首页> 外国专利> METHOD AND APPARATUS FOR DETERMINING PREDICTIVE TAG VALUE EACH OF A PLURALITY OF MUSIC TAG FOR GENERATING MAPPING FUNCTION OF MUSIC

METHOD AND APPARATUS FOR DETERMINING PREDICTIVE TAG VALUE EACH OF A PLURALITY OF MUSIC TAG FOR GENERATING MAPPING FUNCTION OF MUSIC

机译:确定用于生成音乐映射功能的多个音乐标签的预测标签值的方法和装置

摘要

Disclosed are a method and an apparatus for determining respective predictive tag values of a plurality of music tags for generating a mapping function of music. The disclosed method is a method for determining the respective predictive tag values of a plurality of music tags used to generate a mapping function of music to which music tags are not assigned. The method comprises the following steps of: searching for one or more reference music having a positive tag value of a target music tag among a plurality of reference music to which music tags are assigned; calculating first reference music with a minimum music distance from the music, among the one or more reference music, and calculating second reference music with a minimum music distance from the first reference music, wherein the music distance is calculated based on n number of pieces of sound information constituting music; calculating first dissimilarity between the music and the first reference music and second dissimilarity between the first reference music and the second reference music; and determining a predictive tag value of the target music tag as positive when the first dissimilarity is greater than the second dissimilarity, and determining the predictive tag value of the target music tag as negative when the first dissimilarity is smaller than the second dissimilarity. The purpose of the present invention is to expand a single class classifier (NNDD) utilizing only positive information of music tags to a multiple predictor (MLNNDD).
机译:公开了一种用于确定多个音乐标签的各个预测标签值以生成音乐的映射功能的方法和设备。所公开的方法是一种用于确定用于生成未分配音乐标签的音乐的映射功能的多个音乐标签的各个预测标签值的方法。该方法包括以下步骤:在分配了音乐标签的多个参考音乐中搜索具有目标音乐标签的正标签值的一个或多个参考音乐;在一个或多个参考音乐中,计算与该音乐具有最小音乐距离的第一参考音乐,并计算与该第一参考音乐与最小音乐距离的第二参考音乐,其中基于n个构成音乐的声音信息;计算音乐与第一参考音乐之间的第一差异以及第一参考音乐与第二参考音乐之间的第二差异;当所述第一差异大于所述第二差异时,将所述目标音乐标签的预测标签值为正;所述第一差异小于所述第二差异时,将所述目标音乐标签的预测标签值为负。本发明的目的是将仅利用音乐标签的肯定信息的单个类别分类器(NNDD)扩展到多个预测器(MLNNDD)。

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