Skip to content

v0.2.0 — Ensembles, restoration, transcription, audio editor

Choose a tag to compare

View all tags
@ericcayers-ai ericcayers-ai released this 12 Jul 08:21
· 41 commits to main since this release
v0.2.0
b1e6c02

A large expansion across every phase of roadmap.md, plus a full repository-health pass. Everything is tested (74 tests, ~84% coverage) and runs with no model downloads.

Try it

pip install -e .
neiro ui                                 # local interface (127.0.0.1)
neiro separate song.wav --preset vocals-ensemble
neiro transcribe song.wav --out song.mid
neiro enhance old.wav                    # auto-repair from analysis

New in this release

Separation — ensemble engine with weighted complex-spectrogram fusion (mean/median/max/min) and test-time augmentation; ensembles are themselves manifests. New vocals-ensemble preset.

Restoration — de-clip, mains-hum removal, spectral-gate denoise, normalize — applied automatically as conditioning chains from detected conditions, or on demand via neiro enhance.

Transcription — audio → MIDI with YIN pitch tracking, a timeline compiler doing reversible groove-preserving quantization (grid for notation, micro-offsets for feel), and a dependency-free Standard MIDI File writer. Auto-splits dense mixes first. Optional polyphonic Basic Pitch backend.

Audio editor — a basic waveform + spectrogram editor (trim, delete, silence, fade, gain, normalize, reverse) with non-destructive undo, in the browser.

Local interface — drag-and-drop Simple mode on 127.0.0.1: analysis card, separation with a stem mixer, transcription with a synced piano-roll, restoration, and the editor. Path-traversal protected; nothing leaves the machine.

Analysis — added mains-hum (50/60 Hz) and echo/delay detection.

Repository health

  • CI across Ubuntu/macOS/Windows × Python 3.10–3.12 with lint, format, and coverage; tag-triggered release workflow.
  • Ruff lint + format, coverage config, py.typed.
  • Docs: architecture, adding models, performance + a runnable benchmark.
  • CONTRIBUTING, CODE_OF_CONDUCT, SECURITY, issue/PR templates, Dependabot.

Licensing

Engine is MIT. Model weights carry their own licenses — each manifest records its terms and neiro models shows them.

Full changelog: CHANGELOG.md

🤖 Generated with Claude Code

Full Changelog: v0.1.0...v0.2.0

Assets 4
Loading