首页> 外国专利> METHOD AND DEVICE FOR MINIMIZING DEVIATION OF PRONUNCIATION EVALUATING VALUE USING LOW LEVEL MATRIX DECOMPOSITION

METHOD AND DEVICE FOR MINIMIZING DEVIATION OF PRONUNCIATION EVALUATING VALUE USING LOW LEVEL MATRIX DECOMPOSITION

机译:低水平矩阵分解最小化语音评估值偏差的方法和装置

摘要

Disclosed is a method for minimizing a deviation of pronunciation evaluating values using a low level matrix decomposition, which comprises the steps of: obtaining an evaluating value matrix of at least one evaluator for at least one pronunciation; setting an initial subjective value matrix representing subjective components of the evaluator and an initial true value matrix representing components excluding the subjective components of the evaluator based on the evaluating value matrix; obtaining a loss function containing non-correlation restraint function and a reconstruction error function based on the evaluating value matrix, the initial true value matrix and the initial subjective value matrix; calculating a corrected true matrix having a value of the loss function as a minimum value and a corrected subjective value matrix; and calculating a corrected evaluating value matrix based on the corrected true value matrix and the corrected subjective value matrix.;COPYRIGHT KIPO 2019
机译:公开了一种使用低级矩阵分解来最小化发音评估值的偏差的方法,该方法包括以下步骤:获得至少一个评估者的至少一个评估者的评估值矩阵;基于评估值矩阵,设置代表评估者的主观成分的初始主观值矩阵和代表评估者的主观成分之外的成分的初始真实值矩阵;基于评估值矩阵,初始真实值矩阵和初始主观值矩阵,获得包含非相关约束函数和重构误差函数的损失函数;计算以损失函数的值为最小值的校正后的真实矩阵和校正后的主观值矩阵;并基于校正后的真实值矩阵和校正后的主观值矩阵计算校正后的评估值矩阵。; COPYRIGHT KIPO 2019

著录项

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