Mel 频率下基于 LPC 的语音信号深度特征提取算法

针对传统语音信号二次特征提取方法在保证识别率的前提下,实时性较差的问题,提出一种Mel频率下基于线性预测系数(linear predictive coefficient,LPC)的改进的语音信号深度特征提取算法.该方法根据人耳的听觉特性把LPC在Mel频率下进行非线性变换,再进行微分、高阶微分和按比例重组等步骤,得到一种既考虑声道激励又兼顾人耳听觉的新特征参数,从而大大减少传统语音信号深度特征提取的计算量,在不影响识别效率的情况下,极大提高系统的实时性.最后,将该算法在智能轮椅平台进行有效性验证,大量实验表明,语音控制系统实时性差的问题在使用该算法后能够得到明显改善,该算法既保证了特征提取识别率,也有效地改善了系统的实时性.在一定程度上使语音控制智能轮椅更具实用性.

A further speech signal features extraction algorithm based on LPC Mel frequency scale

According to the bad real-time performance of the traditional further speech signal features extraction algorithm in the premise of ensuring the recognition rate, a further speech signal features extraction algorithm based on linear predictive coefficient(LPC) Mel frequency scale is put forward in this paper. This method transforms LPC with Mel-frequency in a nonlinear way, calculates the derivative, high order differential and combines the feature according to a certain proportion to realize a new features parameter which takes both the channel incentives and the human auditory into account. So the calculation quantity of the traditional speech signal further features extraction is decreased sharply. The real-time performance of the system is improved in the premise of ensuring the recognition rate. Through the intelligent wheelchair platform to verify the validity of the algorithm a lot of experiments show that the problem of real-time performance is not good of traditional algorithm can be improved effectively; this algorithm can improve the real-time performance and the practicability, on the basis of ensuring the recognition rate of the further features extraction.