公开(公告)号：KR20100138623A

公开(公告)日：2010-12-31

申请号：KR20090057224

申请日：2009-06-25

Applicant: KOREA ELECTRONICS TELECOMM

Inventor： CHO HOON YOUNG ,  KIM DONG HYUN ,  HWANG KYU WOONG ,  KIM YOUNG IK ,  KIM SANG HUN ,  PARK SANG KYU

IPC: G10L15/06 ,  G10L15/04

Abstract: PURPOSE: A method for voice recognition for performing the voice recognition about the noises is provided to differentially change the generation degree by division of the signal according to reliability information. CONSTITUTION: An interval classifying unit(110) compartmentalizes the inputted voice signal into the plural periods of the specific unit. A Fineness determining unit(120) determines the granularity of the grid generation among plural periods about one section. A grating generating unit(130) generates a lattice for the voice recognition according to granularity. A decoding unit(140) acquires a voice recognition result by performing the search space.

Abstract translation: 目的：提供一种用于执行噪声的语音识别的方法，以通过根据可靠性信息划分信号来差异地改变生成度。 构成：间隔分类单元（110）将输入的语音信号分隔成特定单元的多个周期。 细度确定单元（120）确定围绕一个部分的多个周期中的网格生成的粒度。 光栅产生单元（130）根据粒度产生用于语音识别的格子。 解码单元（140）通过执行搜索空间来获取语音识别结果。