基于蚁群算法的普通话测试系统评分机制改进

An Improvement of the Scoring Mechanism of Mandarin Testing System Based on ACO

在分析现有国家计算机辅助普通话水平测试系统评分机制特点和原理的基础上,提出了一种使用蚁群算法对评分系统里评价模型进行优化的方法.实验结果表明,蚁群算法克服了传统归一化算法收敛慢、计算量大等缺点,它可以准确的估计评价模型向量并且不会产生局部相位的波形恶化,使得各处向量的功率谱有明显的增益.改进后的评价模型能对语音信号进行良好的识别,具有良好的通用性和全局性.

A new method using ACO to optimize the evaluation model of scoring mechanism was proposed based on an analysis of the characteristics and working principle of the scoring mechanism of the present computer-assisted Madarin test.The experiment result indicates that the ACO algorithm does not have the drawbacks of traditional normalization algorithm such as slow convergence and large amount of calculation;it can accurately estimate the vector and the evaluation phase of the wave without producing localized deterioration,and make the power spectrum of vectors gain remarkably.The improved model,with good versatility and generality,can identify the speech signal clearly.