Spaces:
Running
Running
| # reference: https://huggingface.co/spaces/r3gm/Audio_separator | |
| import gradio as gr | |
| import shutil | |
| import numpy as np | |
| from pathlib import Path | |
| import os | |
| import time | |
| import torch | |
| from uvr_processing import process_uvr_task | |
| def get_device_info(): | |
| if torch.cuda.is_available(): | |
| device = f"GPU ({torch.cuda.get_device_name(0)})" | |
| else: | |
| device = "CPU" | |
| return f"当前运行环境: {device}" | |
| def inference(audio_file: str, | |
| stem: str = "vocal",) -> list[str]: | |
| # audio_file = '/private/var/folders/02/_9ymjkz12xq8m_xh5592pl840000gn/T/gradio/74c3de047a439ea3cfb8e2d1ee6e5a85ea999d3eb30537b88d386aac177902d0/Spare Zeit und Aufwand mit den Servicevertragen von Mercedes-Benz Trucks..m4a' | |
| if not audio_file: | |
| raise ValueError("The audio path is missing.") | |
| if not stem: | |
| raise ValueError("Please select 'vocal' or 'background' stem.") | |
| audio_file = Path(audio_file) | |
| mdxnet_models_dir = Path("./mdx_models") | |
| output_dir = Path("./output") | |
| outputs = [] | |
| start_time = time.time() | |
| background_path, vocals_path = process_uvr_task( | |
| mdxnet_models_dir=mdxnet_models_dir, | |
| input_file_path=audio_file, | |
| output_dir=output_dir, | |
| ) | |
| end_time = time.time() | |
| execution_time = end_time - start_time | |
| outputs.append(str(background_path)) | |
| outputs.append(str(vocals_path)) | |
| return outputs | |
| def get_gui(theme, title, description): | |
| with gr.Blocks(theme=theme) as app: | |
| # Add title and description | |
| gr.Markdown(title) | |
| gr.Markdown(description) | |
| gr.Markdown(get_device_info()) | |
| audio_input = gr.Audio(label="Audio file", type="filepath") # type: str | Path | bytes | tuple[int, np.ndarray] | None | |
| download_button = gr.Button("Inference") | |
| file_output = gr.File(label="Result", file_count="multiple", interactive=False) | |
| download_button.click( | |
| inference, | |
| inputs=[audio_input], | |
| outputs=[file_output], | |
| ) | |
| return app | |
| if __name__ == "__main__": | |
| title = "<center><strong><font size='7'>Vocal BGM Separator</font></strong></center>" | |
| description = "This demo uses the MDX-Net models to perform Ultimate Vocal Remover (uvr) task for vocal and background sound separation." | |
| theme = "NoCrypt/miku" | |
| BASE_DIR = "." # os.path.dirname(os.path.dirname(os.path.abspath(__file__))) | |
| mdxnet_models_dir = os.path.join(BASE_DIR, "mdx_models") | |
| output_dir = os.path.join(BASE_DIR, "output") | |
| # confirm entry points from client | |
| # client_local = Client("http://127.0.0.1:7860") | |
| # client = Client(f"{HF_USERNAME}/{HF_SPACENAME}", hf_token=HF_TOKEN) | |
| # client_local.view_api() | |
| # entry point for GUI | |
| # predict(audio_file, api_name="/inference") -> result | |
| app_gui = get_gui(theme, title, description) | |
| # entry point for API | |
| # predict(audio_file, api_name="/predict") -> output | |
| app_api = gr.Interface( | |
| fn=inference, | |
| inputs=gr.Audio(type="filepath"), | |
| outputs=gr.File(file_count="multiple"), | |
| ) | |
| app = gr.TabbedInterface( | |
| interface_list=[app_gui, app_api], | |
| tab_names=["GUI", "API"] | |
| ) | |
| app.queue(default_concurrency_limit=40) | |
| app.launch() |