Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| from speechbrain.pretrained import SepformerSeparation as separator | |
| import torchaudio | |
| # model = separator.from_hparams(source="speechbrain/sepformer-wham-enhancement", savedir='pretrained_models/sepformer-wham-enhancement') | |
| model = separator.from_hparams(source="speechbrain/sepformer-dns4-16k-enhancement", savedir='pretrained_models/sepformer-dns4-16k-enhancement') | |
| def predict_song(audio_path): | |
| est_sources = model.separate_file(path=audio_path) | |
| torchaudio.save("enhanced_wham.wav", est_sources[:, :, 0].detach().cpu(), 16000) | |
| return "enhanced_wham.wav" | |
| # Create title, description and article strings | |
| title = "Denoise Audio Using Sepformer" | |
| description = "Using SepFormer model implemented with SpeechBrain" | |
| article = "Tham khao Hunggingface [speechbrain/sepformer-wsj02mixt](https://huggingface.co/speechbrain/sepformer-wsj02mix)." | |
| # Create the Gradio demo | |
| demo = gr.Interface(fn=predict_song, # mapping function from input to output | |
| inputs=gr.Audio(type="filepath"), # what are the inputs? | |
| outputs=gr.File(file_count="multiple", file_types=[".wav"]), # our fn has two outputs, therefore we have two outputs | |
| title=title, | |
| description=description, | |
| article=article) | |
| # Launch the demo! | |
| demo.launch() |