CAJ | 학술논문

为了解决在说话人改变检测常用算法中寻找说话人改变点（Speaker change point，SCP）过程中存在的盲目性和遍历性问题，使用均值漂移的核密度估计寻找SCP：首先使用均值漂移寻找峰值点，然后计算峰值点的最大似然值，最后通过寻找最大值的方法寻找SCP．试验结果表明，提出的算法的F。值与贝叶斯信息准则（Bayesian information criterion，BIC）和DISTBIC接近．
위료해결재설화인개변검측상용산법중심조설화인개변점（Speaker change point，SCP）과정중존재적맹목성화편력성문제，사용균치표이적핵밀도고계심조SCP：수선사용균치표이심조봉치점，연후계산봉치점적최대사연치，최후통과심조최대치적방법심조SCP．시험결과표명，제출적산법적F。치여패협사신식준칙（Bayesian information criterion，BIC）화DISTBIC접근．
To settle out the problem that searching of speaker change point (SCP) was blind and exhaustive, kernel density estimation by using mean shift was proposed to seek SCP. Peak point was sought with mean shift first, maximum likelihood ratio of peak points was calculated secondly, and SCP was got by searching max value lastly. The experiments showed that the proposed algorithm got a comparable F1 value along Bayesian information criterion (BIC) and DISTBIC.