一种基于后验概率差值的拒识算法

An out-of-vocabulary rejection based on likelihood difference for speech recognition

本文提出了一种孤立词语音识别系统中基于后验概率差值的拒识算法。研究了作为拒识特征时,输入词的后验概率和后验概率差值之间的区别,并将多层感知人工神经网络用于拒识特征的学习。相比现存的几种拒识算法,本算法几乎不需要额外的计算和存储量。当识别率为98.2％时,拒识率达到了95.4％。

This paper proposes an algorithm based on likelihood difference for isolated-word speech recognition system. We explore the behaviors of likelihood and likelihood difference as discriminative features for Out-Of-Vocabulary (OOV) rejection. The Multi Layer Perceptron (MLP) is used to map the features to the OOV rejection. Compared with the existent methods, the additional amount of computation and storage is negligible in the presented algorithm, and the rejection rate can be 95.4% with an acception rate of 98.2%.