Fixes #421 and #293

This release adds a few more python wheels to PyPI

• Add python 3.11 support
• Add MacOS aarch64
• Add Linux aarch64

• Fix a bug where the ladspa plugin results in a reverberant output when running with attenuation limit

Minor release with improved logging.

Bugfix release

• Better ladspa defaults for consistent noise suppression
• Revert DNN inference framework to previous version due to audible artifacts

• Many bugfixes
• Improved DeepFilterNet3 model
• Multi-Frame Filtering (MVDR/Wiener Filter)

• New DeepFilterNet3 model with a slightly improved noise suppression performance.
• Fix a bug, where the native implementation produced a slightly worse noise suppression compared to the pytorch implementation.
• Add attenuation limiting for the LADSPA plugin.

• Fix a regression that caused a larger memory consumption
• Related to above: Update onnx models for deep-filter Rust implementation and improve inference speed

Rust-only version of DeepFilterNet implemented as a real-time loop.

This makes processing of large files relatively easy. Usage:

$ deep-filter -h
deep_filter 0.3.0
Hendrik Schröter
Noise supression using deep filtering
USAGE:
    deep-filter [OPTIONS] [FILES]...
ARGS:
    <FILES>...
OPTIONS:
    -D, --compensate-delay
            Compensate delay of STFT and model lookahead
    -h, --help
            Print help information
    -m, --model <MODEL>
            Path to model tar.gz
    -o, --out-dir <OUT_DIR>
            [default: out]
        --pf
            Enable postfilter
        --reduce-mask <REDUCE_MASK>
            If used with multiple channels, reduce the mask with max (1) or mean (2) [default: 1]
    -v, --verbose
            Logging verbosity
    -V, --version
            Print version information

LADSPA plugin for real-time noise reduction

E.g. for usage in a Pipewire filter chain. See readme.

Fix Windows PyPi package