Audio-to-Audio
MambaSSM
Safetensors
speech-enhancement
universal speech enhancement
multiple input sampling rates
language-agnostic
Instructions to use nvidia/RE-USE with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MambaSSM
How to use nvidia/RE-USE with MambaSSM:
from mamba_ssm import MambaLMHeadModel model = MambaLMHeadModel.from_pretrained("nvidia/RE-USE") - Notebooks
- Google Colab
- Kaggle
Download inference_chunk.sh from nvidia/RE-USE: direct link, hf CLI and curl.
- Browser
- Download file 473 Bytes
-
https://huggingface.co/nvidia/RE-USE/resolve/main/inference_chunk.sh
- Command line
-
hf download hf://nvidia/RE-USE/inference_chunk.sh
-
curl -L -o inference_chunk.sh https://huggingface.co/nvidia/RE-USE/resolve/main/inference_chunk.sh
473 Bytes
| CUDA_VISIBLE_DEVICES='0' python ./inference_chunk.py \ | |
| --input_folder ./long_noisy_audio \ | |
| --output_folder ./long_enhanced_audio \ | |
| --checkpoint_file ./exp/30x1_lr_00002_norm_05_vq_065_nfft_320_hop_40_NRIR_012_pha_0005_com_04_early_peak_GAN_tel_mic/g_01134000.pth \ | |
| --config ./recipes/USEMamba_30x1_lr_00002_norm_05_vq_065_nfft_320_hop_40_NRIR_012_pha_0005_com_04_early_001.yaml \ | |
| --chunk_size_in_seconds 5\ | |
| --hop_length_portion 0.5\ | |
| #--BWE 32000 \ | |