• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>International Speech Communication Association >Speaker-Independent Emotion Recognition based on Feature VectorClassification
【24h】

Speaker-Independent Emotion Recognition based on Feature VectorClassification

机译：基于特征矢量的扬声器的情感识别

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

This paper proposes a new feature vector classification forspeech emotion recognition. The conventional feature vectorclassification applied to speaker identification categorizedfeature vectors as overlapped and non-overlapped. Thismethod discards all of the overlapped vectors in modeltraining, while non-overlapped vectors are used toreconstruct corresponding speaker models. Although theconventional classification showed strong performance inspeaker identification, it has limitations in constructingrobust models when the number of overlapped vectors issignificantly increased such as in emotion recognition. Toovercome such a drawback, we propose a more sophisticatedclassification method which selects discriminative vectorsamong overlapped vectors and adds the vectors in modelreconstruction. On experiments based on an LDC emotioncorpus, our classification approach exhibited superiorperformance when compared to the conventional method.
机译：本文提出了一种新的特征向量分类，陷入困境的情感识别。传统的特征矢量应用于扬声器识别分类的载体，其重叠和非重叠。 Thismethod丢弃了模型中的所有重叠矢量，而非重叠的矢量使用Toreconstruct相应的扬声器模型。虽然强化分类表现出强大的性能验证器识别，但由于在诸如情感认可之类的重叠矢量的数量增加的重叠矢量的数量增加时，它的施工型巨型模型有局限性。 Toolectcome这样的缺点，我们提出了一种更复杂的Classification方法，该方法选择鉴别的Vectorsamong重叠的载体，并在型号中添加了型号的载体。在基于LDC Emotioncorpus的实验中，与传统方法相比，我们的分类方法表现出优越的表现。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

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