/
13-00-ACA-Mood.tex
216 lines (202 loc) · 9.78 KB
/
13-00-ACA-Mood.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
% move all configuration stuff into includes file so we can focus on the content
\input{include}
\subtitle{module 13: mood recognition}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}
% generate title page
\input{include/titlepage}
\section[overview]{lecture overview}
\begin{frame}{introduction}{overview}
\begin{block}{corresponding textbook section}
%\href{http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6331125}{Chapter 8: Musical Genre, Similarity, and Mood} (pp.~158--161)
chapter~13
\end{block}
\begin{itemize}
\item \textbf{lecture content}
\begin{itemize}
\item introduction to emotion and mood
\item models for mood
\item linear regression
\end{itemize}
\bigskip
\item<2-> \textbf{learning objectives}
\begin{itemize}
\item describe Russel's arousal-valence plane
\item discuss commonalities and differences between mood recognition and genre classification
\item implement linear regression in Matlab
\end{itemize}
\end{itemize}
\inserticon{directions}
\end{frame}
\section[intro]{introduction}
\begin{frame}{mood recognition}{introduction}
\begin{itemize}
\item \textbf{objective}:identify mood/emotion of a song
\item<2-> \textbf{terminology}:
\begin{itemize}
\item \textit{Music Mood Recognition} and \textit{Music Emotion Recognition} usually used synonymously
\end{itemize}
\bigskip
\item<3-> \textbf{processing steps} (similar to genre and similarity tasks)
\begin{itemize}
\item extract features
\item classify (possibly regression)
\end{itemize}
\end{itemize}
\end{frame}
\section[mood]{mood \& emotion}
\begin{frame}{mood recognition}{challenges}
\question{What is the difference between \textit{mood} and \textit{emotion}}
many definitions out there but general consensus on
\begin{itemize}
\item \textit{emotion}:
\begin{itemize}
\item temporary, evanescent
\item (directly) related to external stimuli
\end{itemize}
\item \textit{mood}:
\begin{itemize}
\item longer term, stable
\item diffuse affect state
\end{itemize}
\end{itemize}
\end{frame}
\begin{frame}{mood recognition}{challenges}
\begin{itemize}
\item \textbf{ground truth data}
\begin{itemize}
\item \textit{verbalization} of emotions/moods usually misleading
\item not easily \textit{quantifiable}/categorizable
\item changing over time?
\end{itemize}
\bigskip
\item \textbf{research questions}
\begin{itemize}
\item<2-> are \textit{basic emotions} (happiness, anger, fear, \ldots) representative for music perception?
\item<2-> should aesthetic emotions be distinguished from other emotions (guilt, shame, disgust, ...)?
\item<3-> \textit{aroused vs.\ transported}/\textit{evoked vs.\ conveyed} moods?
\end{itemize}
\end{itemize}
\end{frame}
\section[models]{mood models}
\begin{frame}{mood recognition}{models}
\vspace{-3mm}
\begin{itemize}
\item classification into \textbf{label clusters}
\only<1>{\footfullcite{hu_exploring_2007}
\begin{scriptsize}
\begin{tabular}{ccccc}
\\ \hline
\bf{\emph{Cluster 1}} & \bf{\emph{Cluster 2}} & \bf{\emph{Cluster 3}} & \bf{\emph{Cluster 4}} & \bf{\emph{Cluster 5}}\\
\hline
\bf{\textnormal{Rowdy}} & Amiable/Good Natured & Literate & Witty & Volatile\\
\bf{\textnormal{Rousing}} & Sweet & Wistful & Humorous & Fiery\\
\bf{\textnormal{Confident}} & Fun & Bittersweet & Whimsical & Visceral\\
\bf{\textnormal{Boisterous}} & Rollicking & Autumnal & Wry & Aggressive\\
\bf{\textnormal{Passionate}} & Cheerful & Brooding & Campy & Tense/Anxious\\
\bf{} & & Poignant & Quirky & Intense\\
\bf{} & & & Silly & \\
\end{tabular}
\end{scriptsize}
}
\item<2-> \textbf{mood model}, circumplex model
\only<2>{\footfullcite{russel_circumplex_1980}
\begin{figure}
\centering
\scalebox{.8}{\input{pict/genre_moodmodel}}
\end{figure}
}
\end{itemize}
\end{frame}
\section{regression}
\begin{frame}{mood recognition}{mood model: regression modeling}
\begin{itemize}
\item \textbf{mapping}
\begin{itemize}
\item (N-dimensional) observation (feature) to 2-dimensional coordinate (valence/arousal)
\end{itemize}
\bigskip
\item \textbf{training}
\begin{itemize}
\item find model to minimize error between data points and ``prediction''
\end{itemize}
\end{itemize}
\end{frame}
\begin{frame}{regression}{regression}
\begin{itemize}
\item linear regression: fit a linear function to a series of points $(x_j,y_j)$
\begin{equation*}
y_n = m\cdot x_n + b
\end{equation*}
\figwithmatlab{LinearRegression}
\item other regression approaches: SVR, DNNs, etc.
