Permalink
Browse files

additional tweaks to intro

  • Loading branch information...
1 parent e02f227 commit 22dbd092c8a0143f7680cd56a959f5701bd336f9 @DavidEscott committed Apr 24, 2012
Showing with 15 additions and 24 deletions.
  1. +15 −24 final_report/final.tex
View
@@ -67,30 +67,21 @@ \section{INTRODUCTION}
they see it is not surprising that machine learning algorithms
struggle to identify emotions exhibited by individuals.
-The use of multi-modal features is common in the Human Emotion
-analysis literature. The IEMOCAP dataset is only one such example, and
-researchers have found that some facial components such as cheek,
-mouth, forehead, and eyebrow movements correspond more strongly than
-others to specific emotional expressions. However this data is almost
-always accompanied by audio data which provides its own modality and
-information content.
-
-% I did not understand this paragraph on first reading. I'm just
-% guessing on what the mower paper is all about
-The proper analysis of multiple input modalities is crucial to proper
-comprehension in speech perception, as seen in the McGurk
-effect\cite{mcgurk-effect}. In \cite{mower-mcgurk}, the authors extend
-this to the area of emotion perception; by testing if mismatched audio
-and visual features affect emotional perception of human
-observers. These results demonstrate the importance of incorporating
-multiple modalities in order to replicate human emotional
-perception. While the relationship between multi-modal features has
-been recognized in the literature, the challenge of finding a
-consistent logical way of combining audio (human voice) and video
-(motion-capture data) features in emotion classification problems is
-still an open problem, and motivates this paper as we explore
-techniques to best incorporate multi-modal features to improve emotion
-classification performance.
+The commonly used emotional data sets often include multiple
+modalities of features that describe physiological, auditory, or
+visual information. The proper analysis of multiple input modalities
+is known to be crucial to proper comprehension in speech perception,
+as seen in the McGurk effect\cite{mcgurk-effect}. In
+\cite{mower-mcgurk}, the authors extend this to the area of emotion
+perception; by testing if mismatched audio and visual features affect
+emotional perception of human observers. These results demonstrate the
+importance of incorporating multiple modalities in order to replicate
+human emotional perception. While the relationship between multi-modal
+features has been recognized in the literature, the challenge of
+finding a consistent logical way of combining auditory and visual
+features in emotion classification problems is still an open
+problem. In this paper we explore techniques to learn multi-modal
+features that will improve emotion classification.
% 2) present the problem in detail CURRENT WORK
Most research in emotion recognition employs feature selection methods

0 comments on commit 22dbd09

Please sign in to comment.