請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/74317
標題: | 深度學習於多模態情感辨別——注意力截斷循環神經網絡 Deep Learning for Multimodal Emotion Recognition -- Attentive Residual Disconnected RNN |
作者: | Erick Chandra 章智傑 |
指導教授: | 許永真(Jane Yung-jen Hsu) |
關鍵字: | Emotion Recognition,Disconnected Recurrent Neural Network,Attention Mechanism,Residual Network, |
出版年 : | 2019 |
學位: | 碩士 |
摘要: | Human communicates using verbal and non-verbal cues. One of the most essential elements that complements the understanding of communication is emotion. Emotion is expressed not only in words, but also facial expressions, body language, tone, etc. Therefore, we formulate the emotion recognition as a multimodal task.
Emotions are usually described in a sequence along with the utterances. In recent years, RNN-based models have been known to be good at modeling the entire sequence and capturing long-term dependencies. However, it lacks the ability to extract local key patterns and position-invariant features. Hence, we adopt Deep Attentive Residual Disconnected RNN model which incorporates the concept from both RNN and CNN to enhance the ability to capture spatial and temporal features. We utilize CMU MOSEI dataset comprising of language, visual, and acoustic modalities for training and evaluating our model. The results show that Deep Attentive Residual Disconnected RNN model outperforms the baseline. Besides, the use of multimodal approach also solidifies the recognition better compared to those of single modalities. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/74317 |
DOI: | 10.6342/NTU201901646 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-108-1.pdf 目前未授權公開取用 | 1.51 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。