def SpectralSub(signal, wlen, inc, NIS, a, b):
"""
谱减法滤波
:param signal:
:param wlen:
:param inc:
:param NIS:
:param a:
:param b:
:return:
"""
wnd = np.hamming(wlen)
y = enframe(signal, wnd, inc)
fn, flen = y.shape
y_a = np.abs(np.fft.fft(y, axis=1))
y_a2 = np.power(y_a, 2)
y_angle = np.angle(np.fft.fft(y, axis=1))
Nt = np.mean(y_a2[:NIS, ], axis=0)
y_a2 = np.where(y_a2 >= a * Nt, y_a2 - a * Nt, b * Nt)
X = y_a2 * np.cos(y_angle) + 1j * y_a2 * np.sin(y_angle)
hatx = np.real(np.fft.ifft(X, axis=1))
sig = np.zeros(int((fn - 1) * inc + wlen))
for i in range(fn):
start = i * inc
sig[start:start + flen] += hatx[i, :]
return sig
def Nmfcc(x, fs, p, frameSize, inc):
"""
计算mfcc系数
:param x: 输入信号
:param fs: 采样率
:param p: Mel滤波器组的个数
:param frameSize: 分帧的每帧长度
:param inc: 帧移
:return:
"""
# 预处理-预加重
xx = lfilter([1, -0.97], [1], x)
# 预处理-分幀
xx = enframe(xx, frameSize, inc)
# 预处理-加窗
xx = np.multiply(xx, np.hanning(frameSize))
# 计算FFT
xx = np.fft.fft(xx)
# 计算能量谱
xx = np.multiply(np.abs(xx), np.abs(xx))
# 计算通过Mel滤波器的能量
xx = xx[:, :frameSize // 2 + 1]
bank = melbankm(p, frameSize, fs, 0, 0.5 * fs, 0)
ss = np.matmul(xx, bank.T)
# 计算DCT倒谱
n_dct = 20
M = bank.shape[0]
m = np.array([i for i in range(M)])
mfcc = np.zeros((ss.shape[0], n_dct))
for n in range(n_dct):
mfcc[:, n] = np.sqrt(2 / M) * np.sum(np.multiply(
np.log(ss), np.cos((2 * m - 1) * n * np.pi / 2 / M)),
axis=1)
return mfcc
def pitch_Corr(x, wnd, inc, T1, fs, miniL=10):
"""
自相关法基音周期检测函数
:param x:
:param wnd:
:param inc:
:param T1:
:param fs:
:param miniL:
:return:
"""
y = enframe(x, wnd, inc)
fn = y.shape[0]
if isinstance(wnd, int):
wlen = wnd
else:
wlen = len(wnd)
voiceseg, vsl, SF, Ef = pitch_vad(x, wnd, inc, T1, miniL)
lmin = fs // 500 # 基音周期的最小值
lmax = fs // 60 # 基音周期的最大值
period = np.zeros(fn)
for i in range(vsl):
ixb = voiceseg[i]['start']
ixd = voiceseg[i]['duration']
for k in range(ixd):
ru = np.correlate(y[k + ixb, :], y[k + ixb, :], 'full')
ru = ru[wlen:]
tloc = np.argmax(ru[lmin:lmax])
period[k + ixb] = lmin + tloc
return voiceseg, vsl, SF, Ef, period
def vad_specEN(data, wnd, inc, NIS, thr1, thr2, fs):
from scipy.signal import medfilt
x = enframe(data, wnd, inc)
X = np.abs(np.fft.fft(x, axis=1))
if len(wnd) == 1:
wlen = wnd
else:
wlen = len(wnd)
df = fs / wlen
fx1 = int(250 // df + 1) # 250Hz位置
fx2 = int(3500 // df + 1) # 500Hz位置
km = wlen // 8
K = 0.5
E = np.zeros((X.shape[0], wlen // 2))
E[:, fx1 + 1:fx2 - 1] = X[:, fx1 + 1:fx2 - 1]
E = np.multiply(E, E)
Esum = np.sum(E, axis=1, keepdims=True)
P1 = np.divide(E, Esum)
E = np.where(P1 >= 0.9, 0, E)
