LREC 2000 2nd International Conference on Language Resources & Evaluation

Previous Paper   Next Paper

Title Automatic Speech Segmentation in High Noise Condition
Authors Ivanov Rosen (Department “Computer Systems and Technologies”, Technical University, 4 “H.Dimitar”, 5300 Gabrovo, Bulgaria,
Session Session SP4 - Tools for Evaluation and Processing of Spoken Language Resources
Full Paper, 80.pdf
Abstract The accurate segmentation of speech and end points detection in adverse condition is very important for building robust automatic speech recognition (ASR) systems. Segmentation of speech is not a trivial process - in high noise conditions it is very difficult to determine weak fricatives and nasals at end of the words. An efficient threshold (a priory defined) independent speech segmentation algorithm, robust to level of disturbance signals, is developed. The results show a significant improvement of robustness of proposed algorithm with respect to traditional algorithms.