inference uses PyTorch's restricted weights-only checkpoint loader rather than automatically loading legacy pickled model objects.Read 12 Oct 2026 · github.com
Song structure
Chain-of-Score uses section tags such as chorus start and end markers to support long-form coherence.Read 12 Oct 2026 · github.com
Vocal generation
YesRead 22 Sep 2026 · github.com

Where Specifiction places it

Alternatives to BACHAll alternatives →