Kangarroar's picture
Update app.py
288512c
raw
history blame
1.73 kB
title = "DIFF-SVC"
description = """
<p>
<body style="background-color: #18181a; color: white;"></body>
<center>
<h1>DIFF-SVC Inference Cloud</h1>
This is a Cloud Inference where you can render your models with your wav files
<p>Enter a link:</p>
<input type="text" id="link-input"/>
<p>Upload a WAV file:</p> <input type="file" id="wav-input" accept=".wav"/>
<button id="render-button">Render</button>
<p>Diff-SVC prediction:</p>
<p id="prediction-output"></p>
</center>
</p>
"""
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 *
############
logging.getLogger('numba').setLevel(logging.WARNING)
# 工程文件夹名,训练时用的那个
project_name = "Unnamed"
model_path = f'./checkpoints/Unnamed/model_ckpt_steps_192000.ckpt'
config_path=f'./checkpoints/Unnamed/config.yaml'
hubert_gpu=True
svc_model = Svc(project_name,config_path,hubert_gpu, model_path)
print('model loaded')
wav_fn='raw/cilliafinal.wav'#支持多数音频格式,无需手动转为wav
demoaudio, sr = librosa.load(wav_fn)
key = -8 # 音高调整,支持正负(半音)
# 加速倍数
pndm_speedup = 20
wav_gen='queeeeee.wav'#直接改后缀可以保存不同格式音频,如flac可无损压缩
f0_tst, f0_pred, audio = run_clip(svc_model,file_path=wav_fn, key=key, acc=pndm_speedup, use_crepe=True, use_pe=True, thre=0.05,
use_gt_mel=False, add_noise_step=500,project_name=project_name,out_path=wav_gen)