Oriel — AI & technology worth noticing, every day.

nvidia/NVIDIA-NemotronLabs-RE-USE

AI model · https://huggingface.co/nvidia/NVIDIA-NemotronLabs-RE-USE

First seen by Oriel on .

AI summary

NVIDIA NemotronLabs RE-USE is a commercially usable 9.6M-parameter multilingual model that restores degraded audio across multiple sampling rates while preserving speech fidelity, speaker identity, and intelligibility.

  • It accepts mono .wav audio from 8–48 kHz and handles noise, reverberation, clipping, bandwidth limits, codec artifacts, packet loss, and low-quality microphones.
  • On the full URGENT 2025 subset, it leads listed models in NISQA, PESQ, STOI, speaker similarity, and character accuracy, despite using only 9.6M parameters.
  • Training used roughly 2,500 hours of synthetic, single-speaker multilingual audio spanning 8–48 kHz, plus noise, room impulse responses, and microphone impulse responses.
  • The architecture combines convolutional encoding and decoding with 30 bidirectional Mamba layers; inference targets Linux on NVIDIA Ampere A100 with Triton acceleration.
  • Overlapping speech is processed by preserving the main speaker and suppressing others; NVIDIA recommends NVIDIA-NemotronLabs-RE-USE-laughter when laughter preservation matters.

#speech enhancement #mamba #audio processing #multilingual ai

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Models: 212, latest observation ; +88 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Models: 36,607, latest observation ; +7,332 over observed period since

Aliases

SourceExternal IDURL
Hugging Face Modelsnvidia/NVIDIA-NemotronLabs-RE-USEhttps://huggingface.co/nvidia/NVIDIA-NemotronLabs-RE-USE Read discussion