Learning Discriminative Representations for Speech Emotion Features using Convolutional Neural Network

カテゴリ: 部門大会

論文No: SS3-4

グループ名: 【C】平成28年電気学会電子・情報・システム部門大会講演論文集

発行日: 2016/08/31

タイトル(英語): Learning Discriminative Representations for Speech Emotion Features using Convolutional Neural Network

著者名: 暢 雁楠(早稲田大学),吉江 修(早稲田大学)

著者名(英語): Yannan Chang|Osamu Yoshie

キーワード: Speech Emotion Recognition|Convolutional Neural Networks|Deep Learning

要約(日本語): As an essential way of human-computer cognitions, increasing attention has been directed to the study of Speech Emotion Recognition (SER). Although advances have been made in research of SER, a challenging problem is that how to choose efficient features for this task is difficult. In our research, based on a deep learning method of convolutional neural networks (CNN), we propose to involve both of pooling feature-maps and convolutional feature-maps into the fully connected layer to learn Discriminative Emotional Features (DEF). In our research, rough features are learned in the lower layers, emotion-specific features are extracted in the higher layers, and support vector machine (SVM) is used to make the classification.

PDFファイルサイズ: 1,202 Kバイト