Development of Secure Web System with Voice Interface to Identify Child Speakers

本研究开发了基于网络的儿童说话人识别系统，该系统可用于网络过滤系统，以保护儿童免受互联网的危害，开发方法如下：(1)使用我们的语音网络系统组织实际话语的收集。(2)开发了基于hmm的分类器原型系统。(3)比较了基于hmm的方法和人的听觉能力在区分儿童和成人话语方面的差异。我们考虑了有助于区分儿童和成人的语言特征。(5)提出了基于24类隐马尔可夫模型（hmm）和支持向量机（svm）相结合的分类器。在实验中，我们证明了通过使用真实网络用户的声音来识别儿童说话者的准确性。

In this study, the web-based system identifying child speakers, which can be adopted in web filtering system to protect children from the dangers of the Internet, has been developed as follows : (1) The collection of actual utterances was organized by using our voice-enabled web system. (2) The prototype system having the HMM-based classifier was developed. (3) The HMM-based method and the human's hearing abilities in distinguishing between child and adult utterances were compared. (4) We considered the linguistic features that help distinguish between children and adults. (5) The classifier based on combining 24-class HMMs (Hidden Markov Models) and SVMs (Support Vector Machines) was proposed. In the experiments, we proved improvements in accuracies in identifying child speakers by using voices captured from real web users.