ウェーブレット解析と線形予測法による個人成分抽出

概要

論文の詳細を見る
Recently, the significance of smooth communication between human and mechanical systems in various fields has been increased. Several analysis methods of speech recognition such as cepstrum analysis, linear predictive analysis etc., have been suggested and remarkably evolved. They are mainly related with the analysis-synthesis coding of speech signals in order to realize the characteristic parameters extraction of high quality in the low bit rate coding. On the other hand, the specification of speaker and the perception of emotion also come to be significant in the human-machine interface techniques. However, there are a lot of unsolved problems in the speech recognition methods related with speaker specification and emotion recognition. Wavelets often give a better signal representation using Multiresolution analysis, with balanced resolution at any time and frequency. Therefore, this study introduces Wavelet analysis in order to improve the traditional speech recognition methods as a time-frequency signal processing technical skill. The purpose of this study is to suggest a new speech recognition method by integrating Wavelet analysis with the traditional speech processing methods making possible to realize the speaker specification and the emotion perception. The new method may make possible to apply to speech recognition-based IT or security systems as human-machine interface.
沼津工業高等専門学校の論文
2010-01-31