-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
86 lines (69 loc) · 3.52 KB
/
utils.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
import pyworld as pw
from hyperparams import hyperparams
import tensorflow as tf
import numpy as np
import librosa
hp = hyperparams()
def get_sp(fpath: str): # 原始人的sp
wav, _ = librosa.load(fpath, sr=hp.SR, mono=True, dtype=np.float64) # librosa.load 返回音频信号值 & 采样率
# mono=False声音保持原通道数
f0, timeaxis = pw.harvest(wav, hp.SR)
sp = pw.cheaptrick(wav, f0, timeaxis, hp.SR, fft_size=hp.N_FFT) # sp:频谱包络；pw.cheaptrick 谐波频谱包络估计算法
# sp = pw.cheaptrick(wav, mean_f0, timeaxis, hp.SR, fft_size=hp.N_FFT)
coded_sp = pw.code_spectral_envelope(sp, hp.SR, hp.CODED_DIM)
# 将频谱包络sp再 ？压缩？;返回值是：ndarray
# pw.code_spectral_envelope ：减小频谱包络 和 非周期性的 尺寸 。
# https://blog.csdn.net/weixin_32393347/article/details/88623256
coded_sp = coded_sp.T # ndarray 的 转置矩阵
return np.array(coded_sp)
# 0314_黄圣杰
def get_f0(fpath:str): # 求原始/目标人的均值，然后再拿出去做差
wav, _ = librosa.load(fpath, sr=hp.SR, mono=True, dtype=np.float64) # librosa.load 返回音频信号值 & 采样率
f0, timeaxis = pw.harvest(wav, hp.SR) # f0是一维数组，每帧会有一个f0
# return sum(f0) / sum(timeaxis)
return f0
# total_f0 = 0
# num = 0
# for i in range(timeaxis):
# total_f0 += f0
# num = num+1
# average_f0 = total_f0/num
# return average_f0
def get_mel(fpath:str):
# print("进入 get_mel 函数")
# print("fpath = "+fpath)
fpath = fpath.strip() # 防止输入进来的路径，还带着回车符号'\n'
wav, sr = librosa.load(fpath, sr=hp.SR, mono=True, dtype=np.float64) # 返回音频信号值 & 采样率
# print("librosa 成功载入音频")
f0, timeaxis = pw.harvest(wav, hp.SR)
sp = pw.cheaptrick(wav, f0, timeaxis, hp.SR, fft_size=hp.N_FFT) # sp:频谱包络；pw.cheaptrick 谐波频谱包络估计算法
# win size800，hop200，fft 1024
mel = librosa.feature.melspectrogram(y=wav, sr=sr, n_mels=80, n_fft=1024, hop_length=200)
# shape=(n_mels, t) = (80, t)
return mel
def learning_rate_decay(init_lr, global_step, warmup_steps=4000.):
step = tf.cast(global_step + 1, dtype=tf.float32) # step = 1.0
# 数据类型转换 cast(x, dtype, name=None)
# 第一个参数 x: 待转换的数据（张量）
# 第二个参数 dtype： 目标数据类型
# 第三个参数 name： 可选参数，定义操作的名称
return init_lr * warmup_steps ** 0.5 * tf.minimum(step * warmup_steps ** -1.5, step ** -0.5)
def control_weight(global_step, function_type='logistic'):
# KL散度的权重
steps = tf.cast(global_step + 1, dtype=tf.float32)
if function_type == 'logistic':
return 1/(1 + tf.exp(-hp.K * (steps - hp.X0)))
elif function_type == 'linear':
return tf.math.minimum(1, steps/hp.X0) # 似乎需要 1. 才能运行'linear'情况
else:
raise Exception('No Supported VAE LOSS WEIGHT FUNCTION.')
# tmp = tf.get_variable('global_step', initializer=0, dtype=tf.int32, trainable=False)
# print("1 ******")
# print(learning_rate_decay(0.001, tmp,)) # Tensor("mul_1:0", shape=(), dtype=float32)
# print("2 ******")
# with tf.Session() as sess:
# sess.run(tf.global_variables_initializer())
# print("3 ******")
# print(sess.run(tmp)) # 0
# print("4 ******")
# print(sess.run(learning_rate_decay(0.001, tmp, ))) # 2.5e-07 = 0.001 / 4000