malar
/
speech-scoring
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

clean-up

master
Malar Kannan 2017-11-08 11:08:19 +05:30
parent 7cbfebbf1a
commit b3a6aa2f6a
2 changed files with 3 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
speech_data.py
View File

@ -11,12 +11,8 @@ import os
import random import random
import csv import csv
import gc import gc
# import progressbar
from tqdm import tqdm from tqdm import tqdm
# def prog_bar(title):
# widgets = [title, progressbar.Counter(), ' [', progressbar.Bar(), '] - ', progressbar.ETA()]
# return progressbar.ProgressBar(widgets=widgets)
def siamese_pairs(rightGroup, wrongGroup): def siamese_pairs(rightGroup, wrongGroup):
group1 = [r for (i, r) in rightGroup.iterrows()] group1 = [r for (i, r) in rightGroup.iterrows()]
@ -25,10 +21,7 @@ def siamese_pairs(rightGroup, wrongGroup):
rightRightPairs = [i for i in itertools.combinations(group1, 2)] rightRightPairs = [i for i in itertools.combinations(group1, 2)]
random.shuffle(rightWrongPairs) random.shuffle(rightWrongPairs)
random.shuffle(rightRightPairs) random.shuffle(rightRightPairs)
# return (random.sample(same,10), random.sample(diff,10))
# return rightRightPairs[:10],rightWrongPairs[:10]
return rightRightPairs[:32],rightWrongPairs[:32] return rightRightPairs[:32],rightWrongPairs[:32]
# return rightRightPairs,rightWrongPairs
def create_spectrogram_tfrecords(audio_group='audio'): def create_spectrogram_tfrecords(audio_group='audio'):
''' '''
@ -38,14 +31,10 @@ def create_spectrogram_tfrecords(audio_group='audio'):
audio_samples = pd.read_csv( './outputs/' + audio_group + '.csv' audio_samples = pd.read_csv( './outputs/' + audio_group + '.csv'
, names=['word','phonemes', 'voice', 'language', 'rate', 'variant', 'file'] , names=['word','phonemes', 'voice', 'language', 'rate', 'variant', 'file']
, quoting=csv.QUOTE_NONE) , quoting=csv.QUOTE_NONE)
# audio_samples = audio_samples.loc[audio_samples['word'] ==
# 'sunflowers'].reset_index(drop=True)
audio_samples['file_path'] = audio_samples.loc[:, 'file'].apply(lambda x: 'outputs/' + audio_group + '/' + x) audio_samples['file_path'] = audio_samples.loc[:, 'file'].apply(lambda x: 'outputs/' + audio_group + '/' + x)
audio_samples['file_exists'] = apply_by_multiprocessing(audio_samples['file_path'], os.path.exists) audio_samples['file_exists'] = apply_by_multiprocessing(audio_samples['file_path'], os.path.exists)
audio_samples = audio_samples[audio_samples['file_exists'] == True].reset_index() audio_samples = audio_samples[audio_samples['file_exists'] == True].reset_index()
# audio_samples['rate_int'] = apply_by_multiprocessing(audio_samples['rate'], str.isdigit)
# audio_samples = audio_samples[audio_samples['rate_int'] == True].reset_index().drop(['level_0'],axis=1)
# audio_samples['rate'] = audio_samples['rate'].astype(int)
def _float_feature(value): def _float_feature(value):
return tf.train.Feature(float_list=tf.train.FloatList(value=value)) return tf.train.Feature(float_list=tf.train.FloatList(value=value))
@ -99,6 +88,7 @@ def create_spectrogram_tfrecords(audio_group='audio'):
} }
)) ))
writer.write(example.SerializeToString()) writer.write(example.SerializeToString())
group_prog.close()
prog.close() prog.close()
writer.close() writer.close()
@ -120,7 +110,6 @@ def find_max_n(trf):
def read_siamese_tfrecords(audio_group='audio'): def read_siamese_tfrecords(audio_group='audio'):
records_file = os.path.join('./outputs',audio_group+'.tfrecords') records_file = os.path.join('./outputs',audio_group+'.tfrecords')
record_iterator = tf.python_io.tf_record_iterator(path=records_file) record_iterator = tf.python_io.tf_record_iterator(path=records_file)
# input1,input2 = [],[]
input_pairs = [] input_pairs = []
output_class = [] output_class = []
max_n = find_max_n(records_file) max_n = find_max_n(records_file)
@ -128,8 +117,6 @@ def read_siamese_tfrecords(audio_group='audio'):
for string_record in record_iterator: for string_record in record_iterator:
example = tf.train.Example() example = tf.train.Example()
example.ParseFromString(string_record) example.ParseFromString(string_record)
# word = example.features.feature['word'].bytes_list.value[0]
# input_words.append(word)
example.features.feature['spec2'].float_list.value[0] example.features.feature['spec2'].float_list.value[0]
spec_n1 = example.features.feature['spec_n1'].int64_list.value[0] spec_n1 = example.features.feature['spec_n1'].int64_list.value[0]
spec_n2 = example.features.feature['spec_n2'].int64_list.value[0] spec_n2 = example.features.feature['spec_n2'].int64_list.value[0]
@ -138,20 +125,12 @@ def read_siamese_tfrecords(audio_group='audio'):
spec1 = np.array(example.features.feature['spec1'].float_list.value).reshape(spec_n1,spec_w1) spec1 = np.array(example.features.feature['spec1'].float_list.value).reshape(spec_n1,spec_w1)
