A Robust and Adaptive Word Boundary Detection Method

摘要

Word boundary detection is an important research topic of speech processing. A new method for word boundary detection is proposed, which divides the single band into multi bands and extracts the mean square deviation of the multi bands as a feature parameter for word boundary detection. The new method computers the estimation of the background noise and updates the estimation to adapt the change of the background noise. Experimental results show that adaptive band spectral mean square deviation is robust against with background noise well. The new method proposed in this paper achieved a good detection effect under different noise environment.