/
10-02-ACA-Alignment.tex
282 lines (264 loc) · 12.9 KB
/
10-02-ACA-Alignment.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
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
% move all configuration stuff into includes file so we can focus on the content
\input{include}
\subtitle{module 10.2: audio-to-audio \& audio-to-score alignment}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\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=6331124}{Chapter 7: Alignment} (pp.~146--150)
sections~10.2 \& 10.3
\end{block}
\begin{itemize}
\item \textbf{lecture content}
\begin{itemize}
\item Audio-to-Audio alignment
\begin{itemize}
\item use cases
\item features
\item distance measures
\item typical accuracy
\end{itemize}
\item Audio-to-Score alignment
\end{itemize}
\bigskip
\item<2-> \textbf{learning objectives}
\begin{itemize}
\item elaborate on possible use cases for audio-to-audio alignment
\item give examples for features and distance measures for alignment
\item discuss differences between audio-to-audio and audio-to-score alignment
\end{itemize}
\end{itemize}
\inserticon{directions}
\end{frame}
\section[A2A]{Audio to audio alignment}
\begin{frame}{audio-to-audio alignment}{introduction}
\begin{itemize}
\item \textbf{objective}
\begin{itemize}
\item align two sequences of audio
\end{itemize}
\bigskip
\item<2-> \textbf{use cases}
\begin{itemize}
\item \textit{quick browsing} for certain parts in recordings
\item \textit{timing adjustment} (backing vocals, loops, \ldots)
\item \textit{automated dubbing}
\item \textit{musicological analysis} (relative timing of several performances)
\end{itemize}
\bigskip
\item<3-> \textbf{processing steps}
\begin{itemize}
\item extract suitable features
\item compute distance matrix
\item compute alignment path
\end{itemize}
\end{itemize}
\end{frame}
\begin{frame}{audio-to-audio alignment}{alignment path computation}
\vspace{-3mm}
\begin{columns}
\column{.4\linewidth}
\vspace{4mm}
\begin{itemize}
\item prerequisite:\\ Module 10.1~---~Dynamic Time Warping
\end{itemize}
\column{.6\linewidth}
\vspace{-13mm}
\figwithmatlab{DtwPath}
\end{columns}
\end{frame}
\begin{frame}{audio-to-audio alignment}{features}
\vspace{-2mm}
\only<1-2>{
\begin{itemize}
\item \textbf{use case examples}
\begin{itemize}
\item \textbf{quick browsing} --- find the same part across files\\
$\Rightarrow$ use \textit{pitch based} features
\item \textbf{timing adjustment} --- backing vocals to lead vocals\\
$\Rightarrow$ use \textit{intensity based} features
\item \textbf{automated dubbing} --- same speaker several recordings\\
$\Rightarrow$ use \textit{intensity based} and \textit{timbre based} features
\end{itemize}
\smallskip
\item<2-> \textbf{feature categories}
\begin{itemize}
\item \textbf{intensity}: energy, onset probability, \ldots
\item \textbf{tonal}: pitch chroma, \ldots
\item \textbf{timbral}: MFCCs, spectral shape, \ldots
\end{itemize}
\item<2->[] plot from \footfullcite{kirchhoff_evaluation_2011}
\end{itemize}
}
\only<3>{
\begin{figure}
\centerline{\includegraphics[scale=.5]{graph/ATAA_features}}
\end{figure}
}
\end{frame}
\begin{frame}{audio-to-audio alignment}{feature-dependency of results}
\figwithmatlab{DtwFeatures}
features (left to right): pitch chroma, RMS, MFCCs
\end{frame}
\begin{frame}{audio-to-audio alignment}{compute distance matrix --- distance measures}
\begin{itemize}
\item typical distance measures
\only<1-4>{
\begin{itemize}
\item<1-4> \emph{{Euclidean distance}}:
$d_\mathrm{E}(s) = \sqrt{\sum\limits_{j = 0}^{11}{\big(\nu_\mathrm{e}(j)-\nu_\mathrm{t,s}(j)\big)^2}} $
\item<2-4> \emph{{Manhattan distance}}:
$d_\mathrm{M}(s) = \sum\limits_{j = 0}^{11}{\big|\nu_\mathrm{e}(j)-\nu_\mathrm{t,s}(j)\big|} $
\item<3-4> \emph{{Cosine distance}}:
$d_\mathrm{C}(s) = 1-\left( \frac{\sum\limits_{j = 0}^{11}{\nu_\mathrm{e}(j)\cdot\nu_\mathrm{t,s}(j)}}{\sqrt{\sum\limits_{j = 0}^{11}{\nu_\mathrm{e}(j)^2}}\sqrt{\cdot \sum\limits_{j = 0}^{11}{\nu_\mathrm{t,s}(j)^2}}}\right) $
\item<4-4> \emph{{Kullback-Leibler divergence}}:
