DeepFilterNet is a low-complexity speech enhancement framework for full-band audio using deep filtering. The framework includes a pre-compiled binary with no Python dependencies, a real-time version, and a LADSPA plugin for noise reduction on microphones. DeepFilterNet also includes a Deep Multi-Frame Filtering paper and a DeepFilterNet2 paper for speech enhancement on embedded devices. The framework includes usage instructions and a Python wrapper, and it supports Linux, MacOS, and Windows. The training is tested only under Linux, and currently, only audio files with a sampling rate of 48kHz are supported. If used, users should cite DeepFilterNet or DeepFilterNet2 accordingly.
https://github.com/Rikorose/DeepFilterNet