Description: 本 文 首先 介绍了语音识别的研究和发展状况，然后循着语音识别系统的
处理过程，介绍了语音识别的各个步骤，并对每个步骤可用的几种方法在实
验基础上进行了分析对比。研究了语音信号的预处理和特征参数提取，包括
语音信号的数字化、分帧加窗、预加重滤波、端点检测及时域特征向量和变
换域特征向量.其中端点检测采用双门限法.通过实验比对特征参数的选取，
采用12阶线性预测倒谱系数作为识别参数。详细分析了特定人孤立词识别算
法，选定动态时间弯折为识别算法，并重点介绍其设计实现。
在 Vi su alC++环境下，设计并实现一个特定人、孤立词语音识别系统，
系统可以识别数字0-9等简单指令。该系统还具备演示、学习功能，可以演
示语音处理的各个步骤，还可以根据需要添加新的指令。
最 后 ， 重点从端点检测算法和动态时间弯折识别算法对系统进行改进。
实验表明，改进后的系统识别率有很大提高，达到95 ，为进一步开发实用
性语音识别系统产品打下了基础。-This article introduced the first speech recognition research and development, and then follow the voice recognition system
Processing, speech recognition, introduced the various steps, each step of the methods available in the real
A post-mortem conducted on the basis of the analysis and comparison. Research on the speech signal pre-processing and feature extraction, including
Digitized voice signals, sub-frame window, pre-emphasis filtering, endpoint detection feature vector in time domain and variable
Eigenvector for the domain. One endpoint detection method using dual-threshold. Through experiments over the selection of characteristic parameters,
The use of 12-order linear prediction cepstral coefficients as recognition parameters. Detailed analysis of the specific operator who isolated word recognition
Law, selected Dynamic Time Warping Algorithm for identifying and focusing on the achievement of its design.
In Vi su alC++ environment, design and realization of a s Platform: |
Size: 2491392 |
Author:周文超 |
Hits:
Description: 本 文 首先 介绍了语音识别的研究和发展状况，然后循着语音识别系统的处理过程，介绍了语音识别的各个步骤，并对每个步骤可用的几种方法在实验基础上进行了分析对比。研究了语音信号的预处理和特征参数提取，包括括语音信号的数字化、分帧加窗、预加重滤波、端点检测及时域特征向量和变换域特征向量.其中端点检测采用双门限法.通过实验比对特征参数的选取，采用12阶线性预测倒谱系数作为识别参数。详细分析了特定人孤立词识别
-This paper first introduces the research and development of speech recognition, and then follow the process of the speech recognition system, describes the various steps of the speech recognition, several methods are available and each step were analyzed and compared on an experimental basis. Study of the speech signal preprocessing and characterized parameter extraction, including voice signals of the digitization, stars frame windowed pre heavier filtering, endpoint detection timely domain feature vectors and transform domain feature vector. Wherein the endpoint detection using double doors limit law. By experiment than the characteristic parameters of the selected 12 order linear prediction cepstral coefficients as recognition parameters. A detailed analysis of a specific isolated word recognition Platform: |
Size: 2491392 |
Author:mx |
Hits:
Search - feature vector of voice signal