audio-clean-booster 0.1.0

Speech cleanup helpers using open-source enhancers plus reference-guided matching.

audio-clean-booster

Open-source speech cleanup helpers for denoising, reference matching, and chunked listening comparisons.

The package can:

• Run DeepFilterNet3 speech denoising.
• Run ClearVoice MossFormer2 speech enhancement.
• Match an enhanced file to a provided clean reference with local loudness, spectral shaping, and strict reference targeting.
• Build an HTML page with 10-second audio chunks for side-by-side comparison.

Install

Base install:

pip install audio-clean-booster

Install all model backends:

pip install "audio-clean-booster[all]"

For local development:

pip install -e ".[all]"

CLI

Run DeepFilterNet3:

acb deepfilter noisy_16k.wav noisy_16k_deepfilter.wav

Run MossFormer2:

acb mossformer noisy_16k.wav noisy_16k_mossformer2.wav

Create a strict reference-guided match:

acb reference-match noisy_16k_mossformer2.wav noisy_16k_clean_original.wav noisy_16k_final.wav --mode strict

Build a comparison page:

acb compare --source noisy:noisy_16k.wav --source mine:noisy_16k_clean_original.wav --source final:noisy_16k_final.wav

This writes compare_chunks.html and chunk WAV files under compare_chunks/.

Publishing

python -m build
python -m twine upload dist/*

Before publishing, update the Homepage and Repository URLs in pyproject.toml.