摘要： 在自动语音识别系统的实际应用中,词表外(Out-of-Vocabulary,OOV)语音的检测与拒识非常重要。针对语音识别的实时性要求,提出了一种新的基于支持向量机的OOV快速拒识算法,并将其应用于基于DSP实现的孤立词语音识别系统中。实验结果表明,该算法计算简单,实时性好,且效果显著,拒识率达80%以上。通过与传统神经网络方法的比较,证明该算法具有更好的性能和应用潜力。

关键词: 支持向量机, OOV拒识, 语音识别, 神经网络

Abstract: In practical automatic speech recognition systems, out-of-vocabulary (OOV) utterance detection and rejection represent an important part. This paper proposes a new OOV rejection algorithm based on support vector machine and uses it in isolated-word speech recognition system based on DSP hardware environment. The experimental results show that this algorithm is very simple and efficient, which attains over 80% OOV rejection rate. By comparing the results with traditional neural networks method, the performance of this algorithm is better and has more potential in application.

中图分类号: 

• TP391.42