基于在线垃圾模型的语音确认方法

Utterance Verification Based On On-line Garbage Modeling Evaluation Approach

该文针对关键词检测和实用语音识别中OOV(Out-Of-vocabulary)问题的检测和拒绝进行研究 ,通过使用判别式分析 ,利用L -Best本地分数和N -best言语假设判别分数 ,进行言语判别 (utteranceverification) ;该文进行了两组实验 ,分别针对OOV问题 ,在小词汇量特定人孤立词识别系统、小词汇量非特定人的孤立词识别系统中进行研究。

For the OOV (Out-Of-vocabulary) problem in key word spotting and practical speech recognition, we can finish the utterance detection with discriminant analysis using L-best local scores and N-best scores, and provide two examinations. Especially for OOV problem, one based on small vocabulary isolated word recognition system for speaker dependent and the other based on small vocabulary isolated recognition system for speaker independent.