Eb0 = E[:, 0::4]
Eb1 = E[:, 1::4]
Eb2 = E[:, 2::4]
Eb3 = E[:, 3::4]
Eb = Eb0 + Eb1 + Eb2 + Eb3
prob = np.divide(Eb + K, np.sum(Eb + K, axis=1, keepdims=True))
Hb = -np.sum(np.multiply(prob, np.log10(prob + 1e-10)), axis=1)
for i in range(10):
Hb = medfilt(Hb, 5)
Me = np.mean(Hb)
eth = np.mean(Hb[:NIS])
Det = eth - Me
T1 = thr1 * Det + Me
T2 = thr2 * Det + Me
voiceseg, vsl, SF, NF = vad_revr(Hb, T1, T2)
return voiceseg, vsl, SF, NF, Hb
def pitch_vad(x, wnd, inc, T1, miniL=10):
"""
使用能熵比检测基音，实际上就是语音分段
:param x:
:param wnd:
:param inc:
:param T1:
:param miniL:
:return:
"""
y = enframe(x, wnd, inc)
fn = y.shape[0]
if isinstance(wnd, int):
wlen = wnd
else:
wlen = len(wnd)
Sp = np.abs(np.fft.fft(y, axis=1))
Sp = Sp[:, :wlen // 2 + 1]
Esum = np.sum(np.multiply(Sp, Sp), axis=1)
prob = Sp / np.sum(Sp, axis=1, keepdims=True)
H = -np.sum(np.multiply(prob, np.log10(prob + 1e-16)), axis=1)
H = np.where(H < 0.1, np.max(H), H)
Ef = np.sqrt(1 + np.abs(Esum / H))
Ef = Ef / np.max(Ef)
zseg = findSegment(np.where(Ef > T1)[0])
zsl = len(zseg.keys())
SF = np.zeros(fn)
for k in range(zsl):
if zseg[k]['duration'] < miniL:
zseg.pop(k)
else:
SF[zseg[k]['start']:zseg[k]['end']] = 1
return zseg, len(zseg.keys()), SF, Ef
def pitch_Ceps(x, wnd, inc, T1, fs, miniL=10):
"""
倒谱法基音周期检测函数
:param x:
:param wnd:
:param inc:
:param T1:
:param fs:
:param miniL:
:return:
"""
y = enframe(x, wnd, inc)
fn = y.shape[0]
if isinstance(wnd, int):
wlen = wnd
else:
wlen = len(wnd)
voiceseg, vsl, SF, Ef = pitch_vad(x, wnd, inc, T1, miniL)
lmin = fs // 500 # 基音周期的最小值
lmax = fs // 60 # 基音周期的最大值
period = np.zeros(fn)
y1 = y[np.where(SF == 1)[0], :]
y1 = np.multiply(y1, np.hamming(wlen))
xx = np.fft.fft(y1, axis=1)
b = np.fft.ifft(2 * np.log(np.abs(xx) + 1e-10))
Lc = np.argmax(b[:, lmin:lmax], axis=1) + lmin - 1
period[np.where(SF == 1)[0]] = Lc
return voiceseg, vsl, SF, Ef, period
def vad_corr(y, wnd, inc, NIS, th1, th2):
x = enframe(y, wnd, inc)
Ru = STAc(x.T)[0]
Rum = Ru / np.max(Ru)
thredth = np.max(Rum[:NIS])
T1 = th1 * thredth
T2 = th2 * thredth
voiceseg, vsl, SF, NF = vad_forw(Rum, T1, T2)
return voiceseg, vsl, SF, NF, Rum
def vad_pro(data, wnd, inc, NIS, thr1, thr2, mode):
"""
使用比例法检测端点
:param data:
:param wnd:
:param inc:
:param NIS:
:param thr1:
:param thr2:
:param mode:
:return:
"""
from scipy.signal import medfilt
x = enframe(data, wnd, inc)
if len(wnd) == 1:
wlen = wnd
else:
wlen = len(wnd)
if mode == 1: # 能零比
a = 2
b = 1
LEn = np.log10(1 + np.sum(np.multiply(x, x) / a, axis=1))
EZRn = LEn / (STZcr(data, wlen, inc) + b)
