DeepFilterNet Plus introduces real-time noise suppression for Linux and multi-platform WAV cleaning

DeepFilterNet Plus has launched as a maintained fork of the DeepFilterNet framework, specifically optimized for high-reliability speech enhancement. The software utilizes neural network-based deep filtering to remove background noise from full-band audio at a 48 kHz sample rate.

On Linux systems, the tool integrates into audio workflows via a LADSPA plugin, making it compatible with PipeWire and Easy Effects for real-time microphone suppression. Additionally, it provides a cross-platform command-line utility for cleaning recorded WAV files on Linux, macOS, and Windows, supported by pretrained models and a codebase written in Rust and Python.

Technical improvements and audio processing architecture

• Real-time Safety: The implementation utilizes real-time safe LADSPA processing designed to avoid blocking the audio callback, preventing system hangs during processing.
• Overload Protection: A specialized mechanism allows audio to pass through unprocessed if the processing worker cannot maintain pace with the incoming signal, preventing audio glitches.
• Audio Fidelity: The framework supports full-band audio processing at 48 kHz and includes improved stereo inference and multi-channel audio model handling.
• Integration and Compatibility: It maintains existing DeepFilterNet plugin identifiers to ensure compatibility with existing PipeWire filter-chain configurations.
• Development Stack: The project is built using Rust for core performance and includes Python components for model development and research workflows.

Why it matters

By providing a LADSPA plugin, DeepFilterNet Plus offers Linux users a high-quality, open-source alternative for real-time noise cancellation within the PipeWire ecosystem. Its ability to handle audio overloads without blocking makes it a robust choice for professional audio setups and streaming on Linux-based workstations.