基于“码袋”算法的话者自动辨认系统

Automatic speaker identification based on"bag of codes"

在话者自动辨认系统中,话者数量是决定辨认时间的最主要因素。因而在大数量注册话者的辨认中如何减少辨认所需要的运算时间是一个关键问题。针对这一问题,提出了一种新的基于"码袋"的话者模型设计算法,它通过统计"码袋"中每个码字在话者语音中的概率分布来实现话者模型的设计。实验结果表明该算法在保证较高辨认率的同时,有效地降低了话者自动辨认系统的计算复杂度。

The most dominating factor of the identification time is the number of speakers.Thus, how to reduce the computational cost of evaluating large speaker database is the key problem.Due to this, a＂bag of codes＂algorithm is proposed.This novel algorithm can generate speaker models by estimating the probability distribution of codes in speech data.Experiments prove that the new algorithm can reduce the computational complexity in the speaker identification system with high identification accuracy.