Determination of regression classes tree structure for a speech recognizer

摘要

Tree nodes combine the word constituting elements in dependence on a rate for the element spacings in acoustic chamber. The combination of regression classes to a class lying nearer to a tree root is carried out according to a correlation rate. At the start of the determination of the tree structure each word constituting element forms a basic regression class, then successively each two regression classes with largest correleation rate are combined to a new regression class. In the next step of tree structure forming the combination is considered instead of the original classes until a regression class, representing the tree root, is formed. Independent claims are included for use of the tree structure and a speech recognition system.