To provide a method of extracting the features of narrow-band envelopes associated with sound quality and individuality of voices and identifying the differences in the sound quality based on the individual differences of an utterance mechanism.
The speech information identification method includes a process of extracting the envelopes of the speeches individually inputted from a plurality of speakers by each band, calculating correlation matrices and making a correlation matrix dictionary, a process of extracting the envelopes of the speeches inputted by the specific speakers by every prescribed band and calculating the correlation matrices, a process of identifying the calculated correlation matrices and the respective correlation matrices of the correlation matrix dictionary, and a process of selecting the speaker having the most resembling correlation matrix from the correlation matrix dictionary and identifying the speaker.
KAZAMA MICHIKO
YAMAZAKI YOSHIO
