论文部分内容阅读
汉语实际发音中声母能量小,其频率成分多且分布分散;而韵母能量相对较大,其频率成分较少且集中于中低频。运用时频分析方法,将声母和韵母发音的这些差异同时体现在时频平面的时频原子特征上,提出利用所得到的Gabor原子参数的不同对汉语孤立字进行声韵分割的新方法,并通过引入遗传算法降低匹配追踪算法搜索原子的运算量。对115个汉语孤立字的仿真实验显示,该方法的分割正确率可达80.87%。
In Chinese, the initial consonant energy is small, its frequency component is more and distributed, while the vowel energy is relatively larger, its frequency content is less and concentrated in the middle and low frequency. By using the time-frequency analysis method, these differences of phonetic and vowel pronunciation are simultaneously reflected in the time-frequency atomic features in the time-frequency plane. A new method of vocalization segmentation based on Gabor atomic parameters is proposed. Introducing genetic algorithm to reduce the computational complexity of matching pursuit algorithm. The simulation experiments on 115 isolated Chinese characters show that the segmentation accuracy of this method is up to 80.87%.