抄録
We have developed the new real time speech-displaying system called “KanNon” which helps deaf person to understand speaker's speech contents by spectrogram-reading. The term of the spectrogram-reading means understanding the speech contents by seeing formant transition which appears in the sound spectrogram. This idea is based on that the speaker's speech is characterized by formant transition. The KanNon system displays not only sound spectrogram, but also pitch frequency and loudness of speech and characters by speech-recognition system as real time scrolling image. For the purpose of displaying formant patterns clearly with high accuracy, we applied Burg method combining with the minimum cross-entropy (Burg-MCE) method to estimate the autoregressive model parameters, and also applied the mel-scale to the frequency axis in the sound spectrogram instead of the hertz-scale. Finally, we show more effective display for the spectrogram-reading in the KanNon system.
