Research And Design Of Confidence Measures In Speech Recognition Based On A Combination Score

Voice confidence degree assessment is a combination of voice recognition and voice signal processing of a technical, the paper objective of the study is to establish an English voice confidence degree evaluation system that focuses on the standard voice and data modeling software was compared to the voice data rating.The entire system consists of three parts: voice signals cutting, voice content certification rating and voice confidence measures. Which voice signals cutting through voice recognition technology, based on the acoustic model energy threshold, graphics than a way to correct pronunciation sectors. Voice content certification is to voice the voice signal cutting out factors each time zone, and with good pre-training voice content pronunciation acoustic models as compared to the certification gives voice to the designated testing voice content and content consistency. Voice for the rating system core, the scoring method was used with standard software voice signal characteristics of voice in comparison confidence measures, scoring three standard reference features : models in a few words of concealing techniques, and dynamic atmosphere of the time, the vector quantified characteristics. Different characteristics of the design parameters for reasonable scoring mechanism, and three parameters weighted arithmetic to establish a reasonable joint rating system.Based on the above theoretical research to MATLAB6.5 for the development environment, to achieve a preliminary system design. Test results showed that the system functions to the expected design goals.