極・零モデルに基づく無声破裂音の分析と特徴抽出

概要

論文の詳細を見る
An autoregressive moving-average with adaptive orders of poles and zeros is adopted as a model for the production process of voiceless stop consonants. Analysis of 90 utterances of CV-syllables containing /p/, /t/ and /k/ indicates that the features of the three voiceless stop consonants can be quite explicitly represented by the trajectories of poles and zeros from the noise burst to the onset of the following vowel. If the orders of the model are optimized for voiceless stop consonants, the sum of the order of the AR scheme n and that of the MA scheme m is less than the order of the conventional all-pole model for vowels. Speech analysis with both adaptive and fixed orders of the model reveals the relation between the reduction of the amount of computation in the feature extraction and the decrease in the estimation accuracy. Separability of the three consonants is also examined for the estimated parameters.
社団法人日本音響学会の論文
1982-09-01