Audio-to-Audio
MambaSSM
Safetensors
streaming speech-enhancement
speech-enhancement
universal speech enhancement
multiple input sampling rates
language-agnostic
Instructions to use nvidia/Real-time_RE-USE with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MambaSSM
How to use nvidia/Real-time_RE-USE with MambaSSM:
from mamba_ssm import MambaLMHeadModel model = MambaLMHeadModel.from_pretrained("nvidia/Real-time_RE-USE") - Notebooks
- Google Colab
- Kaggle
| CUDA_VISIBLE_DEVICES='0' python offline_inference.py \ | |
| --input_folder ./noisy_audio \ | |
| --output_folder ./offline_enhanced_audio \ | |
| --config recipes/USEMamba_12x1_lr_00002_norm_05_vq_067_nfft_320_hop_160_NRIR_012_pha_0005_com_04_early_005_release_random_layer_GAN_longer_1k.yaml \ | |
| --Exit_layer 8 \ | |
| --look_ahead_frames 0 \ | |
| #--BWE 16000 \ | |
| # Exit_layer can be between 3~12 , look_ahead_frames can be between 0~2 |