SYSTEM AND METHOD FOR DYNAMIC TIME EXPANSION AND CONTRACTION BASED ON GAUSSIAN MODEL FOR SPEECH PROCESSING

摘要

PROBLEM TO BE SOLVED: To provide a hierarchical statistical model representing a sound pattern by a Gaussian dynamic time expanding/contracting method. SOLUTION: This model is characterized in that a 1st layer represents a general sound space, a 2nd layer presents each speaker space, and a 3rd layer represents time structure information which is included in each registered speech speaking and based upon equal time intervals. Those three layers are hierarchically formed, the 2nd layer is extracted from the 1st layer, and the 3rd layer is extracted from the 2nd layer. This model is useful for a speech processing field, specially, a field of recognition of words and speakers using a retrieval recognition mode. COPYRIGHT: (C)2004,JPO&NCIPI