【摘要】 话音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，文章提出了基于子带矢量量化（ＳＢＶＱ）的话者模型。该话者模型由对应语音文本不同频段的各个子带上的矢量量化码本（ＳＢＶＱ码本）构成。ＳＢＶＱ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。识别时，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可进一步提升系统的性能。实验表明，文章提出的ＳＢＶＱ话者模型具有较好的时间鲁棒性。更多还原

【Abstract】 The time robustness of a speaker recognition system is one of the key problems for the system’s practical performance. In order to improve the time robustness, a speaker model based on subband VQ is presented in this paper. This model is composed of the VQ codebooks in subbands. This model can reflect the different effects of different subbands,and can localize the influence of some factors such as the time interval into one or more subbands. This model can improve the time robustness of the speaker recognition system. It was demonstrated by the experiments.更多还原