口腔出力と鼻腔出力の分離ピックアップ方式による摩擦音・破擦音の識別

概要

論文の詳細を見る
Fundamental studies for automatic recognition of Japanese fricatives and affricates are performed. The oral and nasal outputs are measured simultaneously through two separate microphones. For eliminating the effect of the superposed low frequency wave the noise rejected differential zero crossing rate(N. R. -D. Z. C. R. )is proposed. Maximum value of the N. R. -D. Z. C. R. during the starting period of syllables is used as one of the parameters for classifying the fricatives and affricates. For the vowels the value of the N. R. -D. Z. C. R starts to increase simultaneously with the increase of the nasal output energy. Concerning to the voiceless fricatives and affricates, however, the increase of the N. R. -D. Z. C. R. is found to be earlier than that of the energy. In the case of the voiced fricatives, the situation is just opposite. The time difference between the increase of the N. R. -D. Z. C. R. and that of the energy is also used as another parameter. In the first experiment, a plate is used to shield the microphones. The recognition accuracy of 90 percent is obtained for the fricatives and affricates in isolated CV's uttered by ten males. In the second experiment, the shield plate is removed. However, the recognition result is still higher than88percent and it is considered that the two proposed parameters are very effective for recognizing the fricatives and affricates.
社団法人日本音響学会の論文
1983-05-01