app.py

import io

import gradio as gr
import librosa
import numpy as np
import soundfile
import torch
from inference.infer_tool import Svc
import logging
logging.getLogger('numba').setLevel(logging.WARNING)

model_name = "logs/48k/suiji.pth"
config_name = "configs/suiji.json"

svc_model = Svc(model_name, config_name)
sid_map = {
    "岁己（本音）": "suiji"
}
def vc_fn(sid, input_audio, vc_transform):
    if input_audio is None:
        return "You need to upload an audio", None
    sampling_rate, audio = input_audio
    # print(audio.shape,sampling_rate)
    duration = audio.shape[0] / sampling_rate
    audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
    if len(audio.shape) > 1:
        audio = librosa.to_mono(audio.transpose(1, 0))
    if sampling_rate != 16000:
        audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
    print(audio.shape)
    out_wav_path = io.BytesIO()
    soundfile.write(out_wav_path, audio, 16000, format="wav")
    out_wav_path.seek(0)
    sid = sid_map[sid]
    out_audio, out_sr = svc_model.infer(sid, vc_transform, out_wav_path)
    _audio = out_audio.cpu().numpy()
    return "Success", (48000, _audio)

app = gr.Blocks()
with app:
    with gr.Tabs():
        with gr.TabItem("Basic"):
            gr.Markdown(value="""
                #### 这是 sovits 3.0 48kHz AI岁己（本音）歌声（划重点）音色转换的在线demo
                
                #### 目前模型训练状态：700000steps / 640epochs
                
                #### 如果要训练自己的数据请访问：[Github仓库](https://github.com/innnky/so-vits-svc)、[教程《svc相关》](https://www.yuque.com/jiuwei-nui3d/qng6eg)
                
                #### 如果要在本地推理请使用 git lfs clone 该仓库，安装 requirements.txt 后运行 app.py 即可
                
                #### 更建议参考仓库[README.md上的推理部分](https://github.com/innnky/so-vits-svc#%E6%8E%A8%E7%90%86)，在本地使用 inference_main.py 处理
                
                #### 3060Ti 8G可推理一条20(建议) - 30s的音频，过长音频可分割后批量处理
            """)
            sid = gr.Dropdown(label="音色", choices=["岁己（本音）"], value="岁己（本音）")
            vc_input3 = gr.Audio(label="输入音频（长度请控制在25s左右，过长可能会爆内存）")
            vc_transform = gr.Number(label="变调（整数，可以正负，半音数量，升高八度就是12）", value=0)
            vc_submit = gr.Button("转换", variant="primary")
            vc_output1 = gr.Textbox(label="输出日志")
            vc_output2 = gr.Audio(label="输出音频")
        vc_submit.click(vc_fn, [sid, vc_input3, vc_transform], [vc_output1, vc_output2])
    app.launch()