Study on the Subspace Dimension of Single Phoneme Signals Based on PCA
Xianju Liu, Nikos E. Mastorakis, Zhongxiao Li, Xiaodong Zhuang · 2019
The subspace dimension of single phoneme signals is studied based on principal component analysis. It is found by experiments that the number of principal components of most unvoiced and voiced sounds is significantly different. It is further revealed that the number of principal components increases linearly with increasing length of signal vector, which is a new feature for the single phoneme signal. In addition, the effect of different frame shift on the number of principle components is investigated, which guarantees the effectiveness of the existing experimental results.