基于子带VQ及ANN的话者确认系统的设计  

The Design of Speaker Verification System Based on Subband VQ and ANN

作　　者：方绍武[1] 戴蓓倩[1] 徐文盛[1] 刘鸣[1] 

机构地区：[1]中国科学技术大学电子科学与技术系,合肥230026

出　　处：《电路与系统学报》1999年第4期24-29,共6页Journal of Circuits and Systems

基　　金：国家自然科学基金

摘　　要：话者识别系统的时间鲁棒性是影响话者识别系统实用化的关键问题之一。为了提高系统的时间鲁棒性，本文提出了基于子带矢量量化（SBVQ）及人工神经网络（ANN）的话者模型。将语音文本的有效频段划分为几个子带，分别求取于带上的矢量量化码本（SBVQ码本），利用BP型人工神经网络（BPNN）对训练数据在各个子带上的量化误差进行拟合，即可训练出话者模型（SBVQ码本及BPNN的极值矩阵、确认阈值）。该话者模型反映了不同频段对话者识别系统性能的不同影响，并可将时间间隔等因素对系统性能的影响局限在某个子带内从而提高模型的时间鲁棒性。实验表明，本文提出的（SBVQ＋BPNN）话者模型具有较好的时间鲁棒性。The time robustness of a speaker recognition system is one of the key problems for the system's practicalperformance. In order to improve the time robustness, a speaker model based on sub-band VQ and ANN is presented. Thewhole valid band is divided into several sub-bands, Their VQ code-books are calculated independently, and then thequantization errors are fitted using BPNN. By this training aPproach, a speaker model, which is comPosed of sub-band VQcode-books, weighting matrix and a thresholds can be determined. This model reflects the different effects of different sub-bands, and the influence of some factors such as the time interval can be localized within one or more sub-bands. Theexperiments show that the (SBVQ+BPNN) speaker model has good time robustness.

关 键 词：鲁棒性 子带矢量 神经网络 话者模型 语音识别 

分 类 号：TN912.34[电子电信—通信与信息系统]