forked from rohitkumara/LMS-ANN-Project
-
Notifications
You must be signed in to change notification settings - Fork 1
/
LMS_tf.py
172 lines (157 loc) · 5.32 KB
/
LMS_tf.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
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
import tensorflow as tf
import numpy as np
import soundfile as sf
import sounddevice as sd
import time
import matplotlib.pyplot as plt
import matplotlib.ticker as ticker
from pathlib import Path
class LMSFilter:
def __init__(self, noise_fn, tap=32, epoch=10, p=0.9, lr=0.001, batch_size = 50, plot = False):
self.noise_fn = noise_fn
self.tap = tap
self.epoch = epoch
self.p = p
self.lr = lr
self.plot = plot
self.batch_size = batch_size
def input_from_history(self,data):
y = np.size(data) - self.tap
ret = np.zeros([y,self.tap])
for i in range(y):
ret[i,:] = data[i:i+self.tap]
return ret
def read_speech(self,data):
print('Reading clean data...')
rootdir = Path('clean/')
filelist = [f for f in rootdir.glob('**/*') if f.is_file()]
filelist = sorted(filelist)
data, Fs = self.read_wav(str(filelist[0]))
for i in range(len(filelist)):
if i == 0:
continue
temp, Fs = self.read_wav(str(filelist[i]))
data = np.append(data, temp)
l = int((self.p)*data.size)
train_data = data[0:l]
test_data = data[l:-1]
return train_data,test_data,Fs
def read_wav(self, FILE_NAME):
self.data,samplerate = sf.read(FILE_NAME)
return self.data, samplerate
def get_data(self,filename):
data,Fs= self.read_wav(filename)
return data, Fs
def data_equalization(self,data, noise):
noise_len = noise.shape[0]
data_len = data.shape[0]
data = (np.amax(noise)/np.amax(data))*data
n = int(data_len/noise_len)+1
noi_temp = np.tile(noise,n)
trainX = noi_temp[0:data_len] + data
trainY = noi_temp[0:data_len]
#print(trainX.shape,trainY.shape)
return trainX, trainY
def data_preprocessing(self,trainX, trainY):
trainX = trainX/np.amax(trainX)
trainY = trainY/np.amax(trainY)
trainX_use = self.input_from_history(trainX)
trainX_use = trainX_use.reshape((trainX_use.shape[0],self.tap))
trainY_use = trainY[self.tap-1:-1].reshape((trainY.size-self.tap))
return trainX_use, trainY_use
def save_file(self,filename,data,Fs):
sf.write(filename,data,Fs)
def play_file(self,data,Fs):
try:
ti = np.shape(data)[0]/Fs
print('Time in sec:', ti)
sd.play(data, Fs)
time.sleep(ti)
sd.stop()
except:
sd.stop()
def measure_snr(self,noisy, noise):
data = noisy - noise
pwr_noise = (np.sum(noise**2))/noise.size
pwr_data = (np.sum(data**2))/data.size
snr = pwr_data/pwr_noise
return snr
def driver(self):
#data preprocessing step
noise, Fs = self.get_data(self.noise_fn)
data,test_data, Fs = self.read_speech(self.p)
#print(data.shape,noise.shape)
trainX_o, trainY_o = self.data_equalization(data, noise)
trainX, trainY = self.data_preprocessing(trainX_o,trainY_o)
print(trainX.shape, trainY.shape)
X = tf.placeholder(tf.float32, [None, self.tap])
Y = tf.placeholder(tf.float32, [None, 1])
W = tf.Variable(tf.random_normal([1, self.tap], stddev=0.1))
#LMS Algorithm
out = tf.matmul(X,tf.transpose(W))
yhat = out
err = Y - yhat
err = tf.reduce_mean(tf.square(err))
opt = tf.train.GradientDescentOptimizer(self.lr).minimize(err)
init_snr = self.measure_snr(trainX_o,trainY_o)
print('INIT SNR:', init_snr)
init_all = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init_all)
j=0
av_cost = np.inf
strt = time.time()
snr_plt = []
for j in range(self.epoch):
av_cost = 0
for i in range(int(trainY.shape[0]/self.batch_size)):
batch_X = trainX[i:i+self.batch_size,:].reshape([self.batch_size,self.tap])
batch_Y = trainY[i:i+self.batch_size].reshape([self.batch_size,1])
sess.run(opt, feed_dict = {X:batch_X, Y:batch_Y})
av_cost += sess.run(err, feed_dict = {X:batch_X, Y:batch_Y})
yout = sess.run(yhat, feed_dict = {X:trainX})
snr = self.measure_snr(yout,trainY_o[self.tap-1:-1].reshape([yout.size,1]))
snr_plt.append(snr)
print('Epoch:',j, 'Sq. Error:', av_cost,'SNR:',snr)
end = time.time()
print('Time taken',(end-strt))
sav_file = 'lms{}.npy'.format(self.epoch)
np.save(sav_file,snr_plt)
if(self.plot == True):
fig, ax = plt.subplots()
ax.plot(snr_plt, linewidth=4.0)
start, end = ax.get_ylim()
ax.yaxis.set_ticks(np.arange(start, end, 0.5))
ax.yaxis.set_major_formatter(ticker.FormatStrFormatter('%0.1f'))
fig.suptitle('SNR vs Number of Iterations while training the LMS model', fontsize=26)
plt.ylabel('SNR (in dB)', fontsize=24)
plt.xlabel('Number of iterations', fontsize=24)
for tick in ax.xaxis.get_major_ticks():
tick.label.set_fontsize(20)
for tick in ax.yaxis.get_major_ticks():
tick.label.set_fontsize(20)
plt.show()
predict = yout
print('SNR of INPUT:', init_snr)
#play_file(trainX_o,Fs)
print('SNR of OUTPUT:', self.measure_snr(predict,trainY_o[self.tap-1:-1].reshape([yout.size,1])))
#play_file(predict,Fs)
print('')
print('')
trainX_o, trainY_o = self.data_equalization(test_data, noise)
start = time.time()
trainX, trainY = self.data_preprocessing(trainX_o, trainY_o)
yout = sess.run(yhat, feed_dict = {X:trainX})
predict = yout
end = time.time()
print('Time Taken:', (end-start))
snr_test = self.measure_snr(predict ,trainY_o[self.tap-1:-1].reshape([yout.size,1]))
print('SNR of INPUT:', self.measure_snr(trainX_o, trainY_o))
#play_file(trainX_o,Fs)
print('SNR of OUTPUT:', snr_test)
#play_file(predict,Fs)
return snr, snr_test
def main():
rnnfil = RNNFilter('Mockingbird.wav', epoch = 2)
_,_ = rnnfil.driver()
if __name__ == "__main__": main()