NVIDIA Releases NemotronLabs RE-USE Universal Speech Enhancement Model
NVIDIA has released NemotronLabs RE-USE, a 9.6-million-parameter multilingual universal speech enhancement model designed to restore degraded audio while preserving speaker fidelity. The model is ready for commercial use and supports multiple input sampling rates and diverse audio degradations. This release provides a lightweight, highly capable, and commercially ready tool for cleaning noisy or distorted audio across multiple languages. It benefits developers and creators working on local AI, speech processing, and audio restoration applications. The model handles additive noise, reverberation, clipping, bandwidth limitation, codec artifacts, packet loss, and low-quality microphones across sampling rates from 8 kHz to 48 kHz. A specialized variant, NVIDIA-NemotronLabs-RE-USE-laughter, is available for better laughter preservation, and the model targets main speaker preservation during overlapping speech.
## BACKGROUND
Speech enhancement models are machine learning systems trained to remove background noise, reverberation, and distortions from recorded audio to improve clarity and intelligibility. Codec artifacts refer to unwanted distortions or degradation in sound quality that occur during the compression and decompression process of audio data using a codec.