摘要
水下目标识别是潜艇在海战中,先敌发现并有效进行水声对抗的关键技术。然而,如何根据声纳接收到的舰船辐射噪声对三类目标进行分类识别是长期困扰人们的问题。研究了四种语音识别中常用的方法——线性预测系数(LPC),线性预测倒谱系数(LPCC),美尔倒谱系数(MFCC)和最小均方无失真响应(MVDR),在水下目标识别中的应用效果,并比较了这四种方法在无噪声情况下的识别概率,以及在不同信噪比下的识别概率,并通过比较找到在无噪声和有噪声情况下的最佳方法。实验表明,在无噪声的情况下,MFCC方法总体识别率最高,第一类目标MFCC方法的识别率最高,第二类目标MFCC和MVDR方法识别率相似,好于其他两者,第三类目标MVDR方法识别率最高。在加入噪声的情况下,MVDR方法对三类目标的识别和抗噪声性能明显好于其余三者。
In submarine battle,underwater target recognition is a key technology used for early finding enemy and then taking effective acoustic countermeasure to defeat the enemy.However,a problem that puzzles people for a long time is how to classify and identify three kinds of ship targets based on the received ship radiation noise by sonar system.This paper studies the effects of applying four frequently-used speech recognition methods on underwater target classification.The four methods are linear prediction coefficient（LPC）,linear prediction cepstrum coefficient（LPCC）,Mel frequency cepstrum coefficient（MFCC） and Minimum Variance Distortionless Response（MVDR）.The paper verifies the results by comparing the rates of recognition for different target samples under different SNRs,and finds out the best method in the condition of having or no noise.Experiments show that,without noise,the overall recognition rate of MFCC is highest;MFCC method has the highest recognition rate to the first kind of targets;MFCC and MVDR methods have similar recognition rate to the second kind of targets and better than the other two;and MVDR method has the highest recognition rate to the third kind of target.In the case of adding noise,the recognition and anti-noise performances of MVDR method to the three kinds of targets are significantly better than the other three.
出处
《声学技术》
CSCD
2012年第5期530-534,共5页
Technical Acoustics
关键词
语音识别
线性预测系数
线性预测倒谱系数
美尔倒谱系数
最小均方无失真响应
speech recognition
linear prediction coefficient
linear prediction cepstrum coefficien
mel frequency cepstrum coefficient
minimum variance distortionless response
