webui.py

import gradio as gr
import subprocess
import os

def audio_model_inference(files, output_folder, model_path, denoise, margin, chunks, n_fft, dim_t, dim_f):
    # 构建命令行调用字符串
    cmd = f"separate.py {' '.join(files)}"
    if output_folder:
        cmd += f" -o {output_folder}"
    if model_path:
        cmd += f" -m {model_path}"
    if denoise:
        cmd += " -d"
    if margin:
        cmd += f" -M {margin}"
    if chunks:
        cmd += f" -c {chunks}"
    if n_fft:
        cmd += f" -F {n_fft}"
    if dim_t:
        cmd += f" -t {dim_t}"
    if dim_f:
        cmd += f" -f {dim_f}"
    
    # 执行命令行调用
    result = subprocess.run(cmd, shell=True, capture_output=True, text=True)
    
    # 检查命令是否成功执行
    if result.returncode != 0:
        return f"错误：{result.stderr}"
    
    # 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='binary', file_count='multiple'),
    gr.Textbox(label="输出文件夹", default="./"),
    gr.Textbox(label="模型路径", default="./models/MDX_Net_Models/UVR-MDX-NET-Inst_HQ_3.onnx"),
    gr.Checkbox(label="启用降噪", default=False),
    gr.Number(label="边距", default=0.1),
    gr.Number(label="块大小", default=1024),
    gr.Number(label="FFT大小", default=2048),
    gr.Number(label="时间维度", default=512),
    gr.Number(label="频率维度", default=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()