Voice recognition methods for the development of an intelligent virtual consultant

Abstract

In this article, methods of voice recognition were studied for the development of an intelligent virtual consultant, which plays an important role as an auxiliary tool when working with business process information systems, or rather with an ERP system. This study is intended to describe an intelligent virtual consultant in assessing the usefulness and effectiveness of using real functions in a business process modeling software system for enterprise resource planning (ERP). An intelligent virtual consultant is integrated with a Web-ERP prototype. In this study, natural language processing (NLP) was based using Levenberg Marquardt (LM) and Broyden-Fletcher Goldwarb-Shanno (BFGS) methods with the Mel kepstral frequency coefficients algorithm. A comparative analysis of these methods was also carried out to determine the effectiveness of the methods used. The big advantage of this virtual consultant is the voice input of long text data fields when working in Web-ERP. This system is relevant, as it significantly expands the search for business workflows. The system allows you to convert speech into text by receiving information system (IS) instructions, which, in turn, are transmitted to an ontological database, where a query based on terms is executed to obtain a set of commands available for execution. The introduction of an intelligent virtual consultant not only expands the application's capabilities for users through the system, but also allows users to interpret data using speech synthesis. This prototype will be used for further research on the integration of intelligent virtual consultant and data analysis.