CAJ | 학술논문

针对电话手机语音的文本无关说话人确认中,训练集语音和测试集语音来自不同信道所产生失配而导致系统性能下降的问题,采用一种基于高维空间映射的方法对系统进行补偿.在分析了已有的说话人确认系统的基础上,提出了一种基于特征参数映射支持向量机模型 (PSVM)的说话人确认系统.首先用大量已知信道类型的语音训练出信道空间及映射矩阵,然后训练语音和测试语音都通过映射,消除因信道不同而导致的失配影响.在NIST数据库上的实验结果表明,这种方法弥补了训练语音和测试语音的失配,说话人确认系统的性能有了明显的提高.
침대전화수궤어음적문본무관설화인학인중,훈련집어음화측시집어음래자불동신도소산생실배이도치계통성능하강적문제,채용일충기우고유공간영사적방법대계통진행보상.재분석료이유적설화인학인계통적기출상,제출료일충기우특정삼수영사지지향량궤모형 (PSVM)적설화인학인계통.수선용대량이지신도류형적어음훈련출신도공간급영사구진,연후훈련어음화측시어음도통과영사,소제인신도불동이도치적실배영향.재NIST수거고상적실험결과표명,저충방법미보료훈련어음화측시어음적실배,설화인학인계통적성능유료명현적제고.
s: Mismatches occurring between training and testing cause performance degradation. For better decision-making in a text-independent speaker verification system, channel compensation by high-dimensional projection was introduced and a new projected support vector machine model was proposed through the based system. First, voices from a number of known channels were used to train channel spaces and projected matrix; then the parameters were projected onto the high-dimension in order to eliminate the effect of different channels. Text-independent speaker verification experiments conducted on the NIST speaker recognition evaluation corpus indicate remarkable improvement on the systems performance.