diff --git a/speech_data.py b/speech_data.py
index c48950d..b254bee 100644
--- a/speech_data.py
+++ b/speech_data.py
@@ -1,7 +1,8 @@
 import pandas as pd
 from pandas_parallel import apply_by_multiprocessing
-import dask as dd
-import dask.dataframe as ddf
+# import dask as dd
+# import dask.dataframe as ddf
+import tensorflow as tf
 import numpy as np
 from spectro_gen import generate_aiff_spectrogram
 from sklearn.model_selection import train_test_split
@@ -91,13 +92,14 @@ def create_spectrogram_data(audio_group='audio'):
     audio_samples.to_pickle('outputs/{}-spectrogram.pkl'.format(audio_group))
 
 def create_spectrogram_tfrecords(audio_group='audio'):
+    # audio_group = 'story_words'
     audio_samples = pd.read_csv( './outputs/' + audio_group + '.csv'
         , names=['word','phonemes', 'voice', 'language', 'rate', 'variant', 'file']
         , quoting=csv.QUOTE_NONE)
     # audio_samples = audio_samples.loc[audio_samples['word'] ==
     #                                   'sunflowers'].reset_index(drop=True)
-    audio_samples['file_paths'] = audio_samples.loc[:, 'file'].apply(lambda x: 'outputs/' + audio_group + '/' + x)
-    audio_samples['file_exists'] = apply_by_multiprocessing(audio_samples['file_paths'], os.path.exists)
+    audio_samples['file_path'] = audio_samples.loc[:, 'file'].apply(lambda x: 'outputs/' + audio_group + '/' + x)
+    audio_samples['file_exists'] = apply_by_multiprocessing(audio_samples['file_path'], os.path.exists)
     audio_samples = audio_samples[audio_samples['file_exists'] == True].reset_index()
     # audio_samples['spectrogram'] = apply_by_multiprocessing(audio_samples['file_paths'],generate_aiff_spectrogram)#.apply(
     # audio_samples['window_count'] = audio_samples.loc[:,'spectrogram'].apply(lambda x: x.shape[0])
@@ -111,17 +113,29 @@ def create_spectrogram_tfrecords(audio_group='audio'):
 
     def _bytes_feature(value):
       return tf.train.Feature(bytes_list=tf.train.BytesList(value=value))
-    writer = tf.python_io.TFRecordWriter(output_path)
 
-    for sample in audio_samples:
+    writer = tf.python_io.TFRecordWriter('./outputs/' + audio_group + '.tfrecords')
+    audio_samples = audio_samples[:100]
+    for (i,sample) in audio_samples.iterrows():
+        spectrogram = generate_aiff_spectrogram(sample['file_path'])
+        spec_n = spectrogram.shape[0]
+        spec_w = spectrogram.shape[1]
+        spec = spectrogram.reshape(-1)
         example = tf.train.Example(features=tf.train.Features(
           feature={
-            'label': _int64_feature([label]),
-            'path': _bytes_feature([image_path]),
-            'instance' : _bytes_feature([instance_id])
+            'word': _bytes_feature([sample['word'].encode('utf-8')]),
+            'phoneme': _bytes_feature([sample['phonemes'].encode('utf-8')]),
+            'voice': _bytes_feature([sample['voice'].encode('utf-8')]),
+            'language': _bytes_feature([sample['language'].encode('utf-8')]),
+            'rate':_int64_feature([sample['rate']]),
+            'variant': _bytes_feature([sample['variant'].encode('utf-8')]),
+            'file': _bytes_feature([sample['file'].encode('utf-8')]),
+            'spec':_float_feature(spec),
+            'spec_n':_int64_feature([spec_n]),
+            'spec_w':_int64_feature([spec_w])
           }
         ))
-    writer.write(example.SerializeToString())
+        writer.write(example.SerializeToString())
     writer.close()
 
 def create_tagged_data(audio_samples):
@@ -177,7 +191,8 @@ def speech_model_data():
 if __name__ == '__main__':
     # sunflower_pairs_data()
     # create_spectrogram_data()
-    create_spectrogram_data('story_words')
+    # create_spectrogram_data('story_words')
+    create_spectrogram_tfrecords('story_words')
     # create_padded_spectrogram()
     # create_speech_pairs_data()
     # print(speech_model_data())