Semantic analysis of voice messages based on a formalized context

Abstract

The report is devoted to the problem of using formalized contextual information for the semantic analysis of voice messages in conversational systems with a speech interface. The paper proposes an approach based on several fundamental principles: the transition from the acoustic pattern to the semantic representation bypassing a separate stage of textual presentation of information, saving and processing contextual information at all levels in a single knowledge base, transferring the linguistic processing stage to the semantic analysis block (which allows to take into account not only statistical but also semantic links at this level), of applying feedback from the semantic level to the lower level to adjust the result of links work. To implement the approach, the original signal processing technique based on instantaneous harmonic analysis, convolutional neural networks for solving the classification problem, as well as the capabilities of the OSTIS methodology and technology were used.