Spoken Language Proceeding Based on Non-Extensive Information Theory

2014 Fiscal Year Final Research Report

• Project/Area Number: 24650079
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Perception information processing/Intelligent robotics
• Research Institution: Tokyo Institute of Technology
• Principal Investigator: SHINODA KOICHI 東京工業大学, 情報理工学(系)研究科, 教授 (10343097)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: 音声情報処理 / 映像情報処理

Outline of Final Research Achievements

We have developed a methodology for spoken language processing based on non-extensive statistical theory, which is an extension from the conventional extensive statistical theory. We first developed q-log spectral subtraction (q-LMSN) to achieve robustness against the difference of environmental noises and of channels. We proved that it was significantly better than the conventional CMN. Next, we developed a recognition a method using q-Gaussian mixtures for output probabilities in GMMs and in HMMs. We applied it to speech recognition and to video semantic indexing and proved its effectiveness.

Free Research Field

統計的パターン認識