\end{itemize}
\end{frame}
%\begin{frame}{linear regression}{introduction to regression 2/2}
%\begin{itemize}
%\item minimize error between model and data (here: least squares)
%\end{itemize}
%\begin{scriptsize}
%\begin{eqnarray*}
%e_n^2 &=& (y_n - mx_n - b)^2\\
%E &=& \sum (y_n - mx_n - b)^2\\
%\end{eqnarray*}
%\vspace{-10mm}
%\begin{columns}[T]
%\column{.5\linewidth}
%\begin{eqnarray*}
%\visible<2->{\frac{\partial E}{\partial b} = \sum -2(y_n - mx_n - b) &=& 0}\\
%\visible<3->{-2\sum y_n +2\sum mx_n +2\sum b &=& 0}\\
%\visible<4->{\sum mx_n + \sum b &=& \sum y_n}\\
%\visible<5->{m\sum x_n + \mathcal{N}b &=& \sum y_n}
%\end{eqnarray*}
%\column{.5\linewidth}
%\begin{eqnarray*}
%\visible<2->{\frac{\partial E}{\partial m} = \sum -2x_n(y_n - mx_n - b) =& 0&}\\
%\visible<3->{-2\sum x_ny_n +2\sum mx_n^2 +2\sum bx_n =& 0&}\\
%\visible<4->{\sum mx_n^2 + \sum bx_n = \sum x_ny_n&&}\\
%\visible<5->{m\sum x_n^2 + b\sum x_n = \sum x_ny_n&&}
%\end{eqnarray*}
%\end{columns}
%\bigskip
%\visible<6>{
%\begin{eqnarray*}
%&\Rightarrow&\\
%m &=& \frac{\mathcal{N}\sum x_ny_n-\sum x_n\sum y_n}{\mathcal{N}\sum x_n^2 -\left(\sum x_n\right)^2}\\
%b &=& \frac{\sum y_n}{\mathcal{N}}-m\frac{\sum x_n}{\mathcal{N}}\\
%\end{eqnarray*}
%}
%\end{scriptsize}
%\end{frame}
\section{results}
\begin{frame}{mood recognition}{range of results}
\begin{itemize}
\item highly dependent on data
\pause
\item \textbf{5 mood clusters}:\\ 40--60\% classification rate
\bigskip
\item \textbf{mood model}:\\ 0.1--0.4 absolute prediction error (unit circle)
\end{itemize}
\end{frame}
\section{summary}
\begin{frame}{summary}{lecture content}
\begin{itemize}
\item \textbf{emotion and mood}
\begin{itemize}
\item emotion: temporary, related to external stimuli
\item mood: long term, diffuse affective state
\end{itemize}
\bigskip
\item \textbf{features}
\begin{enumerate}
\item baseline features are identical to genre and similarity tasks
\end{enumerate}
\bigskip
\item \textbf{inference}
\begin{enumerate}
\item often done as regression (as opposed to classification)
\end{enumerate}
\end{itemize}
\inserticon{summary}
\end{frame}
\end{document}