Spoken term detection system with high retrieval accuracy, high speed and small resources using Deep Neural Network

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K00241
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Perceptual information processing
• Research Institution: Iwate Prefectural University
• Principal Investigator: Yoshiaki Itoh 岩手県立大学, ソフトウェア情報学部, 教授 (90325928)
• Co-Investigator(Kenkyū-buntansha): 李 時旭 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 主任研究員 (50415642)
• Co-Investigator(Renkei-kenkyūsha): Ogura Kanayo 岩手県立大学, ソフトウェア情報学部, 講師 (10432139)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 音声言語処理 / 音声検索

Outline of Final Research Achievements

This research aims the realization of high retrieval accuracy, speed up and small resources for spoken term detection among video data or voice data. The research introduced deep learning so called DNN (Deep Neural Network). The developed method utilizes the conventional retrieval method for spoken term detection and extracts candidates in the first step. It realized the high retrieval accuracy and speed up by performing detailed matching between a query and the small number of extracted candidates in the second step. Furthermore, we realized the speed up and small resources by the method of pre-retrieval for all syllable bigrams.When a spoken query is given, we developed the spoken term detection system that realized high retrieval accuracy, speed up and small resources.

Free Research Field

音声言語処理