提出一种基于改进竞争模型加权似然比融合的方法来进行关键词的语音确认,并将误差函数用于SVM对加权向量和阈值进行优化.通过构建两个基线系统,将不同竞争模型的加权方式的比较和3种模型的比较作了仿真实验.所得结果显示,相对于反词模型和常规竞争模型,加权似然比融合方法得到的似然比对关键词识别的平均错误率分别降低了26.53%和7.78%. This paper proposes a new method based on improved likelihood ratio likelihood ratio fusion method for speech recognition of key words, and uses the error function in SVM to optimize the weight vector and threshold.We construct two baseline systems to weight the different competition models The comparison between the two models and the comparison of the three models.The simulation results show that the likelihood ratio of weighted likelihood ratio fusion method reduces the average error rate of keyword recognition compared with the anti-word model and the conventional competition model 26.53% and 7.78%.