app.py

import gradio as gr
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import json
import os
import tempfile
import shutil
import requests
from pathlib import Path
###################################################
from utils.hparams import hparams
from preprocessing.data_gen_utils import get_pitch_parselmouth,get_pitch_crepe
import numpy as np
import matplotlib.pyplot as plt
import IPython.display as ipd
import utils
import librosa
import torchcrepe
from infer import *
import logging
from infer_tools.infer_tool import *
import io


spk_dict = {
    "雷电将军": {"model_name": './models/genshin/raiden.ckpt', "config_name": './models/genshin/config.yaml'}
}
print(spk_dict)

project_name = "Unnamed"
model_path = spk_dict['雷电将军']['model_name']
config_path= spk_dict['雷电将军']['config_name']
hubert_gpu = False
svc_model = Svc(project_name, config_path, hubert_gpu, model_path)

print(svc_model)

def vc_fn(sid, audio_record, audio_upload, tran, pndm_speedup=20):
    print(sid)
    if audio_upload is not None:
        audio_path = audio_upload
    elif audio_record is not None:
        audio_path = audio_record
    else:
        return "你需要上传wav文件或使用网页内置的录音！", None

    tran = int(tran)
    pndm_speedup = int(pndm_speedup)
    print('model loaded')
    # demoaudio, sr = librosa.load(audio_path)
    key = tran # 音高调整，支持正负（半音）
    # 加速倍数
    wav_gen='./output.wav'
    
    # Show the spinner and run the run_clip function inside the 'with' block
    f0_tst, f0_pred, audio = run_clip(svc_model, file_path=audio_path, key=key, acc=pndm_speedup, use_crepe=True, use_pe=True, thre=0.1,
                                            use_gt_mel=False, add_noise_step=500, project_name=project_name, out_path=wav_gen)
    
    audio, sr = librosa.load(wav_gen)
    f0_gen,_=get_pitch_parselmouth(*svc_model.vocoder.wav2spec(wav_gen),hparams)
    f0_tst[f0_tst==0]=np.nan#ground truth f0
    f0_pred[f0_pred==0]=np.nan#f0 pe predicted
    f0_gen[f0_gen==0]=np.nan#f0 generated
    fig=plt.figure(figsize=[15,5])
    plt.plot(np.arange(0,len(f0_tst)),f0_tst,color='black',label="f0_tst")
    plt.plot(np.arange(0,len(f0_pred)),f0_pred,color='orange',label="f0_pred")
    plt.plot(np.arange(0,len(f0_gen)),f0_gen,color='red',label="f0_gen")
    plt.axhline(librosa.note_to_hz('C4'),ls=":",c="blue",label="C4")
    plt.axhline(librosa.note_to_hz('G4'),ls=":",c="green",label="G4")
    plt.axhline(librosa.note_to_hz('C5'),ls=":",c="orange",label="C5")
    plt.axhline(librosa.note_to_hz('F#5'),ls=":",c="red",label="F#5")
    #plt.axhline(librosa.note_to_hz('A#5'),ls=":",c="black",label="分割线")
    plt.legend()
    plt.savefig('./temp.png')

    return "Success", (sr, audio), gr.Image.update("temp.png") # hparams['audio_sample_rate']


app = gr.Blocks()
with app:
    with gr.Tabs():
        with gr.TabItem("Basic"):
            gr.Markdown(value="""
            本模型基于diffsvc训练，使用雷电将军语音数据，模型与数据都来源于[Erythrocyte/Diff-SVC_Genshin_Datasets](https://huggingface.co/datasets/Erythrocyte/Diff-SVC_Genshin_Datasets)，
            
            本页面支持**60s以内**的**无伴奏**wav格式，或使用**网页内置**的录音（二选一）
            
            转换效果取决于源音频语气、节奏是否与目标音色相近，以及音域是否超出目标音色音域范围
            """)
            speaker_id = gr.Dropdown(label="音色", choices=['雷电将军'], value="雷电将军")
            record_input = gr.Audio(source="microphone", label="录制你的声音", type="filepath", elem_id="audio_inputs")
            upload_input = gr.Audio(source="upload", label="上传音频（长度小于60秒）", type="filepath",
                                    elem_id="audio_inputs")
            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_speedup = gr.Number(label="加速倍数", value=20)
            vc_submit = gr.Button("转换", variant="primary")
            out_audio = gr.Audio(label="Output Audio")
            gr.Markdown(value="""
                        无用信息
                        """)
            out_message = gr.Textbox(label="Output")
            gr.Markdown(value="""f0曲线可以直观的显示跑调情况：
                        """)
            f0_image = gr.Image(label="f0曲线")
        vc_submit.click(vc_fn, [speaker_id, record_input, upload_input, vc_transform, vc_speedup],
                        [out_message, out_audio, f0_image])

    app.launch()