HushMic brings DPDFNet-based real-time noise suppression to Linux via PipeWire

HushMic is a newly released open-source application that provides real-time microphone noise suppression for Linux systems. It creates a PipeWire virtual microphone that applications such as Discord, OBS Studio, and web browsers can select as their audio input, eliminating the need for per-application noise reduction implementations.

The software processes audio locally using CPU-based inference with the DPDFNet speech enhancement model, targeting background sounds like keyboard noise, computer fans, and surrounding conversations while preserving speech clarity. It offers three operating modes (noise suppression, bypass, and mute), adjustable suppression strength, a choice between quality and lightweight processing models, and includes system tray controls, global keyboard shortcuts, command-line automation, and optional headless operation. A built-in microphone testing facility allows users to compare original and processed audio with short recordings for playback.

PipeWire integration and processing architecture

• DPDFNet speech enhancement model: HushMic uses this specific deep learning model for noise suppression, distinguishing it from alternatives based on RNNoise or DeepFilterNet that are already available in the Linux audio ecosystem.
• PipeWire virtual microphone: The application creates a virtual input device that persists across PipeWire restarts and system suspensions, allowing any PipeWire-compatible application to consume processed audio without native noise suppression support.
• Dual inference engine: Processing uses a native engine on supported processors with an ONNX Runtime fallback, keeping all computation local and avoiding cloud dependencies.
• LADSPA plugin availability: HushMic provides a LADSPA plugin for integration with Easy Effects and other compatible audio processing chains, enabling use within existing Linux audio workflows.
• Rust implementation with dual licensing: The application is written in Rust and available under either MIT License or Apache License 2.0, matching common open-source Rust project conventions.

Why it matters

HushMic fills a gap in the Linux desktop audio stack by providing a standalone, locally-processed noise suppression solution that integrates at the PipeWire level rather than requiring per-application support. For streamers, remote workers, and Linux gamers using voice chat, this offers an alternative to NoiseTorch-ng and noise-suppression-for-voice with a different underlying model architecture, potentially improving suppression quality for specific noise profiles while maintaining the privacy benefits of local processing.