-
Notifications
You must be signed in to change notification settings - Fork 17
/
qsr_ctc_wer.py
169 lines (139 loc) · 5.45 KB
/
qsr_ctc_wer.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
from tensorflow import keras
from tensorflow.keras.layers import Input, Lambda, BatchNormalization, Conv1D, GRU, TimeDistributed, Activation, Dense, Flatten
from tensorflow.keras.optimizers import SGD, RMSprop
from tensorflow.keras.callbacks import ModelCheckpoint
from tensorflow.keras.losses import categorical_crossentropy
import os
from tensorflow.keras import layers as L
from tensorflow.keras import backend as K
from tensorflow.keras.models import Model
import tensorflow as tf
import numpy as np
from tqdm import tqdm
import string
from models import build_asr_model
import json
os.environ["CUDA_VISIBLE_DEVICES"]="0"
labels = ['go', 'up', 'on', 'no']
characters = string.ascii_lowercase # set(char for label in labels for char in label)
# Mapping characters to integers
char_to_num = L.experimental.preprocessing.StringLookup(
vocabulary=list(characters), num_oov_indices=0, mask_token=None
)
# Mapping integers back to original characters
num_to_char = L.experimental.preprocessing.StringLookup(
vocabulary=char_to_num.get_vocabulary(), mask_token=None, invert=True
)
# test_x = tf.random.uniform([4, 60, 126, 1])
# test_y = labels
b_size = 64
MAX_word_length = 2
SAVE_PATH = "data_quantum/asr_set/"
load_asr_data = True
def get_asr_data(y_valid, y_train, x_valid, x_train, q_valid, q_train):
char_y_tr = []
char_y_val = []
new_x_tr = []
new_x_val = []
new_q_tr = []
new_q_val = []
for idx, y in enumerate(y_valid):
if labels_10[np.argmax(y)] in labels:
char_y_val.append(labels_10[np.argmax(y)])
new_x_val.append(x_valid[idx])
new_q_val.append(q_valid[idx])
for idx, y in enumerate(y_train):
if labels_10[np.argmax(y)] in labels:
char_y_tr.append(labels_10[np.argmax(y)])
new_x_tr.append(x_train[idx])
new_q_tr.append(q_train[idx])
return char_y_tr, char_y_val, new_x_tr, new_x_val, new_q_tr, new_q_val
if load_asr_data == True:
print("Load Data")
new_x_tr = np.load(SAVE_PATH + "asr_x_tr.npy")
new_x_val = np.load(SAVE_PATH + "asr_x_val.npy")
new_q_tr = np.load(SAVE_PATH + "asr_q_tr.npy")
new_q_val = np.load(SAVE_PATH + "asr_q_val.npy")
with open(SAVE_PATH + "char_y_val.json", 'r') as f:
char_y_val = json.load(f)
with open(SAVE_PATH + "char_y_tr.json", 'r') as f:
char_y_tr = json.load(f)
else:
print("Please Proc. your features")
exit()
print("-- Validation Size: ", np.array(char_y_val).shape, np.array(new_x_val).shape, np.array(new_q_val).shape)
print("-- Training Size: ", np.array(char_y_tr).shape, np.array(new_x_tr).shape, np.array(new_q_tr).shape)
# Get the model
model = build_asr_model(30, 63, 4) # 60 126 1
model.summary()
def encode_single_sample(img, label):
# 5. Transpose the image because we want the time
# dimension to correspond to the width of the image.
# img = tf.transpose(img, perm=[1, 0, 2])
# 6. Map the characters in label to numbers
label = char_to_num(tf.strings.unicode_split(label, input_encoding="UTF-8"))
# 7. Return a dict as our model is expecting two inputs
return {"speech": img, "label": label}
print("=== Making CTC input dataset ...")
train_dataset = tf.data.Dataset.from_tensor_slices((new_q_tr, char_y_tr))
train_dataset = (
train_dataset.map(
encode_single_sample, num_parallel_calls=tf.data.experimental.AUTOTUNE
)
.batch(b_size)
.prefetch(buffer_size=tf.data.experimental.AUTOTUNE)
)
validation_dataset = tf.data.Dataset.from_tensor_slices((new_q_val, char_y_val))
validation_dataset = (
validation_dataset.map(
encode_single_sample, num_parallel_calls=tf.data.experimental.AUTOTUNE
)
.batch(b_size)
.prefetch(buffer_size=tf.data.experimental.AUTOTUNE)
)
epochs = 50
early_stopping_patience = 10
# Add early stopping
early_stopping = keras.callbacks.EarlyStopping(
monitor="val_loss", patience=early_stopping_patience, restore_best_weights=True
)
history = model.fit(train_dataset,
validation_data=validation_dataset,
epochs=epochs,
callbacks=[early_stopping],)
# Get the prediction model by extracting layers till the output layer
prediction_model = keras.models.Model(
model.get_layer(name="speech").input, model.get_layer(name="dense2").output
)
# prediction_model.summary()
# A utility function to decode the output of the network
def decode_batch_predictions(pred, max_length=MAX_word_length):
input_len = np.ones(pred.shape[0]) * pred.shape[1]
# Use greedy search. For complex tasks, you can use beam search
results = keras.backend.ctc_decode(pred, input_length=input_len, greedy=True)[0][0][
:, :max_length
]
# Iterate over the results and get back the text
output_text = []
for res in results:
res = tf.strings.reduce_join(num_to_char(res)).numpy().decode("utf-8")
output_text.append(res)
return output_text
val_port = 3
pred_texts = []
for batch in validation_dataset.take(val_port):
batch_speech = batch["speech"]
preds = prediction_model.predict(batch_speech)
pred_texts.append(decode_batch_predictions(preds))
import itertools
cor_idx = 0
pred_texts = list(itertools.chain.from_iterable(pred_texts))
for idx, word in enumerate(char_y_val[0:b_size*val_port]):
if word != pred_texts[idx]:
cor_idx += 1
# else:
# if len(word) == len(pred_texts[idx]):
# pass CER
print(pred_texts)
print("=== WER:", 100*cor_idx/len(pred_texts), " %")
model.save('checkpoints/' + 'asr_ctc_demo.hdf5')