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

skipping missing files

master
Malar Kannan 2017-11-03 15:20:31 +05:30
parent b4ceeb4eed
commit 22d353f101
1 changed files with 3 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
speech_data.py
View File

@ -6,11 +6,11 @@ import numpy as np
from spectro_gen import generate_aiff_spectrogram from spectro_gen import generate_aiff_spectrogram
from sklearn.model_selection import train_test_split from sklearn.model_selection import train_test_split
import itertools import itertools
import os
import random import random
import csv import csv
import gc import gc
def get_siamese_pairs(groupF1, groupF2): def get_siamese_pairs(groupF1, groupF2):
group1 = [r for (i, r) in groupF1.iterrows()] group1 = [r for (i, r) in groupF1.iterrows()]
group2 = [r for (i, r) in groupF2.iterrows()] group2 = [r for (i, r) in groupF2.iterrows()]
@ -84,6 +84,8 @@ def create_spectrogram_data(audio_group='audio'):
# audio_samples = audio_samples.loc[audio_samples['word'] == # audio_samples = audio_samples.loc[audio_samples['word'] ==
# 'sunflowers'].reset_index(drop=True) # 'sunflowers'].reset_index(drop=True)
file_names = audio_samples.loc[:, 'file'].apply(lambda x: 'outputs/' + audio_group + '/' + x) file_names = audio_samples.loc[:, 'file'].apply(lambda x: 'outputs/' + audio_group + '/' + x)
audio_samples['file_exists'] = apply_by_multiprocessing(file_names,os.path.exists)
audio_samples = audio_samples[audio_samples['file_exists'] == False]
audio_samples['spectrogram'] = apply_by_multiprocessing(file_names,generate_aiff_spectrogram)#.apply( audio_samples['spectrogram'] = apply_by_multiprocessing(file_names,generate_aiff_spectrogram)#.apply(
#generate_aiff_spectrogram) #generate_aiff_spectrogram)
audio_samples['window_count'] = audio_samples.loc[:,'spectrogram'].apply(lambda x: x.shape[0]) audio_samples['window_count'] = audio_samples.loc[:,'spectrogram'].apply(lambda x: x.shape[0])
@ -105,28 +107,8 @@ def create_tagged_data(audio_samples):
Y = to_onehot(Y_f.astype(np.int8)) Y = to_onehot(Y_f.astype(np.int8))
print('casting as array speech pairs') print('casting as array speech pairs')
X = np.asarray(same_data + diff_data) X = np.asarray(same_data + diff_data)
# del same_data
# del diff_data
# gc.collect()
return X,Y return X,Y
# def create_padded_spectrogram(audio_group='audio'):
# audio_samples = pd.read_pickle('outputs/{}-spectrogram.pkl'.format(audio_group))
# daf_audio_samples = ddf.from_pandas(audio_samples,npartitions=4)
# def spec_size(s):
# return s['spectrogram'].shape[0]
# max_samples = daf_audio_samples.apply(spec_size,axis=1, meta=('x', 'i8')).max().compute()
# print('max sample count is ',max_samples)
# def padd_zeros_fixed(sp):
# x = sp['spectrogram']
# bounds = [(0, max_samples - x.shape[0]), (0, 0)]
# sp['spectrogram'] = np.lib.pad(x,bounds,'constant')
# return sp
# padded_audio_samples = daf_audio_samples.apply(padd_zeros_fixed,axis=1, meta=audio_samples).compute()#,max_samples=max_samples)
# import pdb; pdb.set_trace()
# padded_spectrogram = np.asarray(padded_audio_samples['spectrogram'].tolist())
# np.save('outputs/{}-padded_spectrogram.npy'.format(audio_group),padded_spectrogram)
def create_speech_pairs_data(audio_group='audio'): def create_speech_pairs_data(audio_group='audio'):
audio_samples = pd.read_pickle('outputs/{}-spectrogram.pkl'.format(audio_group)) audio_samples = pd.read_pickle('outputs/{}-spectrogram.pkl'.format(audio_group))
# sample_size = audio_samples['spectrogram'][0].shape[1] # sample_size = audio_samples['spectrogram'][0].shape[1]
@ -139,26 +121,11 @@ def create_speech_pairs_data(audio_group='audio'):
np.random.shuffle(X) np.random.shuffle(X)
np.random.set_state(rng_state) np.random.set_state(rng_state)
np.random.shuffle(Y) np.random.shuffle(Y)
# p = np.random.permutation(len(X))
# X = X_i[p]
# Y = Y_i[p]
print('pickling X/Y') print('pickling X/Y')
np.save('outputs/{}-train-X.npy'.format(audio_group), X) np.save('outputs/{}-train-X.npy'.format(audio_group), X)
np.save('outputs/{}-train-Y.npy'.format(audio_group), Y) np.save('outputs/{}-train-Y.npy'.format(audio_group), Y)
save_samples_for('train',tr_audio_samples) save_samples_for('train',tr_audio_samples)
save_samples_for('test',te_audio_samples) save_samples_for('test',te_audio_samples)
# print('generating test siamese speech pairs ')
# X,Y = create_tagged_data(te_audio_samples,max_samples)
# print('pickling X/Y')
# np.save('outputs/{}-test-X.npy'.format(audio_group), X)
# np.save('outputs/{}-test-Y.npy'.format(audio_group), Y)
# # print('train/test splitting speech pairs')
# # tr_pairs, te_pairs, tr_y, te_y = train_test_split(X, Y, test_size=0.1)
# # print('pickling train/test')
# # np.save('outputs/{}-tr_pairs.npy'.format(audio_group), tr_pairs)
# # np.save('outputs/{}-te_pairs.npy'.format(audio_group), te_pairs)
# # np.save('outputs/{}-tr_y.npy'.format(audio_group), tr_y)
# # np.save('outputs/{}-te_y.npy'.format(audio_group), te_y)
def speech_data(audio_group='audio'): def speech_data(audio_group='audio'):
X = np.load('outputs/{}-X.npy'.format(audio_group)) / 255.0 X = np.load('outputs/{}-X.npy'.format(audio_group)) / 255.0