webui.py

import gradio as gr
import subprocess
import os
import soundfile as sf
from pathlib import Path
import separate

def audio_model_inference(file, output_folder, model_path, denoise, margin, chunks, n_fft, dim_t, dim_f):
    filename = str(Path(file)).split('/')[-1]
    # 执行调用
    audio_worker = separate.Predictor(args={
    										"files": Path(file),
    										"output": Path(output_folder),
    										"model_path": Path(model_path),
    										"denoise": denoise,
    										"margin": margin,
    										"chunks": chunks,
    										"n_fft": n_fft,
    										"dim_t": dim_t,
    										"dim_f": dim_f
											})
    vocals, no_vocals, sampling_rate = audio_worker.predict(files)
    sf.write(os.path.join(output_folder, mp3_filename + "_no_vocals.wav"), no_vocals, sampling_rate)
    sf.write(os.path.join(output_folder, mp3_filename + "_vocals.wav"), vocals, sampling_rate)

    # 生成分离后的文件名
    vocals_filename = f"{os.path.splitext(filename)[0]}_vocals.wav"
    no_vocals_filename = f"{os.path.splitext(filename)[0]}_no_vocals.wav"
    
    # 读取输出文件
    vocals_file = f"{os.path.splitext(os.path.basename(files[0]))[0]}_vocals.wav"
    no_vocals_file = f"{os.path.splitext(os.path.basename(files[0]))[0]}_no_vocals.wav"
    vocals_path = os.path.join(output_folder, vocals_file)
    no_vocals_path = os.path.join(output_folder, no_vocals_file)
    
    # 确保文件存在
    if not os.path.exists(vocals_path) or not os.path.exists(no_vocals_path):
        return "错误：输出文件未找到。"
    
    # 读取音频文件
    vocals_audio = open(vocals_path, 'rb').read()
    no_vocals_audio = open(no_vocals_path, 'rb').read()
    
    return (vocals_audio, no_vocals_audio)

# Gradio 界面组件
inputs = [
    gr.File(label="源音频文件", type='filepath', file_count='single'),
    gr.Textbox(label="输出文件夹", value="./"),
    gr.Textbox(label="模型路径", value="./models/MDX_Net_Models/UVR-MDX-NET-Inst_HQ_3.onnx"),
    gr.Checkbox(label="启用降噪", value=False),
    gr.Number(label="边距", value=0.1),
    gr.Number(label="块大小", value=1024),
    gr.Number(label="FFT大小", value=2048),
    gr.Number(label="时间维度", value=512),
    gr.Number(label="频率维度", value=64)
]

outputs = [gr.Audio(label="人声"), gr.Audio(label="无人声")]

# 创建界面
iface = gr.Interface(
    fn=audio_model_inference,
    inputs=inputs,
    outputs=outputs,
    title="音频分离模型",
    description="上传音频文件并配置参数，使用音频分离模型处理它们。"
)

iface.launch()