Yutong Zhao

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2021-94

May 14, 2021

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-94.pdf

The absolute quantity of content on multimedia platforms such as social networks has grown exponentially in the past few years. As such, algorithms for automatic evaluation for digital content become increasingly important in enhancing the relevancy of retrieval, especially the ability to capture abstractions of more abstract aspects such as beauty, artistry, interest- ingness. Particularly, we see a need for models that can evaluate abstract notions of digital content such as style and memorability.

While there has been development and achievement in using artificial neural networks for predicting image memorability as well as for image style transfer, such success have not completely been replicated in different media modalities such as video or audio. Given the broad topic of multimedia, this report seeks to focus on the background and context of audio features in a framework of multimedia evaluation. We also present our experiments for establishing use of audio features in both predicting video memorability as well as in audio style transfer. Furthermore, this work outlines both current challenges as well as the path forward for systematic augmentation and optimization of multimedia attributes such as memorability and style.

Advisors: Gerald Friedland

---

BibTeX citation:

@mastersthesis{Zhao:EECS-2021-94,
    Author= {Zhao, Yutong},
    Title= {On Memorability and Style of Audio Features in Multimedia Evaluation},
    School= {EECS Department, University of California, Berkeley},
    Year= {2021},
    Month= {May},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-94.html},
    Number= {UCB/EECS-2021-94},
    Abstract= {The absolute quantity of content on multimedia platforms such as social networks has grown exponentially in the past few years. As such, algorithms for automatic evaluation for digital content become increasingly important in enhancing the relevancy of retrieval, especially the ability to capture abstractions of more abstract aspects such as beauty, artistry, interest- ingness. Particularly, we see a need for models that can evaluate abstract notions of digital content such as style and memorability.

While there has been development and achievement in using artificial neural networks for predicting image memorability as well as for image style transfer, such success have not completely been replicated in different media modalities such as video or audio. Given the broad topic of multimedia, this report seeks to focus on the background and context of audio features in a framework of multimedia evaluation. We also present our experiments for establishing use of audio features in both predicting video memorability as well as in audio style transfer. Furthermore, this work outlines both current challenges as well as the path forward for systematic augmentation and optimization of multimedia attributes such as memorability and style.},
}

EndNote citation:

%0 Thesis
%A Zhao, Yutong 
%T On Memorability and Style of Audio Features in Multimedia Evaluation
%I EECS Department, University of California, Berkeley
%D 2021
%8 May 14
%@ UCB/EECS-2021-94
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-94.html
%F Zhao:EECS-2021-94