app.py

import io
import os

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_path = "logs/48k/aqua.pth"
config_path = "configs/config.json"

svc_model = Svc(model_path, config_path)


def vc_fn(input_audio, vc_transform, term):
    if not term:
        return "请阅读并同意《AI阿夸模型使用协议》", None
    if input_audio is None:
        return "请上传音频", 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
    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)
    print(audio.shape)
    out_wav_path = io.BytesIO()
    soundfile.write(out_wav_path, audio, 24000, format="wav")
    out_wav_path.seek(0)

    # sid = sid_map[sid]
    sid = "aqua"
    out_audio, out_sr = svc_model.infer(sid, vc_transform, out_wav_path)
    _audio = out_audio.cpu().numpy()
    return "Success", (48000, _audio)


inputs = [
    gr.inputs.Audio(source="upload"),
    gr.inputs.Number(default=0),
    gr.Checkbox(label="您已阅读并同意《AI阿夸模型使用协议》")
]

outputs = [
    "text",
    gr.outputs.Audio(type="numpy")
]

example = [
    ["./raw/大手拉小手.wav", 0, False]
]

des = """
## 在使用此模型前请阅读[AI阿夸模型使用协议](https://huggingface.co/spaces/DoNotSelect/AI-minato_aqua/blob/main/terms.md)
"""

demo = gr.Interface(
    fn=vc_fn,
    inputs=inputs,
    outputs=outputs,
    layout="horizontal",
    theme="huggingface",
    description=des,
    examples=example,
    cache_examples=True
)

if __name__ == "__main__":
    demo.launch()