| Summary: | 碩士 === 國立成功大學 === 電機工程學系 === 102 === This paper proposes a robust template based on the previously proposed ECWRT (enhanced cross word reference template) for template-based ASR with Out-of-Vocabulary (OOV) detection ability, by using correlational weight adjusting method to improve robustness against speech variation named CWCWRT. This work addresses two vital issues: such as outlier rejection in training set and elimination of speech feature coefficients which usually unwanted utterances. Consequently, two main steps are investigated in this paper, firstly, correlational analyzing, and secondly, weight adjusting. OOV detection here is using the two-stage OOV detection to reject. The main idea of proposed algorithm is divided into two stages: count stage, which is to estimation the distribution of output score, and threshold stage, evaluate the confidence score of input feature vector. Two types of platforms including PC and GPCE063A embedded platform are conducted, both inside test and outside test are also applied. The results show that the average recognition rate for inside test is 98.20% in PC simulation and 93.97% in the embedded platform. The outside test results are 95.48% and 90.65% in two platforms respectively. The related and previous works including cross word reference template (CWRT) and ECWRT are also compared the comparison exhibit that the proposed CWCWRT gives higher robustness and accuracy than two baselines. OOV rejection ratio of OOV is 79%.
|
|---|
