计算机语音信号处理与语音识别系统  

A Speech Signal Processing and Phonetic Recognition System

作　　者：朱学芳[1] 徐建平[1] 

机构地区：[1]南京邮电学院计算机科学与技术系

出　　处：《南京邮电学院学报》1998年第5期113-119,共7页Journal of Nanjing University of Posts and Telecommunications(Natural Science)

摘　　要：对计算机语音处理和对单个数码字识别的实现进行了探讨。根据汉语语音的特点，以汉语单音字作为识别对象，对１０个数码字识别进行了研究和实验。通过观察和分析语音信号的时域特性（主要是短时帧能量、短时过零率和帧能量差），并把它们应用于语音端点检测，为系统的建立做了基础准备。选用了语音信号的功率谱差的特征，进行了模板的建立与识别实验。测试结果表明，该系统性能较稳定，单个数码字识别率可达９８６％，说话人识别率达到９２２％。This thesis reports the course of the speech signal processing and the experimental recognition of ten decimal numbers.Based on Chinese characteristics,a Chinese syllable is used as a basic recognition unit.We construct a speaker dependent,experimental system for the single number speech recognition.Based on the sequential characteristics of speech signal,the energy of frames and the zero crossing are used for the speech endpoint detection which is efficient for the speech recognition system.The experimental system consists of the following two parts:using the energy and its difference to detect speech endpoints and to get the silence frames and the voice frames,and dividing the band of 20Hz～4kHz into eighteen frequency bands based on the critical band.The result shows that the recognition rate of the single number is 98.6% and the recognition rate of the speaker is 92.2%.

关 键 词：语声处理 语声识别 WAVE文件格式 临界频带 

分 类 号：TN912.3[电子电信—通信与信息系统]