https://libeldoc.bsuir.by/handle/123456789/30406
Title: | An approach to speech ambiguities eliminating using semantically-acoustical analysis |
Authors: | Zahariev, V. A. Azarov, E. S. Rusetski, K. V. |
Keywords: | материалы конференций;speech processing;semantic technologies;acoustic analysis;semantic analysis;semantically-acoustical analysis |
Issue Date: | 2018 |
Publisher: | БГУИР |
Citation: | Zahariev, V. A. An approach to speech ambiguities eliminating using semantically-acoustical analysis / V. A. Zahariev, E. S. Azarov, K. V. Rusetski // Открытые семантические технологии проектирования интеллектуальных систем = Open Semantic Technologies for Intelligent Systems (OSTIS-2018) : материалы международной научно-технической конференции (Минск, 15 - 17 февраля 2018 года) / редкол. : В. В. Голенков (отв. ред.) [и др.]. – Минск : БГУИР, 2018. – С. 211 – 222. |
Abstract: | An approach to the problem of elimination of ambiguities in speech messages by application of semantically- acoustical analysis is presented in this paper. Authors propose the architecture of the intelligent system that implements this principle. According to this principle the direct transition from speaking to meaning of given phrase is possible with the help of digital signal processing techniques, as well as knowledge formalization methods using semantics networks (semantically- acoustical analysis). A prototype of intelligent system to resolve speech ambiguities of a certain type (homonyms and paronyms) based on the tools provided by the OSTIS technology and GUSLY signal processing framework has been implemented. The main advantages of the proposed solution in comparison to the standard automatic speech recognition systems and possible ways of further development for natural language understanding problem are also reported in this paper |
URI: | https://libeldoc.bsuir.by/handle/123456789/30406 |
Appears in Collections: | OSTIS-2018 |
File | Description | Size | Format | |
---|---|---|---|---|
Zahariev_An.PDF | 482.06 kB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.