Speech Recognition

This is program uses Convolutional Neural Networks, spectrograms, and various image processing techniques to create a software package that can create and recognize words from an "audio dictionary"

Data used for this model

The current accuracy (based on an unseen data set) tends to be around 95% for the datahandle.py implementation

Add:

• discretize the training and implementation of the neural net
• tqdm for training progress (maybe)
• implement in tensorflowRT/TFLite in order to process continuous speech
• implement in tensorflowRT in order to process continuous speech

Bugs:

• Model is using keras V1 model saving/loading protocols, review docs and update to be able to resume model training progress. Currently the callbacks have issues with saving in the ipynb version

References:

• Custom ReduceLROnPlateau Callback
• One Hot Encoding with Keras