Abstract
We develop the speech-displaying system called “KanNon” [1, 2] which supports speech communication of hearing-impaired people. The KanNon system displays a sound spectrogram and corresponding characters by speech recognition, which is scrolling on-line. To develop the KanNon system with the high accurate spectrum estimation of the speech, we adopt new spectral estimation method called “MCE-Burg[3]”, which is the method combining the continuous minimum cross entropy method(C-MCEM)[4, 5] and Burg method[6]. Finally, we compare sound spectrograms estimated by the Burg method and the MCE-Burg method, and show that results by the MCE-Burg method has sharper contrasts and peaks of formant compared with Burg method.
