モーメントを用いた連続音声のピッチ周波数分布に関する研究

Abstract

In the speech signal, not only linguistic information but other various information is included. This paper describes about the shape of the pitch distribution using cepstral technique to extract these additional information. The symmetricity of the pitch frequency distribution is evaluated by using a moment (the mean, the variance, the skewness, and the kurtsis). Generally a pitch frequency distribution of a continuous speech has long tail to higher frequency. To improve the symmetricity, the logarithm transform is applied to the calculated data. Investigations were made on how the conversational speed, the voice power, and the speech intelligibility effects on the distribution shape.