p@24: # -*- coding: utf-8 -*-
p@24: """
p@24: Created on Thu Jul 23 21:55:58 2015
p@24: 
p@24: @author: paulochiliguano
p@24: """
p@24: 
p@24: 
p@24: import tables
p@24: import numpy as np
p@24: import cPickle
p@24: import sklearn.preprocessing as preprocessing
p@24: 
p@24: #Read HDF5 file that contains log-mel spectrograms
p@24: filename = '/homes/pchilguano/msc_project/dataset/7digital/features/\
p@24: feats.h5'
p@24: with tables.openFile(filename, 'r') as f:
p@24:     features = f.root.x.read()
p@24:     #filenames = f.root.filenames.read()
p@24: 
p@24: #Pre-processing of spectrograms mean=0 and std=1
p@24: n_per_example = np.prod(features.shape[1:-1])
p@24: number_of_features = features.shape[-1]
p@24: flat_data = features.view()
p@24: flat_data.shape = (-1, number_of_features)
p@24: scaler = preprocessing.StandardScaler().fit(flat_data)
p@24: flat_data = scaler.transform(flat_data)
p@24: flat_data.shape = (features.shape[0], -1)
p@24: 
p@24: f = file('/homes/pchilguano/msc_project/dataset/7digital/features/\
p@24: feats.pkl', 'wb')
p@24: cPickle.dump(flat_data, f, protocol=cPickle.HIGHEST_PROTOCOL)
p@24: f.close()