Deep learning based speech recognition system design

Natural language processing consists of research and applications on how computers can understand and manipulate natural writing or spoken language. Speech processing is a sub-field of natural language processing that includes speech signals and signal processing methods. Speech signals are mostly processed through digital representations and translated into written language with different methods. This process, which usually consists of training and testing phases, includes training the model using the labeled data at hand and measuring the consistency of the trained model with different labeled data. Throughout history, many researchers have developed different approaches and methods to translate spoken language into writing. Today, online speech recognition models developed by private companies are used in many areas of work. These developed models are realized by using Hidden Markov Model (HMM), artificial neural networks, algorithms used for noise removal, deep learning algorithms and phoneme dictionaries together. The use of these models is increasing day by day in various fields such as smart home systems, automotive, military and health. The models used are mostly online, do not allow new developments by the user, and there are still many aspects that need to be improved due to insufficient language support. In this thesis, two different natural language-to-text transformation models have been created. As an alternative to traditional methods, the first model uses end-to-end deep learning method with less processing load and complexity; The second one was carried out following a pre-processed course with traditional methods. The success of these models in different conditions such as speaker addiction, data set size, and duration of education was tried to be determined. In addition, a network-based software has been developed to collect labeled data from users in order to create the necessary data set for the training and testing stages of both models. Key words: Speech recognition, deep learning, speech-to-text conversion, signal processing, natural language processing