Creating a voice assistant using neural network technologies

Abstract

The article explores the development of a voice assistant using neural network technologies. The main purpose of the work is to demonstrate the possibility of using neural network solutions to create effective voice assistant programs. In the course of the research, a Python program was created, which was tested for speed and accuracy. The program is capable of performing a wide range of tasks, including team personalization, creation of complex algorithms, integration with Smart Home systems and ensuring a high level of privacy through local data processing. The problem of interpreting commands is solved by using the algorithm of fuzzy comparison. Algorithms for finding the Levenshtein and Damerau — Levenshtein distances, as well as the Jaro — Winkler similarity, are implemented for accurate recognition of naked queries. Their effectiveness in terms of work accuracy and optimization is compared. The program interface, created using the Custom Tkinter library, is described, and the general principle of operation of the voice assistant is presented. The VOSK and Silero TTS libraries based on neural networks are used in the key elements of the program — voice input and speech synthesis. Statistics on the use of the program were compiled and an assessment of the speed and accuracy of its key components was carried out, conclusions were drawn about the prospects for using a solution based on neural networks in application development.