2000 Fiscal Year Final Research Report Summary
STUDY ON FEATURE EXTRACTION IN CLASSIFYING SOUNDS LIKE SNORE BY USING MATHEMATICAL METHODS
Project/Area Number 
09480052

Research Category 
GrantinAid for Scientific Research (B).

Allocation Type  Singleyear Grants 
Section  一般 
Research Field 
計算機科学

Research Institution  NAGOYA UNIVERSITY 
Principal Investigator 
HASEGAWA Katsuo NAGOYA UNIVERSITY, GRADUATE SCHOOL GRADUATE SCHOOL OF MATHEMATICS, PROFESSOR, 大学院・多元数理科学研究科, 教授 (70004463)

CoInvestigator(Kenkyūbuntansha) 
HIRASHITA Fumiyasu NAGOYA UNIVERSITY, GRADUATE SCHOOL GRADUATE SCHOOL OF MATHEMATICS, PROFESSOR, 大学院・多元数理科学研究科, 教授 (80314061)
MIYAO Masaru NAGOYA UNIVERSITY, GRADUATE SCHOOL GRADUATE SCHOOL OF MATHEMATICS, PROFESSOR, 大学院・多元数理科学研究科, 教授 (70157593)
MIHASHI Koushin NAGOYA UNIVERSITY, GRADUATE SCHOOL GRADUATE SCHOOL OF MATHEMATICS, PROFESSOR, 大学院・多元数理科学研究科, 教授 (30022594)

Project Period (FY) 
1997 – 2000

Keywords  Analysis of snoring sounds / Fourier Transformation / Visualization / Feature extraction / Spectrogram / Mathematical method 
Research Abstract 
After four years of research on feature extraction in classifying sounds like snore by using mathematical methods, we can show you at least four important results. As written in the proposal, we had intended to focus to survey the pattern recognition mechanism. However, it gradually became clear the recognition is deeply related with the structure of brain and mind, most active field all over the world. During the analysis of recorded snoring sounds using Fourier transformation, we found the existence of life security system working for 24 hours ready in emergency to recover from suffocation caused by tongue plug. First, the patterns of many snoring sounds were transformed into visual graphic patterns using Fourier transformation. Second, using mathematics like linear algebra and differential geometry, many models of neural networks were surveyed and learning mechanism was analyzed. Then we found the fact each neuron is making a decision comparing the total input amount of stimulation with its threshold so that the network can make closed area that can be regarded as a memory category. This process of cogitation inspired us to discover new algorithm of inference designated as the Half Defined Inference Method (HDIM). We also discovered the very much fundamental theory designated as the Two Face Theory (TFT).
