app.py

# coding=utf-8
import time
import gradio as gr
import librosa
import numpy as np
import soundfile
import io
from inference.infer_tool import Svc

model_path = "model/model.pth"
config_path = "model/config.json"
svc_model = Svc(model_path, config_path)


def sovits(input_audio, vc_transform):
    start = time.perf_counter()
    if input_audio is None:
        return "请上传音频", None, None
    sampling_rate, audio = input_audio
    duration = audio.shape[0] / sampling_rate
    if duration > 30:
        return "请上传小于30s的音频，长音频的转换请在本地进行", None, None
    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 != 24000:
        audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=24000)
    out_wav_path = io.BytesIO()
    soundfile.write(out_wav_path, audio, 24000, format="wav")
    out_wav_path.seek(0)
    sid = 'xiaoke'
    out_audio, out_sr = svc_model.infer(sid, vc_transform, out_wav_path)
    _audio = out_audio.cpu().numpy()
    return "生成成功!", (48000, _audio), f"生成耗时 {round(time.perf_counter()-start, 2)} s"

download_audio_js = """

() =>{{

    let root = document.querySelector("body > gradio-app");

    if (root.shadowRoot != null)

        root = root.shadowRoot;

    let audio = root.querySelector("#{audio_id}").querySelector("audio");

    if (audio == undefined)

        return;

    audio = audio.src;

    let oA = document.createElement("a");

    oA.download = Math.floor(Math.random()*100000000)+'.wav';

    oA.href = audio;

    document.body.appendChild(oA);

    oA.click();

    oA.remove();

}}

"""

if __name__ == '__main__':
    with gr.Blocks() as app:
        gr.Markdown(
            "# <center> AI小可\n"
        )
        with gr.Row():
            with gr.Column():
                input_audio = gr.inputs.Audio(label="待转换音频")
                vc = gr.inputs.Number(label="音高调整", default=0)
                btn = gr.Button(value="生成")
            with gr.Column():
                o1 = gr.Textbox(label="Output Message")
                o2 = gr.Audio(label="Output Audio", elem_id=f"vc-audio")
                o3 = gr.Textbox(label="Extra Info")
                download = gr.Button("Download Audio")
            btn.click(sovits, inputs=[input_audio, vc], outputs=[o1, o2, o3])
            download.click(None, [], [], _js=download_audio_js.format(audio_id=f"vc-audio"))
    app.queue(concurrency_count=1).launch()