A multi-confidence feature combination rejection method for robust speech recognition
Junfeng Zhao, Yeping Zhu · 2011
Propose a rejection method which combine multiple confidence measure from different knowledge source to improve the robustness of speech recognition systems in mobile environments. Through the analysis of rejection algorithm based on garbage model and confidence feature selection, we select online garbage score, duration probability and Log likelihood ratio (LLR) to score the confidence for keyword candidates. Experimental results show that the combination of these features can improve rejection performance obviously.