Python Deep Learning Projects

Data transformation

Once we have all the features that we need to feed into the model, we will transform the raw NumPy tensors into the TensorFlow specific format called TFRecords. 

In the following code snippet, we are creating the folders to store all the processed records. The make_example() function creates the sequence example for a single utterance given the sequence length, MFCC features, and corresponding transcript. Multiple sequence records are then written into TFRecord files using the tf.python_io.TFRecordWriter() function:

if os.path.basename(partition) == 'train-clean-100':    # Create multiple TFRecords based on utterance length for training    writer = {}    count = {}    print('Processing training files...') for i in range(min_t, max_t+1): ...