An Effective PCM Based Environment Compensation Approach in Speech Processing for Mobile e-Learning Platform

摘要

This paper presents an efficient environment compensation approach for speech recognition on mobile e-learning platforms, based on the Parallel Model Combination method. The probability density of the corrupted speech is calculated directly from the clean speech model and the noise model, which avoid the estimation error in Log-Normal Approximation. The proposed algorithm accelerates the integration computation by approximating the its value over a rectangle area. Experiment result shows that our approach is robust under low SNR environment, compared with the previous metaphors.