156 likes
Tiny NVIDIA model cleans speech in any language
What is NVIDIA-NemotronLabs-RE-USE?
It restores speech degraded by noise, reverberation, clipping, bandwidth limits, codecs, packet loss, or cheap microphones, while keeping the speaker’s natural tone. The 9.6-million-parameter model works on 8-48 kHz audio and supports many languages.
Who it is for
Voice-app developers, call-center tools, and podcasters can improve audio quality without language-specific tuning.