From a7f1451a7f69fd0a6160dd23344145ed47a32390 Mon Sep 17 00:00:00 2001 From: Malar Kannan Date: Thu, 7 Dec 2017 16:49:34 +0530 Subject: [PATCH] fixed exception in data generation --- segment_data.py | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) diff --git a/segment_data.py b/segment_data.py index b6622b0..de86d61 100644 --- a/segment_data.py +++ b/segment_data.py @@ -113,7 +113,9 @@ def plot_segments(collection_name = 'story_test_segments'): def generate_spec(aiff_file): phrase_sample = pm_snd(aiff_file) phrase_spec = phrase_sample.to_spectrogram(window_length=SPEC_WINDOW_SIZE, maximum_frequency=SPEC_MAX_FREQUENCY) - sg_db = 10 * np.log10(phrase_spec.values) + sshow_abs = np.abs(phrase_spec.values + np.finfo(phrase_spec.values.dtype).eps) + sg_db = 10 * np.log10(sshow_abs) + sg_db[sg_db < 0] = 0 return sg_db,phrase_spec @@ -142,6 +144,7 @@ def create_segments_tfrecords(collection_name='story_test_segments',sample_count fname = g.iloc[0]['filename'] sg_db,phrase_spec = generate_spec(g.iloc[0]['file_path']) phon_stops = [] + phrase_groups.set_postfix(phrase=ph) spec_n,spec_w = sg_db.shape spec = sg_db.reshape(-1) for (i,phon) in g.iterrows(): @@ -177,7 +180,7 @@ def create_segments_tfrecords(collection_name='story_test_segments',sample_count wg_sampled = reservoir_sample(word_groups,sample_count) if sample_count > 0 else word_groups tr_audio_samples,te_audio_samples = train_test_split(wg_sampled,test_size=train_test_ratio) write_samples(tr_audio_samples,'train') - write_samples(te_audio_samples,'test') + # write_samples(te_audio_samples,'test') const_file = './outputs/segments/'+collection_name+'/constants.pkl' pickle.dump((n_spec,n_features,n_records),open(const_file,'wb')) @@ -235,10 +238,8 @@ def read_segments_tfrecords_generator(collection_name='audio',batch_size=32,test random_samples = enumerate(reservoir_sample(te_re_iterator,test_size)) for (i,string_record) in tqdm(random_samples,total=test_size): # (i,string_record) = next(random_samples) - # string_record example = tf.train.Example() example.ParseFromString(string_record) - # example.features.feature['spec'].float_list.value spec_n = example.features.feature['spec_n'].int64_list.value[0] spec_w = example.features.feature['spec_w'].int64_list.value[0] spec = np.array(example.features.feature['spec'].float_list.value).reshape(spec_n,spec_w) @@ -254,9 +255,9 @@ if __name__ == '__main__': # plot_random_phrases() # fix_csv('story_test_segments') # plot_segments('story_test_segments') - # fix_csv('story_test') - pass - # create_segments_tfrecords('story_test') + # fix_csv('story_phrases') + # pass + create_segments_tfrecords('story_phrases', sample_count=1000) # record_generator,input_data,output_data,copy_read_consts = read_segments_tfrecords_generator('story_test') # tr_gen = record_generator() # for i in tr_gen: