pulp-vision/Hush
A speech enhancement model built for voice AI that isolates one speaker from live audio containing background talkers. The 8 MB model runs on the CPU in real time, with a causal design at a 16 kHz sample rate, and targets phone agents, call-centre bots, voice assistants and transcription pipelines. It was trained on more than ten thousand hours of mixed audio and is published under Apache 2.0 for PyTorch and Python 3.9 and above.
Maintenance activity