Skip to content

A2M v2.0.0

Choose a tag to compare

View all tags
@Justagwas Justagwas released this 16 Feb 17:42
· 18 commits to main since this release
v2.0.0
247c2ff
This commit was signed with the committer’s verified signature.
Justagwas Justin
GPG key ID: B9AB78ABCC9F4EE6
Verified
Learn about vigilant mode.

Release Notes for Version: 1.1.0 → 2.0.0

Included Files

  • A2MSetup.exe (Recommended)
    Windows installer for A2M. Size: 99.6 MB
    SHA-256: 4fb7297ca499fa1b912b11e5cb34740b275c3a3b10b3d4e63772bbb0e1269bd5

  • A2M.exe
    Standalone A2M application. Size: 145 MB
    SHA-256: 67ec25596f38a7998d385882863544836e055aaf89b86077fb33d8822f05808d

  • Source code (zip / tar.gz)
    Compressed A2M source code.

Security Notice

Only download from these sources: Official repo ; Sourceforge ; Website.

Changes

Major

  • Migrated runtime inference to ONNX Runtime.
  • Switched model artifact to PianoModel.onnx from https://downloads.justagwas.com/a2m/PianoModel.onnx.
  • Removed legacy PyTorch runtime flow and .pth runtime fallback paths.
  • CPU runtime is now the default shipped mode.
  • Added optional in-app GPU runtime pack installation for CUDA (a2m-onnx-cuda.zip) and DirectML (a2m-onnx-dml.zip).
  • Added a Transcription Engine selector in settings:
    • Legacy v1.0.0
    • Modern v2.0.0
  • Legacy is the default engine; Modern keeps dedicated tuning behavior.

Minor

  • GPU runtime packs are installed per-user under %LOCALAPPDATA%\A2M\runtime_packs\{cuda|dml}.
  • App config now writes to %LOCALAPPDATA%\A2M\a2m_config.json.
  • Added stricter GPU runtime validation to avoid accepting partial/corrupt installs.
  • Fixed ONNX runtime probe stability issues related to repeated PATH/DLL handling.
  • Unified runtime artifact/provider detection across runtime services to reduce mismatch edge cases.
  • Unified GPU validation logic across settings status, GPU toggle, provider switching, and conversion startup.
  • Added explicit GPU status state for unvalidated runtime (Validation pending) to avoid false "GPU active" messaging.
  • Startup now surfaces ONNX/GPU validation progress in the main UI and clears the validation prompt automatically when checks finish.
  • Improved fallback handling so ONNX-runtime-missing flows are handled separately from CUDA dependency failures.
  • CUDA install guidance/prompts are now shown only for CUDA-specific failure reasons.
  • cuDNN install flow is now asynchronous (non-blocking UI) with progress + cancel handling.
  • Fixed frame trimming logic for ONNX output deframing.
  • Continued MIDI post-processing cleanup for short/overlapping note events.
  • GPU mode now prompts users to install required runtime components when unavailable.
  • Provider selection (Auto, CUDA, DirectML) now respects availability state.
  • Refined settings/runtime status behavior and related stale messages.
  • Modern engine feature toggles include:
    • Adaptive thresholds per file
    • Input normalization / denoise
    • Smarter overlap stitching
    • Smart auto calibration
  • Modern controls now hide/show contextually based on selected engine and calibration mode.
  • Added/updated ONNX/GPU runtime-related config fields.

Full Changelog: v1.1.0...v2.0.0

Assets 4
Loading