spec2 = np.array(example.features.feature['spec2'].float_list.value).reshape(spec_n2,spec_w2) spec2 = np.array(example.features.feature['spec2'].float_list.value).reshape(spec_n2,spec_w2)
p_spec1,p_spec2 = padd_zeros(spec1,max_n),padd_zeros(spec2,max_n) p_spec1,p_spec2 = padd_zeros(spec1,max_n),padd_zeros(spec2,max_n)
# input1.append(spec1)
# input2.append(spec2)
input_pairs.append(np.asarray([p_spec1,p_spec2])) input_pairs.append(np.asarray([p_spec1,p_spec2]))
# input_pairs.append([spec1,spec2])
output = example.features.feature['output'].int64_list.value output = example.features.feature['output'].int64_list.value
output_class.append(np.asarray(output)) output_class.append(np.asarray(output))
n_features = spec_w1 n_features = spec_w1
# if len(input_pairs) > 50:
# break
input_data,output_data = np.asarray(input_pairs),np.asarray(output_class) input_data,output_data = np.asarray(input_pairs),np.asarray(output_class)
import pdb; pdb.set_trace()
# tr_x1,te_x1,tr_x2,te_x2,tr_y,te_y = train_test_split(input1,input2,output_class)
tr_pairs,te_pairs,tr_y,te_y = train_test_split(input_data,output_data) tr_pairs,te_pairs,tr_y,te_y = train_test_split(input_data,output_data)
# return (tr_x1,te_x1,tr_x2,te_x2,tr_y,te_y)
n_step,n_features = int(max_n),int(spec_w1) n_step,n_features = int(max_n),int(spec_w1)
return (tr_pairs,te_pairs,tr_y,te_y,n_step,n_features) return (tr_pairs,te_pairs,tr_y,te_y,n_step,n_features)
@ -160,8 +139,6 @@ def audio_samples_word_count(audio_group='audio'):
audio_samples = pd.read_csv( './outputs/' + audio_group + '.csv' audio_samples = pd.read_csv( './outputs/' + audio_group + '.csv'
, names=['word','phonemes', 'voice', 'language', 'rate', 'variant', 'file'] , names=['word','phonemes', 'voice', 'language', 'rate', 'variant', 'file']
, quoting=csv.QUOTE_NONE) , quoting=csv.QUOTE_NONE)
# audio_samples = audio_samples.loc[audio_samples['word'] ==
# 'sunflowers'].reset_index(drop=True)
audio_samples['file_path'] = audio_samples.loc[:, 'file'].apply(lambda x: 'outputs/' + audio_group + '/' + x) audio_samples['file_path'] = audio_samples.loc[:, 'file'].apply(lambda x: 'outputs/' + audio_group + '/' + x)
audio_samples['file_exists'] = apply_by_multiprocessing(audio_samples['file_path'], os.path.exists) audio_samples['file_exists'] = apply_by_multiprocessing(audio_samples['file_path'], os.path.exists)
audio_samples = audio_samples[audio_samples['file_exists'] == True].reset_index() audio_samples = audio_samples[audio_samples['file_exists'] == True].reset_index()
@ -170,19 +147,6 @@ def audio_samples_word_count(audio_group='audio'):
def fix_csv(audio_group='audio'): def fix_csv(audio_group='audio'):
audio_csv_lines = open('./outputs/' + audio_group + '.csv','r').readlines() audio_csv_lines = open('./outputs/' + audio_group + '.csv','r').readlines()
audio_csv_data = [i.strip().split(',') for i in audio_csv_lines] audio_csv_data = [i.strip().split(',') for i in audio_csv_lines]
# audio_samples = pd.read_csv( './outputs/story_words.csv'
# , names=['word','phonemes', 'voice', 'language', 'rate', 'variant', 'file']
# , quoting=csv.QUOTE_NONE)
# voice_set = set(audio_samples['voice'].unique().tolist())
# to_be_fixed = [i for i in audio_csv_data if len(i) > 7]
# def unite_words(entries):
# entries = to_be_fixed[0]
# word_entries = next(((entries[:i],entries[i:]) for (i,e) in enumerate(entries) if e in voice_set),'')
# word_entries[1]
# return
# to_be_fixed[0]
# entries = [unite_words for e in to_be_fixed]
# [i for i in entries if len(i) % 2 != 0]
proper_rows = [i for i in audio_csv_data if len(i) == 7] proper_rows = [i for i in audio_csv_data if len(i) == 7]
with open('./outputs/' + audio_group + '-new.csv','w') as fixed_csv: with open('./outputs/' + audio_group + '-new.csv','w') as fixed_csv:
fixed_csv_w = csv.writer(fixed_csv, quoting=csv.QUOTE_MINIMAL) fixed_csv_w = csv.writer(fixed_csv, quoting=csv.QUOTE_MINIMAL)

2
speech_siamese.py
View File

@ -82,7 +82,7 @@ def siamese_model(input_dim):
def train_siamese(): def train_siamese():
# the data, shuffled and split between train and test sets # the data, shuffled and split between train and test sets
# tr_pairs, te_pairs, tr_y_e, te_y_e = speech_model_data() # tr_pairs, te_pairs, tr_y_e, te_y_e = speech_model_data()
(tr_pairs,te_pairs,tr_y,te_y,n_step,n_features) = read_siamese_tfrecords('story_words') (tr_pairs,te_pairs,tr_y,te_y,n_step,n_features) = read_siamese_tfrecords('story_words_test')
# tr_y = to_categorical(tr_y_e, num_classes=2) # tr_y = to_categorical(tr_y_e, num_classes=2)
# te_y = to_categorical(te_y_e, num_classes=2) # te_y = to_categorical(te_y_e, num_classes=2)
input_dim = (n_step, n_features) input_dim = (n_step, n_features)