The Research On Speaker-Independent Continuous Speech Recognition Based On The High-Dimension Space Covering Theory

The correct rate of endpoint detection is hard to get 100%, since there is co-articulation between two words of continuous speech and there's no clear boundary between these two words. Then the incorrect endpomt is certain to get the incorrect recognition result. In order to solve these problems, we present a novel algorithm of speaker-independent continuous Mandarin figures speech-recognition, which is based on the dynamic searching theory of high dimensional space vertex covering.In this paper the new definitions of some elements such as curves, curved surfaces etc. in high dimensional space (HDS) are presented from the perspective of descriptive geometry and set theory. The primary concepts about Vertex Covering are also presented based on the high dimensional space covering theory (HDSCT). Then we analyze the distribution of speech samples in high dimensional space, propose some kinds of geometrical coverage and study the relationship among them. At last, the high dimensional space regions covering different classes of figures are constructed and a new dynamic searching algorithm based on the high dimensional space covering theory is presented and applied to the speaker-independent continuous speech recognition without endpoint...