Deep regression of social signals in Dyadic Scenarios
Document typeMaster thesis
Rights accessOpen Access
The purpose of this project is to design a general system for emotion recognition through social signals in dyadic using deep learning methods using raw data from audio, video and text transcriptions from publicly available database records. The automatic emotion recognition problem has increased the attention in the scientific community considering the multi applications for emotion detection but also to design more accurate and complex empathic machines. During this project are proposed alternatives for utterance representation of multi-modal data generated from text, audio and video, in order to improve the state of the art system for emotion recognition based on deep learning networks. The proposed framework is based in IEMOCAP database but it has a general scope for any multi-modal database. The performance of this system outperforms the state of the art method and delivers an informative analysis concerning the utterance representation quality. Finally, the conclusions of this work are exposed along with potential future lines of work related to emotion recognition systems and emotion representations.
SubjectsNeural networks (Computer science), Machine learning, Xarxes neuronals (Informàtica), Aprenentatge automàtic
DegreeMÀSTER UNIVERSITARI EN INTEL·LIGÈNCIA ARTIFICIAL (Pla 2017)
All rights reserved. This work is protected by the corresponding intellectual and industrial property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public communication or transformation of this work are prohibited without permission of the copyright holder