for i in range(10):
EZRn = medfilt(EZRn, 5)
dth = np.mean(EZRn[:NIS])
T1 = thr1 * dth
T2 = thr2 * dth
Epara = EZRn
elif mode == 2: # 能熵比
a = 2
X = np.abs(np.fft.fft(x, axis=1))
X = X[:, :wlen // 2]
Esum = np.log10(1 + np.sum(np.multiply(X, X) / a, axis=1))
prob = X / np.sum(X, axis=1, keepdims=True)
Hn = -np.sum(np.multiply(prob, np.log10(prob + 1e-10)), axis=1)
Ef = np.sqrt(1 + np.abs(Esum / Hn))
for i in range(10):
Ef = medfilt(Ef, 5)
Me = np.max(Ef)
eth = np.mean(Ef[NIS])
Det = Me - eth
T1 = thr1 * Det + eth
T2 = thr2 * Det + eth
Epara = Ef
voiceseg, vsl, SF, NF = vad_forw(Epara, T1, T2)
return voiceseg, vsl, SF, NF, Epara
def pitch_Lpc(x, wnd, inc, T1, fs, p, miniL=10):
"""
线性预测法基音周期检测函数
:param x:
:param wnd:
:param inc:
:param T1:
:param fs:
:param p:
:param miniL:
:return:
"""
from scipy.signal import lfilter
from chapter3_分析实验.lpc import lpc_coeff
y = enframe(x, wnd, inc)
fn = y.shape[0]
if isinstance(wnd, int):
wlen = wnd
else:
wlen = len(wnd)
voiceseg, vsl, SF, Ef = pitch_vad(x, wnd, inc, T1, miniL)
lmin = fs // 500 # 基音周期的最小值
lmax = fs // 60 # 基音周期的最大值
period = np.zeros(fn)
for k in range(y.shape[0]):
if SF[k] == 1:
u = np.multiply(y[k, :], np.hamming(wlen))
ar, _ = lpc_coeff(u, p)
ar[0] = 0
z = lfilter(-ar, [1], u)
E = u - z
xx = np.fft.fft(E)
b = np.fft.ifft(2 * np.log(np.abs(xx) + 1e-20))
lc = np.argmax(b[lmin:lmax])
period[k] = lc + lmin
return voiceseg, vsl, SF, Ef, period
plt.rcParams['axes.unicode_minus'] = False
data, fs = soundBase('C7_3_y.wav').audioread()
data -= np.mean(data)
data /= np.max(np.abs(data))
data = lfilter([1, -0.99], 1, data)
N = len(data)
time = [i / fs for i in range(N)] # 设置时间
wlen = 240
inc = 80
overlap = wlen - inc
n2 = [i for i in range(wlen // 2)]
w1 = [i / overlap for i in range(overlap)]
w2 = [i / overlap for i in range(overlap - 1, -1, -1)]
wnd = np.hamming(wlen)
X = enframe(data, wnd, inc)
fn = X.shape[0]
Etmp = np.sum(np.power(X, 2), axis=1)
Etmp /= np.max(Etmp)
T1, r2 = 0.1, 0.5
miniL = 10
mnlong = 5
ThrC = [10, 15]
p = 12
frameTime = FrameTimeC(fn, wlen, inc, fs)
Doption = 0
voiceseg, vosl, SF, Ef, period = pitch_Ceps(data, wlen, inc, T1, fs)
Dpitch = pitfilterm1(period, voiceseg, vosl)
## 共振峰检测
from scipy.signal import lfilter
plt.rcParams['font.sans-serif'] = ['SimHei']
plt.rcParams['axes.unicode_minus'] = False
data, fs = soundBase('C7_2_y.wav').audioread()
data -= np.mean(data)
data /= np.max(np.abs(data))
N = len(data)
time = [i / fs for i in range(N)] # 设置时间
p = 12
wlen, inc = 200, 80
msoverlap = wlen - inc
y = enframe(data, wlen, inc)
