发明名称 SPEECH FEATURE QUANTITY CALCULATING DEVICE AND PROGRAM
摘要 PROBLEM TO BE SOLVED: To provide a means of reducing an influence of environmental noise included in a coefficient series representing a feature quantity of a speech through simple low-load processing for a system which calculates the coefficient series. SOLUTION: A mel-scale band filter processing section 913 passes a filter bank value calculated by filtering the spectrum of a speech signal through a mel-scale band filter to a logarithmic value calculation section 914, which passes a logarithmic value of the filter bank output value to a conversion section 101. The conversion section 101 converts the logarithmic value by using a function or conversion table meeting specified conditions to generate a deformed logarithmic value in which a componeont of a feature quantity of environmetal noise is reduced and passes it to a discrete cosine conversion processing section 915. The discrete cosine conversion processing section 915 generates an MFCC by using the deformed logarithmic value. The MFCC which is thus generated is used by a speech recognition device 92 for speech recognition. COPYRIGHT: (C)2007,JPO&INPIT
申请公布号 JP2007025296(A) 申请公布日期 2007.02.01
申请号 JP20050207775 申请日期 2005.07.15
申请人 YAMAHA CORP 发明人 YOSHIOKA YASUO
分类号 G10L15/02;G10L11/00 主分类号 G10L15/02
代理机构 代理人
主权项
地址