audioreconstructor 1.1.0

Command-line audio reconstruction powered by the Audioreconstruction ONNX model

audioreconstructor

audioreconstructor is the command-line release of Audioreconstruction's ONNX audio enhancer. It installs a small Python launcher; the ONNX model and the native runtime are downloaded only when you explicitly run setup.

Install

pip install audioreconstructor
audioreconstructor setup

setup downloads the native executable, model.onnx, and config.json from the GitHub Release matching the installed package version. It verifies SHA-256 hashes before making them available locally.

The current release supports 64-bit Linux and 64-bit Windows.

Use

Enhance a single file:

audioreconstructor enhance --input song.mp3 --output song_enhanced.flac

Enhance a whole folder (recurses into subfolders):

audioreconstructor enhance --folder ./songs

Reconstructed files are written to a mirrored tree under <folder>/enhanced/, e.g. songs/rock/track.mp3 → songs/enhanced/rock/track.flac.

Choose the ONNX execution provider when needed:

audioreconstructor enhance --input song.mp3 --output song_enhanced.flac --provider cpu

auto is the default. Windows first tries DirectML and falls back to CPU; Linux uses CPU. The native executable reads supported audio through libsndfile and always writes FLAC output.

Note: v1.0.0 used top-level flags (audioreconstructor --setup, audioreconstructor --input ... --output ...). These are now setup, doctor, and enhance subcommands. Run audioreconstructor --help for the full command list.

Verify installation

audioreconstructor doctor

Doctor verifies all cached release assets and runs an actual synthetic-audio ONNX inference test. A healthy installation exits with status 0.

Cache locations

Setup prints the exact locations it uses. By default they are:

• Linux: $XDG_CACHE_HOME/audioreconstructor/<version> or ~/.cache/audioreconstructor/<version>
• Windows: %LOCALAPPDATA%\\audioreconstructor\\Cache\\<version>

Running setup for an upgraded package version removes older cached versions after the new version has been fully verified.