fn = y.shape[0]
Acoef = np.zeros((y.shape[0], p + 1))
resid = np.zeros(y.shape)
synFrame = np.zeros(y.shape)
## 7.2.1
# 求每帧的LPC系数与预测误差
for i in range(fn):
a, _ = lpc_coeff(y[i, :], p)
Acoef[i, :] = a
resid[i, :] = lfilter(a, [1], y[i, :])
# 语音合成
for i in range(fn):
synFrame[i, :] = lfilter([1], Acoef[i, :], resid[i, :])
from chapter2_基础.soundBase import *
from chapter7_语音合成.flipframe import *
from chapter3_分析实验.enframe import enframe
plt.rcParams['font.sans-serif'] = ['SimHei']
plt.rcParams['axes.unicode_minus'] = False
data, fs = soundBase('C7_1_y.wav').audioread()
wlen = 256
wnd = np.hamming(wlen)
overlap = 100
f = enframe(data, wnd, overlap)
plt.figure(figsize=(14, 12))
# 7.1.1
fn_overlap = Filpframe_OverlapA(f, wnd, overlap)
plt.subplot(3, 2, 1)
plt.plot(data / np.max(np.abs(data)), 'k')
plt.title('原始信号')
plt.subplot(3, 2, 2)
plt.title('还原信号-重叠相加法')
plt.plot(fn_overlap / np.max(np.abs(fn_overlap)), 'c')
# 7.1.2
fn_s = Filpframe_OverlapS(f, wnd, overlap)
plt.subplot(3, 2, 3)
plt.plot(data / np.max(np.abs(data)), 'k')
plt.title('原始信号')
plt.subplot(3, 2, 4)
plt.title('还原信号-重叠存储法')
plt.plot(fn_s / np.max(np.abs(fn_s)), 'c')
def vad_TwoThr(x, wlen, inc, NIS):
"""
使用门限法检测语音段
:param x: 语音信号
:param wlen: 分帧长度
:param inc: 帧移
:param NIS:
:return:
"""
maxsilence = 15
minlen = 5
status = 0
y = enframe(x, wlen, inc)
fn = y.shape[0]
amp = STEn(x, wlen, inc)
zcr = STZcr(x, wlen, inc, delta=0.01)
ampth = np.mean(amp[:NIS])
zcrth = np.mean(zcr[:NIS])
amp2 = 2 * ampth
amp1 = 4 * ampth
zcr2 = 2 * zcrth
xn = 0
count = np.zeros(fn)
silence = np.zeros(fn)
x1 = np.zeros(fn)
x2 = np.zeros(fn)
for n in range(fn):
if status == 0 or status == 1:
if amp[n] > amp1:
x1[xn] = max(1, n - count[xn] - 1)
status = 2
silence[xn] = 0
count[xn] += 1
elif amp[n] > amp2 or zcr[n] > zcr2:
status = 1
count[xn] += 1
else:
status = 0
count[xn] = 0
x1[xn] = 0
x2[xn] = 0
elif status == 2:
if amp[n] > amp2 and zcr[n] > zcr2:
count[xn] += 1
else:
silence[xn] += 1
if silence[xn] < maxsilence:
count[xn] += 1
elif count[xn] < minlen:
status = 0
silence[xn] = 0
count[xn] = 0
else:
status = 3
x2[xn] = x1[xn] + count[xn]
elif status == 3:
status = 0
xn += 1
count[xn] = 0
silence[xn] = 0
x1[xn] = 0
x2[xn] = 0
el = len(x1[:xn])
if x1[el - 1] == 0:
el -= 1
if x2[el - 1] == 0:
print('Error: Not find endding point!\n')
x2[el] = fn
SF = np.zeros(fn)
NF = np.ones(fn)
for i in range(el):
SF[int(x1[i]):int(x2[i])] = 1
NF[int(x1[i]):int(x2[i])] = 0
voiceseg = findSegment(np.where(SF == 1)[0])
vsl = len(voiceseg.keys())
return voiceseg, vsl, SF, NF, amp, zcr