计算机工程与应用
計算機工程與應用
계산궤공정여응용
COMPUTER ENGINEERING AND APPLICATIONS
2014年
9期
214-218
,共5页
频率规整线性预测(WLPC)%最小均方误差(LMS)算法%自相关%实时
頻率規整線性預測(WLPC)%最小均方誤差(LMS)算法%自相關%實時
빈솔규정선성예측(WLPC)%최소균방오차(LMS)산법%자상관%실시
Warped Linear Predictive Coding(WLPC)%Least Mean Square(LMS)algorithm%autocorrelation%real-time
在语音识别特征提取过程中，为克服传统自相关法在计算特征参数时实时性较差的缺点，提出一种用于提取频率规整线性预测系数（WLPC）的自适应最小均方误差（LMS）算法。该方法通过自适应LMS技术，不仅能提取出符合人耳的听觉特性的特征参数，而且实现了对WLPC系数的实时提取。实验采用DTW（动态时间规整）算法，对比了自相关法WLPC预测误差和自适应法WLPC两种特征参数对孤立词识别率的影响结果和预测误差，结果证明了采用该算法具有较高的分类准确率和良好的时间性能。
在語音識彆特徵提取過程中，為剋服傳統自相關法在計算特徵參數時實時性較差的缺點，提齣一種用于提取頻率規整線性預測繫數（WLPC）的自適應最小均方誤差（LMS）算法。該方法通過自適應LMS技術，不僅能提取齣符閤人耳的聽覺特性的特徵參數，而且實現瞭對WLPC繫數的實時提取。實驗採用DTW（動態時間規整）算法，對比瞭自相關法WLPC預測誤差和自適應法WLPC兩種特徵參數對孤立詞識彆率的影響結果和預測誤差，結果證明瞭採用該算法具有較高的分類準確率和良好的時間性能。
재어음식별특정제취과정중，위극복전통자상관법재계산특정삼수시실시성교차적결점，제출일충용우제취빈솔규정선성예측계수（WLPC）적자괄응최소균방오차（LMS）산법。해방법통과자괄응LMS기술，불부능제취출부합인이적은각특성적특정삼수，이차실현료대WLPC계수적실시제취。실험채용DTW（동태시간규정）산법，대비료자상관법WLPC예측오차화자괄응법WLPC량충특정삼수대고립사식별솔적영향결과화예측오차，결과증명료채용해산법구유교고적분류준학솔화량호적시간성능。
To overcome the disadvantages that the traditional autocorrelation algorithm has a poor performance in real-time extraction, an adaptive Least Mean Square(LMS)algorithm which is used to extract WLPC coefficients is presented. Based on an adaptive LMS algorithm, the proposed algorithm not only realizes the real-time extraction of the feature parameters which accord with characteristics of human hearing, but also extracts WLPC coefficients in real-time. A speech recognition model based on DTW algorithm is used to estimate the performance of autocorrelation algorithm and adaptive LMS algorithm. The experimental results demonstrate that the proposed algorithm has high classification accuracy and good real-time performance.
