deepfilternet-rs 0.1.0

PyO3 bindings for the official DeepFilterNet Rust realtime runtime

DeepFilterNet-rs

Python bindings for the official DeepFilterNet Rust realtime runtime.

This package exposes a small PyO3 wrapper around DeepFilterNet's Rust DfTract streaming runtime. It is intended for realtime audio enhancement pipelines that need a Python API without shelling out to the deep-filter binary.

Install

pip install deepfilternet-rs

When installing from source, a Rust toolchain is required because the package is built with maturin. Prebuilt wheels do not require Rust on the target machine.

Usage

import numpy as np
from deepfilternet_rs import DeepFilterNetRealtime

processor = DeepFilterNetRealtime(
    model_path=None,
    atten_lim=100.0,
    log_level="warn",
    compensate_delay=True,
    post_filter_beta=0.0,
)

audio = np.zeros(processor.frame_length, dtype=np.float32)
enhanced = processor.process_chunk(audio)
tail = processor.finalize()

API

DeepFilterNetRealtime

Constructor arguments:

Argument	Type	Default	Description
model_path	`str	None`	None
atten_lim	float	100.0	Attenuation limit in dB. 100.0 means no explicit limit.
log_level	`str	None`	None
compensate_delay	bool	True	Drop initial algorithmic-delay samples from output.
post_filter_beta	float	0.0	Post-filter beta. 0.0 disables the post-filter.

Properties:

Property	Description
sample_rate	Backend processing sample rate. Official DeepFilterNet models use 48000 Hz.
frame_length	Frame length in samples. Official DeepFilterNet models use 480 samples.

Methods:

Method	Description
process_chunk(audio)	Process a one-dimensional float32 numpy array and return enhanced float32 samples. The input can be any length; incomplete frames are buffered.
finalize()	Flush buffered samples with zero padding and close the processor.
close()	Clear buffers and close the processor without flushing.

Release

Publishing is handled by GitHub Actions with PyPI Trusted Publishing. Create a GitHub release or run the release workflow manually after configuring the PyPI project trust relationship for this repository.