一种改进的特定人语音识别系统及算法研究  

An improved speaker-dependent speech recognition systems and algorithms

作　　者：赵智琦 房建东[1] 

机构地区：[1]内蒙古工业大学,内蒙古呼和浩特010080

出　　处：《电子设计工程》2014年第16期31-34,37,共5页Electronic Design Engineering

基　　金：内蒙古自治区科技计划项目(20120304);内蒙古自治区科学基金项目(2009MS0913)

摘　　要：针对传统特定人语音识别过程中存在的算法复杂、所占存储空间大等问题,提出了一种改进的基于动态时间规整算法(DTW)的特定人语音识别系统。在对参数提取方法进行详细对比之后,提取美尔频率倒谱系数(MFCC)作为本系统的语音识别参数,有效的解决了人耳响应不同信号灵敏度不同的问题。利用MATLAB环境下语音工具箱Voice Box实现了对若干数字的孤立词识别,识别速度提高了约30%,识别成功率达到95%以上。仿真结果证明,该系统在算法简单,识别成功率高,是一种简单有效的语音识别方法。In speech recognition of the traditional process, there exists the problem of complexity, large storage space, and so on. Aiming at these problems ,proposed an improved algorithm based on dynamic time warping （DTW） speech recognition system for specific people. After a detailed comparison on the parameter extraction method, extraction of Mel frequency cepstrum coefficient （MFCC） as the speech recognition of the system parameters, to effectively solve the ear response signals with different sensitivity to different problems. Using MATLAB environment the Voice toolkit Voice Box to realize the isolated word recognition of several Numbers, the recognition speed by about 30% and the success rate of 95% or more. The simulation results show that, the algorithm is simple, the recognition success rate is high, is a simple and effective method for speech recognition.

关 键 词：语音识别 动态时间规整算法 美尔频率倒谱系数 孤立词 

分 类 号：TN602[电子电信—电路与系统]