Speech and voice recognition system based on machine learning methods

Abstract

Person’s unique biometric identity can be used to distinguish different people and to augment and upgrade the current regular PIN and password systems for gaining access to computers, phones, or restricted access rooms and buildings. The USA banking app uses voice recognition to provide easy and secure multi-factor biometric security, the voice component adding an extra level of liveness detection to the process. The Recurrent Neural Network allows for a bi-directional flow of data which is especially useful for purpose of speech recognition. In our work, we use TensorFlow for machine learning. TensorFlow allows building neural network models to recognize spoken words. Module development includes 4 stages: create (or use a ready-made one) dataset; build NN; train the NN; test NN.