複素1次系をもとにした再帰型短時間スペクトル分析法とその応用

抄録

In the spectral analysis of temporally varying signals such as human voice and sound from musical instruments, the short time spectral analysis is known to be effective because with that method one can change the frame length depending on the frequency involved and choose arbitrarily the analysis interval in the frequency domain. Among the methods, the analysis with the complex first order system by Aoshima and its extension by Chinone are well-known and currently being used. In order to optimize and develop further the methods, it is necessary to investigate their stability because of its recursive nature. In this paper we first discuss the stability of the Chinone's method and by extending the discussion, next propose a new method of IIR short time spectral analysis which utilizes data window in order to analyze even lower level spectral components of signals. Furthermore, the effectiveness of the proposed method for feature extracting the tone and consonants in the process of transcription of music is experimentally demonstrated.