app.py

import gradio as gr
import torch
import torchaudio
from speechbrain.inference.enhancement import SpectralMaskEnhancement

# 加載增強模型
enhance_model = SpectralMaskEnhancement.from_hparams(
    source="speechbrain/metricgan-plus-voicebank",
    savedir="pretrained_models/metricgan-plus-voicebank",
)

def enhance_audio(file_path):
    # 加載音頻文件
    noisy = enhance_model.load_audio(file_path).unsqueeze(0)
    
    # 假設長度為1.0的相對長度
    lengths = torch.tensor([1.0])

    # 使用模型進行語音增強
    enhanced = enhance_model.enhance_batch(noisy, lengths=lengths)

    # 將增強的音頻信號返回
    return enhanced.squeeze(0).cpu().numpy(), 16000

# 創建Gradio接口
demo = gr.Interface(
    fn=enhance_audio,
    inputs=gr.File(type="filepath"),  # 接收語音輸入，通過文件上傳
    outputs=gr.Audio(type="numpy"),  # 返回增強後的語音
    title="Speech Enhancement",
    description="Upload a noisy speech file and get the enhanced output.",
)

# 啟動Gradio應用
demo.launch(share=True)