$d_\mathrm{KL}(s) = \sum\limits_{j = 0}^{11}{\nu_\mathrm{e}(j)\cdot\log\left(\frac{\nu_\mathrm{e}(j)}{\nu_\mathrm{t,s}(j)}\right)}$
\end{itemize}
}
\smallskip
\item<5-> data-driven approach: train classifier with 2-class problem \footfullcite{kirchhoff_evaluation_2011}
\only<5>{
\begin{figure}
\centerline{\includegraphics[scale=.5]{graph/ATAA_feature_training}}
\end{figure}
}
\only<6>{
\begin{figure}
\centerline{\includegraphics[scale=.5]{graph/ATAA_simmatrix_compare}}
\end{figure}
}
\end{itemize}
\end{frame}
\begin{frame}{audio-to-audio alignment}{typical results}
\vspace{-5mm}
\begin{columns}
\column{.7\linewidth}
\vspace{-3mm}\begin{figure}
\centerline{\includegraphics[width= \columnwidth]{graph/ATAA_results}}
\end{figure}
\column{.3\linewidth}
\begin{table}
\begin{tabular}{p{.55\columnwidth}l}
\textbf{originals}\linebreak
\begin{scriptsize}
left:~instrumental\hfill \linebreak
right:~a capella
\end{scriptsize} & \textbf{synced}\\
\includeaudio{originals_splanky} & \includeaudio{synced_splanky}
\end{tabular}
\end{table}
\end{columns}
\hphantom{\footfullcite{kirchhoff_evaluation_2011}}
\inserticon{audio}
\end{frame}
\section[A2S]{Audio to score alignment}
\begin{frame}{audio-to-score alignment}{overview}
\begin{itemize}
\item \textbf{objective}
\begin{itemize}
\item align an audio sequence with a score sequence
\end{itemize}
\bigskip
\item<2-> \textbf{use cases}
\begin{itemize}
\item score viewer
\item music education
\item retrieve matching score/audio via cost function
\item musicological analysis
\end{itemize}
\bigskip
\item<3-> \textbf{processing steps}
\begin{itemize}
\item see audio-to-audio alignment
\end{itemize}
\end{itemize}
\end{frame}
\begin{frame}{audio-to-score alignment}{challenges}
\begin{itemize}
\item features from \textbf{different domains}
\begin{itemize}
\item score contains no timbre info
\item score cannot be expected to contain no loudness info
\item score has no clear ``time axis''
\end{itemize}
\bigskip
\item<2->[$\Rightarrow$] two prototypical for distance/similarity calculation
\begin{itemize}
\item \textit{approach 1}: convert score into audio-like representation
\begin{itemize}
\item MIDI-to-audio
\item use model synthesize
\end{itemize}
\smallskip
\item \textit{approach 2}: convert audio into score-like representation
\begin{itemize}
\item audio-to-MIDI
\item pitch chroma
\item event-based segmentation
\end{itemize}
\end{itemize}
\end{itemize}
\end{frame}
\section[eval]{evaluation}
\begin{frame}{alignment}{evaluation}
\begin{itemize}
\item \textbf{goal}: compare two sequences of time stamps
\bigskip
\item<2-> evaluation \textbf{challenges}
\begin{itemize}
\item pauses/rests, and long held notes: what is the reference path?
\item noise in the begin and end of the recording
\item data not easily available
\begin{itemize}
\item synthesized
\item piano sensors
\item pseudo-ground truth with time stretching
\item automatic annotation with quality assurance
\end{itemize}
\end{itemize}
\end{itemize}
\end{frame}
\begin{frame}{alignment}{evaluation metrics}
\begin{itemize}
\item \textit{audio-to-score}
\begin{itemize}
\item missed note rate
\item misalign rate
\item piece completion
\item average absolute deviation
\item variance of deviation
\end{itemize}
\bigskip
\item \textit{audio-to-audio}
\begin{itemize}
\item mean deviation
\item mean absolute deviation
\item maximum deviation
\item relative number of matching path points
\end{itemize}
\end{itemize}
\end{frame}
\section{summary}
\begin{frame}{summary}{lecture content}
\begin{itemize}
\item \textbf{audio-to-audio alignment}
\begin{enumerate}
\item extract features
\item create distance matrix with suitable distance measure
\item use DTW to find alignment path
\item (use time-stretching to actually align the sequences)
\end{enumerate}
\bigskip
\item \textbf{audio-to-score alignment}
\begin{enumerate}
\item extract usually pitch-based features
\item distance measure
\item use DTW, HMM, etc to extract alignment path
\end{enumerate}
\end{itemize}
\inserticon{summary}
\end{frame}
\end{document}