Title:
音声認識装置
Document Type and Number:
Japanese Patent JP2658104
Kind Code:
B2
Abstract:
PURPOSE:To discriminate a similar word whose phonemic sequence is similar by analyzing a necessary part of an input pattern and a standard pattern and rejecting the standard pattern by which the total distance of both the patterns become minimum. CONSTITUTION:The total distance of an input voice pattern from a parameter memory 8 and a standard pattern from a standard pattern memory 40 is calculated by a total distance calculating means 60, and the standard pattern by which the total distance is minimum is determined by a minimum value deciding means 70. Simultaneously, the input voice pattern and the standard pattern are analyzed partially, a distance difference of each channel of every frame of a detail distance reference means 90 is detected and the number of channels Cn by which the distance difference exceeds a threshold level is counted. Also, the number of frames (x) by which the number of channels Cn exceeds a threshold level is calculated by a difference frame detecting means 92, and when the number of frames (x) exceeds the threshold level, it is decided that a partially different part exists in the input voice pattern and the standard pattern, and the standard pattern by the means 70 is rejected by a rejection deciding means 93. Accordingly, a similar word whose phonemic sequence is similar can be discriminated without an error.
More Like This:
Inventors:
AKAHA MASATERU
TANAKA MYUKI
WATARI MASAO
TANAKA MYUKI
WATARI MASAO
Application Number:
JP32820087A
Publication Date:
September 30, 1997
Filing Date:
December 24, 1987
Export Citation:
Assignee:
SONII KK
International Classes:
G10L17/08; G10L15/02; G10L15/08; G10L15/10; G10L15/28; (IPC1-7): G10L3/00; G10L7/08
Attorney, Agent or Firm:
松隈 秀盛