普通话浊音声母的分隔和辨认

浊音声母的特点是低频信号显著:使用信号经过高频加重滤波器之后和之前的能量比可以反映语音信号能量集中的频率范围。据此,本文提出了一个将浊音声母从可能含有它的孤立字音中分隔出来并且进行分类和辨认的算法。对来自不同省分的八个讲者的语音数据进行分类和辨认实验,区分浊音声母和零声母两类的正确分类率达到98%以上。使用线性判别技术辨认各个浊音声母,平均正确辨认率达到86%。

ISOLATION AND RECOGNITION OF THE VOICED INITIALS IN PUTONGHUA

This paper discussed the speech characteristics of Putonghua voiced initials and the spectrum characteristics of preemphasis, then proposed an algorithm for thte isolation and classification of the initials basing on the measurement of the energy ratio of the speech signal with and without preemphasis. In the classification experiment using all Putonghua syllables with voiced initials spoken by a group of 8 adult speakers (4 male and 4 female) from different provinces of China we obtained an over 98% correct classification rate of the initials into voiced initials and' zero' initials. We also did an initial recognition experiment by applying stepwise linear discriminant analysis and obtained 86% average recognition rate of these initials.