使用惩罚距离进行说话人改变检测

摘要

In order to avoid tuning penalty factor repeatedly when using Bayesian information criterion （BIC） for speaker change criterion, penalty distance was induced from BIC and applied to speaker change detection. Comparing with the distance-BIC the experimental results indicated that the proposed algorithm didn＇t need to tune parameter while F1 increased 2%.%在说话人改变判决中，为了避免使用贝叶斯信息准则（Bayesian information criterion，BIC）需要不断调节惩罚因子的问题，作者从BIC中推导出惩罚距离公式并将其应用于说话人改变检测中．试验结果表明，提出的算法和BIC相比不需要调节参数，与BIC距离相比，F1提高了2％．