论文部分内容阅读
语音关键词识别和确认方法在语音对话系统中得到了广泛的应用 .评价此类系统性能的一个重要指标就是处理非关键词 (垃圾 )的能力 .处理垃圾的传统方法是在离线状态下进行垃圾建模 .但该方法并不能很好的描述大量的系统词库以外的词 ,且训练规模较大 .该文提出了动态垃圾评价方法 ,不对垃圾本身建模 ,而是在识别过程中对输入语音进行可信度评估 ,从而对识别结果进行确认 ,解决了传统垃圾模型灵活性差及在线垃圾建模方法的确认能力不足等问题 .同时由于在垃圾评价中增加了反关键词信息 ,减少了关键词之间的识别错误
The method of speech recognition and verification is widely used in speech dialogue system.An important index for evaluating such system performance is the ability to deal with non-keywords (garbage) .The traditional method to deal with the garbage is to carry out the garbage off-line However, this method does not describe a large number of words outside the system thesaurus, and training a large scale.This paper presents a dynamic garbage evaluation method, not modeling the garbage itself, but in the identification process of input The credibility of speech is evaluated to confirm the recognition results and solve the problems of poor flexibility of traditional trash models and lack of confirmation ability of online spam modeling methods and the like and reduces the key due to the addition of anti-keyword information in rubbish evaluation Wrong word recognition