app.py

import os
import torch
import librosa
import gradio as gr
from scipy.io.wavfile import write
from huggingface_hub import hf_hub_download, snapshot_download
import utils
from models import SynthesizerTrn
from mel_processing import mel_spectrogram_torch
from speaker_encoder.voice_encoder import SpeakerEncoder
import logging
from transformers import Wav2Vec2FeatureExtractor, HubertModel

# 设置日志级别和格式
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

# 模型配置
MODEL_CONFIG = {
    "freevc": {
        "repo_id": "guetLzy/Chinese-FreeVC-Model",
        "files": ["G_17000.pth", "G_35000.pth"]
    },
    "hubert": {
        "repo_id": "guetLzy/chinese-hubert-large-fariseq-ckpt",
    }
}

# 设备设置
device = 'cuda' if torch.cuda.is_available() else 'cpu'

# 可用的模型选项
MODEL_OPTIONS = {
    "Model_17000": "model/G_17000.pth",
    "Model_35000": "model/G_35000.pth",
}

def download_models():
    """下载所有需要的模型文件"""
    os.makedirs("model", exist_ok=True)
    os.makedirs("hubert/chinese-hubert-large-fairseq-ckpt", exist_ok=True)

    freevc_paths = {}
    for model_name, model_path in MODEL_OPTIONS.items():
        path = hf_hub_download(
            repo_id=MODEL_CONFIG["freevc"]["repo_id"],
            filename=os.path.basename(model_path),
            local_dir="model",
            resume_download=True
        )
        freevc_paths[model_name] = path

    hubert_dir = "hubert/chinese-hubert-large-fairseq-ckpt"
    snapshot_download(
        repo_id=MODEL_CONFIG["hubert"]["repo_id"],
        local_dir=hubert_dir,
        repo_type="model",
        resume_download=True
    )
    hubert_paths = {"snapshot": hubert_dir}

    return {
        "freevc": freevc_paths,
        "hubert": hubert_paths
    }

def load_hubert(hubert_dir, status_list):
    """加载HuBERT模型（使用fairseq格式的检查点）"""
    status_list.append("正在加载 HuBERT 模型...")
    logger.info("正在加载 HuBERT 模型...")
    model = HubertModel.from_pretrained(hubert_dir)
    feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(hubert_dir)
    return model.to(device).float().eval(), feature_extractor

def load_freevc(model_path, status_list):
    """加载FreeVC模型（使用本地配置文件）"""
    status_list.append(f"正在从 {model_path} 加载 FreeVC 模型...")
    logger.info(f"正在从 {model_path} 加载 FreeVC 模型...")
    hps = utils.get_hparams_from_file("configs/freevc.json")

    net_g = SynthesizerTrn(
        hps.data.filter_length // 2 + 1,
        hps.train.segment_size // hps.data.hop_length,
        **hps.model
    ).to(device)

    utils.load_checkpoint(model_path, net_g, None, True)
    net_g.eval()

    smodel = SpeakerEncoder("speaker_encoder/ckpt/pretrained_bak_5805000.pt") if hps.model.use_spk else None
    return net_g, smodel, hps

# 预加载模型
status_list = ["正在下载模型..."]
logger.info("正在下载模型...")
model_paths = download_models()
status_list.append(f"模型路径: {model_paths}")
logger.info(f"模型路径: {model_paths}")
status_list.append("正在初始化 HuBERT...")
logger.info("正在初始化 HuBERT...")
hubert_dir = "hubert/chinese-hubert-large-fairseq-ckpt"
hubert_model, hubert_feature_extractor = load_hubert(hubert_dir, status_list)

def voice_conversion(src_audio, tgt_audio, output_name, model_selection):
    """执行语音转换"""
    status_list = ["开始语音转换..."]
    
    try:
        # 加载选中的FreeVC模型
        freevc_model, speaker_model, hps = load_freevc(MODEL_OPTIONS[model_selection], status_list)

        with torch.no_grad():
            # 处理目标音频
            status_list.append("处理目标音频...")
            wav_tgt, _ = librosa.load(tgt_audio, sr=hps.data.sampling_rate)
            wav_tgt, _ = librosa.effects.trim(wav_tgt, top_db=20)
            
            if hps.model.use_spk:
                status_list.append("提取目标音色特征（使用说话人编码器）...")
                g_tgt = speaker_model.embed_utterance(wav_tgt)
                g_tgt = torch.from_numpy(g_tgt).unsqueeze(0).to(device)
            else:
                status_list.append("生成目标音频 Mel 频谱图...")
                wav_tgt = torch.from_numpy(wav_tgt).unsqueeze(0).to(device)
                mel_tgt = mel_spectrogram_torch(
                    wav_tgt,
                    hps.data.filter_length,
                    hps.data.n_mel_channels,
                    hps.data.sampling_rate,
                    hps.data.hop_length,
                    hps.data.win_length,
                    hps.data.mel_fmin,
                    hps.data.mel_fmax
                )

            # 处理源音频
            status_list.append("处理源音频（转换为16kHz）...")
            wav_src, _ = librosa.load(src_audio, sr=16_000)
            inputs = hubert_feature_extractor(
                wav_src,
                return_tensors="pt",
                sampling_rate=16_000
            ).input_values.to(device)
            
            status_list.append("提取源音频特征...")
            c = hubert_model(inputs.float()).last_hidden_state.transpose(1, 2)

            # 执行转换
            status_list.append("执行语音转换...")
            audio = freevc_model.infer(c, g=g_tgt) if hps.model.use_spk else freevc_model.infer(c, mel=mel_tgt)
            
            # 保存结果
            status_list.append("保存转换结果...")
            os.makedirs("output", exist_ok=True)
            output_path = f"output/{output_name}.wav"
            write(output_path, hps.data.sampling_rate, audio[0][0].data.cpu().float().numpy())
            
            status_list.append("转换完成")
            return output_path, "\n".join(status_list)

    except Exception as e:
        logger.error(f"转换错误: {str(e)}")
        status_list.append(f"转换失败: {str(e)}")
        return None, "\n".join(status_list)

# Gradio界面
with gr.Blocks(title="Chinese-FreeVC 语音转换" ,theme='NoCrypt/miku') as app:
    gr.Markdown("## Chinese-FreeVC 语音转换系统")

    with gr.Row():
        with gr.Column():
            src_input = gr.Audio(label="源语音", type="filepath")
            tgt_input = gr.Audio(label="目标音色", type="filepath")
            with gr.Row():  # 输出文件名和模型选择在同一排
                model_dropdown = gr.Dropdown(
                    choices=list(MODEL_OPTIONS.keys()),
                    label="选择模型",
                    value="Model_17000"
                )
                output_name = gr.Textbox(label="输出文件名", value="converted")
            convert_btn = gr.Button("开始转换", variant="primary")
        
        with gr.Column():
            output_audio = gr.Audio(label="转换结果", interactive=False)
            status = gr.Textbox(label="状态", value="待机", interactive=False)

    convert_btn.click(
        fn=voice_conversion,
        inputs=[src_input, tgt_input, output_name, model_dropdown],
        outputs=[output_audio, status],
        api_name="convert"
    )

if __name__ == "__main__":
    app.launch(server_name="0.0.0.0", server_port=7860)