首页> 外国专利> / Apparatus of Talker Dependence/Independence Speech Recognition

/ Apparatus of Talker Dependence/Independence Speech Recognition

机译:/说话者依赖性设备/独立语音识别

摘要

The present invention relates to a speaker-dependent / independent speech recognition apparatus capable of simultaneously recognizing a speaker-dependent word and a speaker-independent word.;This speaker dependent / independent speech recognition apparatus includes speech section detection means for detecting a speech section from a speech signal; Feature vector extracting means for extracting feature vectors from the speech signal from said speech segment detecting means; Vector quantization means for quantizing a feature vector from the feature vector extracting means using a codebook; Parameter estimation means for receiving an output signal of the quantization means and estimating the dependent model parameter; First storage means for storing the dependent model from the parameter estimating means; Second storage means for storing an independent model made from voice data of multiple speakers; A pattern matching means for pattern matching each of the dependent model of the first storage means and the independent model of the second storage means with the output signal of the preprocessing means to generate the dependent model similarity information and the independent model similarity information, respectively; Selection switch means for connecting the vector abbreviation means to any one of the parameter estimating means and the pattern matching means; A decision logic unit for determining a corresponding mode by comparing the dependent model similarity information and the independent model similarity information, and selectively operating according to a mode determined by the decision logic unit to output a recognition result determined using similarity information of the corresponding mode. And a post-processing means for determining a mode corresponding to the input voice using the dependent model similarity information and the independent model similarity information from the pattern matching means and outputting a recognition result determined in the corresponding mode, including a rejection unit.
机译:本发明涉及能够同时识别与说话者相关的单词和与说话者无关的单词的与说话者相关/独立的语音识别设备。该与说话者相关/独立的语音识别设备包括用于从中检测语音部分的语音部分检测装置。语音信号;特征向量提取装置,用于从来自所述语音段检测装置的语音信号中提取特征向量;向量量化装置,其使用码本从特征向量提取装置对特征向量进行量化。参数估计装置,用于接收量化装置的输出信号并估计从属模型参数;第一存储装置,用于存储来自参数估计装置的从属模型;第二存储装置,用于存储由多个扬声器的语音数据构成的独立模型;模式匹配装置,用于将第一存储装置的从属模型和第二存储装置的独立模型分别与预处理装置的输出信号进行模式匹配,以分别生成从属模型相似度信息和独立模型相似度信息;选择开关装置,用于将向量缩写装置连接到参数估计装置和模式匹配装置中的任何一个。判定逻辑单元,用于通过比较从属模型相似度信息和独立模型相似度信息来确定对应模式,并且根据由判定逻辑单元确定的模式来选择性地操作以输出使用对应模式的相似度信息确定的识别结果。以及后处理装置,其使用来自模式匹配装置的从属模型相似性信息和独立模型相似性信息来确定与输入语音相对应的模式,并输出以相应模式确定的识别结果,包括拒绝单元。

著录项

  • 公开/公告号KR100577990B1

    专利类型

  • 公开/公告日2006-08-30

    原文格式PDF

  • 申请/专利权人 LG ELECTRONICS INC.;

    申请/专利号KR19970081825

  • 发明设计人 김락용;

    申请日1997-12-31

  • 分类号G10L17;

  • 国家 KR

  • 入库时间 2022-08-21 21:23:52

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号