计算机工程与应用
計算機工程與應用
계산궤공정여응용
COMPUTER ENGINEERING AND APPLICATIONS
2010年
11期
122-124
,共3页
线性预测残差%HAAR%HOCOR%分级说话人辨识%基音周期
線性預測殘差%HAAR%HOCOR%分級說話人辨識%基音週期
선성예측잔차%HAAR%HOCOR%분급설화인변식%기음주기
Linear Prediction(LP) residue%HAAR%HOCOR%hierarchical speaker identification%pitch period
从线性预测(LP)残差信号中提出了一种新的特征提取方法,这种特征跟单个的说话人的声道密切相关.通过把HAAR小波变换运用于LP残差而获得了一个新的特征(HOCOR).为了进一步提高系统的鲁棒性和辨识率,在采用分级说话人辨识的基础上,将基音周期的高斯概率密度对GMM分类器的似然度进行加权,形成新的似然度进行说话人辨识.试验结果显示,所提出系统的鲁棒性和辨识率都有所提高.
從線性預測(LP)殘差信號中提齣瞭一種新的特徵提取方法,這種特徵跟單箇的說話人的聲道密切相關.通過把HAAR小波變換運用于LP殘差而穫得瞭一箇新的特徵(HOCOR).為瞭進一步提高繫統的魯棒性和辨識率,在採用分級說話人辨識的基礎上,將基音週期的高斯概率密度對GMM分類器的似然度進行加權,形成新的似然度進行說話人辨識.試驗結果顯示,所提齣繫統的魯棒性和辨識率都有所提高.
종선성예측(LP)잔차신호중제출료일충신적특정제취방법,저충특정근단개적설화인적성도밀절상관.통과파HAAR소파변환운용우LP잔차이획득료일개신적특정(HOCOR).위료진일보제고계통적로봉성화변식솔,재채용분급설화인변식적기출상,장기음주기적고사개솔밀도대GMM분류기적사연도진행가권,형성신적사연도진행설화인변식.시험결과현시,소제출계통적로봉성화변식솔도유소제고.
A novel feature extraction method from LP residue signal is proposed.This kind of feature has good relation with vocal tract of speaker.A novel feature(HOCOR) is acquired by applying LP residue with HAAR.In order to improve the robustness and identification rate of the system, hierarchical speaker identification is proposed .Then the likeliness of CMM classifier is weighted by the Gauss probability density of the pitch to form the novel likeliness which is used for speaker identification.The experiment result shows that the robustness and identification rate of the system proposed are both improved.
