SPEECH RECOGNITION METHOD WITH NOISE REDUCTION AND A SYSTEM THEREFOR

Patent №

US 5,201,004

Granted

1993-04-06

Filed 1991

Owner

NEC CORPORATION

Lab

AI components

4

ml · nlp · vision · speech

Assignment

Recorded

Dataset

AIPD

2023_r1 edition

Application

07704160

In a speech recognition system, a method for deciding an input pattern as one of reference patterns includes steps for calculating dissimilarities of the input pattern and the reference patterns and selecting a particular one of the dissimilarities which is lower than a threshold value. In order to reduce noise accompanying an input speech pattern in the input pattern, an average noise level is detected from an input pattern before the input speech pattern is detected and a noise factor is produced corresponding to the average noise level. The dissimilarities are multiplied by the noise factor to produce products and then the products are compared with the threshold value to recognize the input speech pattern. According to one aspect of the speech recognition system, particularly the system using the known clockwise DP matching with a beam search technique, the beam width factor is determined as a function of the noise level.

AI classification

Speech1.00
Natural language0.99
Vision0.64
Machine learning0.54
Planning0.44
AI hardware0.09
Evolutionary computation0.00
Knowledge representation0.00

Ownership

NEC CORPORATION

assignment · 57880415

Assignors

FUJIWARA, RYUHEI, SHIMADA, KEIKO

On an employer assignment, the assignors are typically the inventors.

© 2026 NYSGPT2525 LLC