Robust Method of Distance Estimation to a Speaker for Spoken Dialog System

Research Project

Project/Area Number	26330211
Research Category	Grant-in-Aid for Scientific Research (C)
Allocation Type	Multi-year Fund
Section	一般
Research Field	Perceptual information processing
Research Institution	Aichi University of Technology
Principal Investigator	Jitsuhiro Takatoshi 愛知工科大学, 工学部, 准教授(移行) (60394996)
Co-Investigator(Renkei-kenkyūsha)	TAKEDA Kazuya 名古屋大学, 大学院情報科学研究科, 教授 (20273295) SHIKANO Kiyohiro 奈良先端科学技術大学院大学, 名誉教授 (00263426)
Project Period (FY)	2014-04-01 – 2018-03-31
Project Status	Completed (Fiscal Year 2017)
Budget Amount *help	¥4,810,000 (Direct Cost: ¥3,700,000、Indirect Cost: ¥1,110,000) Fiscal Year 2017: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000) Fiscal Year 2016: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000) Fiscal Year 2015: ¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000) Fiscal Year 2014: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Keywords	音声認識 / 音声対話システム / 音源距離推定 / 音響モデル / VQコードブック / 深層学習 / Deep Belief Network / 発話者距離推定 / Deep Neural Network
Outline of Final Research Achievements	We propose the estimation method of distance from a mouth of a speaker to a microphone by estimating and classifying the feature of speech recorded by a single microphone. A Deep Neural Network (DNN) is training using speech data recorded for each distance. For estimation, short-time speech frames are entered into the DNN, it will estimate the distance for each frame. After that, the estimated distance is obtained for one utterance by majority decision of estimated distance in all frames. In speech recognition experiments of 1 m and 5 m, the proposed method can obtain about 85 % identification rate.