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

Merge branch 'master' of ssh://invnuc/~/Public/Repos/speech_scoring

master
Malar Kannan 2017-11-28 17:05:35 +05:30
parent ec7303223c 20b2d7a958
commit 1928fce4e8
4 changed files with 15 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.gitignore vendored
View File

@ -143,3 +143,5 @@ inputs/audio*
logs/* logs/*
models/* models/*
*.pkl *.pkl
temp/*
trained/*

19
speech_data.py
View File

@ -147,7 +147,7 @@ def read_siamese_tfrecords_generator(audio_group='audio',batch_size=32,test_size
(n_spec,n_features,n_records) = pickle.load(open(const_file,'rb')) (n_spec,n_features,n_records) = pickle.load(open(const_file,'rb'))
def copy_read_consts(dest_dir): def copy_read_consts(dest_dir):
shutil.copy2(const_file,dest_dir) shutil.copy2(const_file,dest_dir+'/constants.pkl')
return (n_spec,n_features,n_records) return (n_spec,n_features,n_records)
# @threadsafe_iter # @threadsafe_iter
def record_generator(): def record_generator():
@ -210,15 +210,14 @@ def record_generator_count(records_file):
record_iterator = tf.python_io.tf_record_iterator(path=records_file) record_iterator = tf.python_io.tf_record_iterator(path=records_file)
count,spec_n = 0,0 count,spec_n = 0,0
for i in record_iterator: for i in record_iterator:
example = tf.train.Example() # example = tf.train.Example()
example.ParseFromString(i) # example.ParseFromString(i)
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]
spec_n = max([spec_n,spec_n1,spec_n2]) # spec_n = max([spec_n,spec_n1,spec_n2])
import pdb; pdb.set_trace()
count+=1 count+=1
record_iterator = tf.python_io.tf_record_iterator(path=records_file) record_iterator = tf.python_io.tf_record_iterator(path=records_file)
return record_iterator,count,spec_n return record_iterator,count #,spec_n
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()
@ -260,8 +259,8 @@ if __name__ == '__main__':
# create_spectrogram_tfrecords('story_all',sample_count=25) # create_spectrogram_tfrecords('story_all',sample_count=25)
# fix_csv('story_words_test') # fix_csv('story_words_test')
#fix_csv('audio') #fix_csv('audio')
# create_spectrogram_tfrecords('story_words_test',sample_count=100,train_test_ratio=0.1) create_spectrogram_tfrecords('story_words_test',sample_count=10,train_test_ratio=0.1)
record_generator_count() #record_generator_count()
# create_spectrogram_tfrecords('audio',sample_count=50) # create_spectrogram_tfrecords('audio',sample_count=50)
# read_siamese_tfrecords_generator('audio') # read_siamese_tfrecords_generator('audio')
# padd_zeros_siamese_tfrecords('audio') # padd_zeros_siamese_tfrecords('audio')

4
speech_model.py
View File

@ -114,7 +114,7 @@ def train_siamese(audio_group = 'audio'):
, epochs=1000 , epochs=1000
, steps_per_epoch=epoch_n_steps , steps_per_epoch=epoch_n_steps
, validation_data=([te_pairs[:, 0], te_pairs[:, 1]], te_y) , validation_data=([te_pairs[:, 0], te_pairs[:, 1]], te_y)
, max_queue_size=32 , max_queue_size=8
, callbacks=[tb_cb, cp_cb]) , callbacks=[tb_cb, cp_cb])
model.save(model_dir+'/siamese_speech_model-final.h5') model.save(model_dir+'/siamese_speech_model-final.h5')
@ -125,4 +125,4 @@ def train_siamese(audio_group = 'audio'):
if __name__ == '__main__': if __name__ == '__main__':
train_siamese('story_phrases') train_siamese('story_words_test')

5
speech_test.py
View File

@ -29,7 +29,7 @@ def test_with(audio_group):
def evaluate_siamese(records_file,audio_group='audio',weights = 'siamese_speech_model-final.h5'): def evaluate_siamese(records_file,audio_group='audio',weights = 'siamese_speech_model-final.h5'):
# audio_group='audio';model_file = 'siamese_speech_model-305-epoch-0.20-acc.h5' # audio_group='audio';model_file = 'siamese_speech_model-305-epoch-0.20-acc.h5'
# records_file = os.path.join('./outputs',eval_group+'.train.tfrecords') # records_file = os.path.join('./outputs',eval_group+'.train.tfrecords')
const_file = os.path.join('./models/'+audio_group+'/',audio_group+'.constants') const_file = os.path.join('./models/'+audio_group+'/','constants.pkl')
arch_file='./models/'+audio_group+'/siamese_speech_model_arch.yaml' arch_file='./models/'+audio_group+'/siamese_speech_model_arch.yaml'
weight_file='./models/'+audio_group+'/'+weights weight_file='./models/'+audio_group+'/'+weights
(n_spec,n_features,n_records) = pickle.load(open(const_file,'rb')) (n_spec,n_features,n_records) = pickle.load(open(const_file,'rb'))
@ -41,7 +41,6 @@ def evaluate_siamese(records_file,audio_group='audio',weights = 'siamese_speech_
total,same_success,diff_success,skipped,same_failed,diff_failed = 0,0,0,0,0,0 total,same_success,diff_success,skipped,same_failed,diff_failed = 0,0,0,0,0,0
all_results = [] all_results = []
for (i,string_record) in tqdm(enumerate(record_iterator),total=records_count): for (i,string_record) in tqdm(enumerate(record_iterator),total=records_count):
# string_record = next(record_iterator)
total+=1 total+=1
example = tf.train.Example() example = tf.train.Example()
example.ParseFromString(string_record) example.ParseFromString(string_record)
@ -178,7 +177,7 @@ def visualize_results(audio_group='audio'):
if __name__ == '__main__': 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.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.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.gpu',weights ='siamese_speech_model-58-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')
# play_results('story_words') # play_results('story_words')
#inspect_tfrecord('./outputs/story_phrases.test.tfrecords',audio_group='story_phrases') #inspect_tfrecord('./outputs/story_phrases.test.tfrecords',audio_group='story_phrases')
# visualize_results('story_words.gpu') # visualize_results('story_words.gpu')