Využití metod strojového učení pro rozpoznávání řeči

Abstract

This thesis is devoted to machine learning methods for speech recognition. The first part deals with teoretical description of methods for speech signal processing and algorithms which can be used for automatic speech recognition. Dynamic time warping, hidden Markov models and deep neural networks are described here. The practical part is focused on the description of the created system, which is based on convolutional neural networks. This system was designed and implemented in Python using Keras and TensorFlow. A~dataset of 15 words was used for training and testing. The use of the system is possible in various areas of Industry 4.0.