Study on predictive discrimination of sound data based on spatial-temporal Gaussian processes

2015 Fiscal Year Final Research Report

• Project/Area Number: 25280067
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Partial Multi-year Fund
• Section: 一般
• Research Field: Perceptual information processing
• Research Institution: The Institute of Statistical Mathematics
• Principal Investigator: Matsui Tomoko 統計数理研究所, モデリング研究系, 教授 (10370090)
• Co-Investigator(Kenkyū-buntansha): TAKEDA KAZUYA 名古屋大学, 情報科学研究科, 教授 (20273295) ; MARKOV KONSTANTIN 会津大学, コンピュータ理工学部, 准教授 (80394998)
• Co-Investigator(Renkei-kenkyūsha): UENO GENTA 統計数理研究所, モデリング研究系, 准教授 (40370093)
• Research Collaborator: PETERS GARETH W. University College London, UK, 統計学科, 講師 (90763061) ; IDO NEVAT A*STAR, Singapore・Institute for Infocom Research(12R), チームリーダー
• Project Period (FY): 2013-04-01 – 2016-03-31
• Keywords: ガウス過程 / 時空間モデリング / 音響信号処理 / 音楽情報処理

Outline of Final Research Achievements

We newly developed a general-purpose software “Monte Carlo Dynamic Classifier（MCDC）tool” for classification and regression with spatial-temporal Gaussian processes. In MCDC tool, state-space models are utilized and the state and observation functions are designed with Gaussian processes.
For modeling of acoustic space with spatial-temporal Gaussian processes, we designed a novel kernel function based on the wave equation in which the wave phase information was taken into account. The effectiveness of the kernel function was experimentally shown through comparison with the conventional method. Moreover, we investigated new methods for music genre classification and music emotion recognition using spatial-temporal Gaussian processes and experimentally showed that the proposed methods outperformed the conventional methods.

Free Research Field

統計的機械学習、音情報処理