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

Merge branch 'master' of ssh://gpuaws/~/repos/speech_scoring

master
Malar Kannan 2017-12-06 17:33:27 +05:30
parent 8e14db2437 3f76207f0d
commit c0369d7a66
3 changed files with 12 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
speech_data.py
View File

@ -6,6 +6,7 @@ import tensorflow as tf
from tensorflow.python.ops import data_flow_ops
import numpy as np
from speech_spectrum import generate_aiff_spectrogram
from speech_pitch import pitch_array
from speech_pitch import compute_mfcc
from sklearn.model_selection import train_test_split
import itertools
@ -20,7 +21,7 @@ from tqdm import tqdm
def siamese_pairs(rightGroup, wrongGroup):
group1 = [r for (i, r) in rightGroup.iterrows()]
group2 = [r for (i, r) in wrongGroup.iterrows()]
rightWrongPairs = [(g1, g2) for g2 in group2 for g1 in group1]#+[(g2, g1) for g2 in group2 for g1 in group1]
rightWrongPairs = [(g1, g2) for g2 in group2 for g1 in group1]+[(g2, g1) for g2 in group2 for g1 in group1]
rightRightPairs = [i for i in itertools.permutations(group1, 2)]#+[i for i in itertools.combinations(group2, 2)]
def filter_criteria(s1,s2):
same = s1['variant'] == s2['variant']
@ -36,7 +37,7 @@ def siamese_pairs(rightGroup, wrongGroup):
random.shuffle(validRWPairs)
random.shuffle(validRRPairs)
# return rightRightPairs[:10],rightWrongPairs[:10]
return validRWPairs[:32],validRRPairs[:32]
return validRRPairs[:32],validRWPairs[:32]
def _float_feature(value):
@ -64,7 +65,8 @@ def create_spectrogram_tfrecords(audio_group='audio',sample_count=0,train_test_r
for (w, word_group) in word_group_prog:
word_group_prog.set_postfix(word=w,sample_name=sample_name)
g = word_group.reset_index()
g['spectrogram'] = apply_by_multiprocessing(g['file_path'],generate_aiff_spectrogram)
g['spectrogram'] = apply_by_multiprocessing(g['file_path'],pitch_array)
# g['spectrogram'] = apply_by_multiprocessing(g['file_path'],generate_aiff_spectrogram)
# g['spectrogram'] = apply_by_multiprocessing(g['file_path'],compute_mfcc)
sample_right = g.loc[g['variant'] == 'low']
sample_wrong = g.loc[g['variant'] == 'medium']
@ -244,7 +246,7 @@ if __name__ == '__main__':
# create_spectrogram_tfrecords('story_all',sample_count=25)
# fix_csv('story_words_test')
#fix_csv('audio')
create_spectrogram_tfrecords('story_words_test',sample_count=10,train_test_ratio=0.1)
create_spectrogram_tfrecords('story_words_pitch',sample_count=0,train_test_ratio=0.1)
#record_generator_count()
# create_spectrogram_tfrecords('audio',sample_count=50)
# read_siamese_tfrecords_generator('audio')

8
speech_model.py
View File

@ -46,10 +46,12 @@ def dense_classifier(processed):
conc_proc = Concatenate()(processed)
d1 = Dense(64, activation='relu')(conc_proc)
# dr1 = Dropout(0.1)(d1)
bn_d1 = BatchNormalization(momentum=0.98)(d1)
# d2 = Dense(128, activation='relu')(d1)
d3 = Dense(8, activation='relu')(d1)
d3 = Dense(8, activation='relu')(bn_d1)
bn_d3 = BatchNormalization(momentum=0.98)(d3)
# dr2 = Dropout(0.1)(d2)
return Dense(2, activation='softmax')(d3)
return Dense(2, activation='softmax')(bn_d3)
def siamese_model(input_dim):
base_network = create_base_rnn_network(input_dim)
@ -127,4 +129,4 @@ def train_siamese(audio_group = 'audio'):
if __name__ == '__main__':
train_siamese('story_words_test')
train_siamese('story_words_pitch')

2
speech_test.py
View File

@ -177,7 +177,7 @@ def visualize_results(audio_group='audio'):
if __name__ == '__main__':
# evaluate_siamese('./outputs/story_words_test.train.tfrecords',audio_group='story_words.gpu',weights ='siamese_speech_model-58-epoch-0.00-acc.h5')
# evaluate_siamese('./outputs/story_words.test.tfrecords',audio_group='story_words',weights ='siamese_speech_model-675-epoch-0.00-acc.h5')
evaluate_siamese('./outputs/story_words_test.train.tfrecords',audio_group='story_words_test.10',weights ='siamese_speech_model-891-epoch-0.02-acc.h5')
evaluate_siamese('./outputs/story_words_pitch.test.tfrecords',audio_group='story_words_pitch',weights ='siamese_speech_model-867-epoch-0.12-acc.h5')
# play_results('story_words')
#inspect_tfrecord('./outputs/story_phrases.test.tfrecords',audio_group='story_phrases')
# visualize_results('story_words.gpu')