%%% {{{ header stuff %%% Sat May 7 13:42:08 EDT 1994 % Edited from the B exam: % %Sat Oct 23 13:35:13 EDT 1993 % Slides for talk at ORST. % Added slides to the talk given at Bill Sears. % Talk at ORST given on October 29. \raggedright \newcommand\thesistitle{\large \bf% Audio System for\\ Technical Readings} \newcommand{\alphaeq}{\mathrel{=_\alpha}} %%% }}} %Duration of talk: 50 minutes. %Demo duration: 10 %%% {{{ %Section1: Theme 3 slides, 4 minutes. %Removed applications slide: \title{Slides for RFB talk} %Title, theme and motivation \begin{slide}{}% <(title )> \label{slide:title} \begin{center} \thesistitle{} \end{center} \vfill \centerline{\large \bf T. V. Raman} \centerline{\bf Cornell University} \centerline{\bf DEC Cambridge Research Lab} \vfill \begin{center} {\large \bf Thesis Adviser: D. Gries} \end{center} \end{slide} \begin{slide}{} \slidetitle{What is formatting?} Information has structure. \begin{itemize} \item Mathematical expressions. $$ e^{e^{e^x}} e^{e^x} e^x$$ \item Tables. \[ \begin{array}{|l|l|l|}\hline 1 & 1 & 1\\ & 1 & 1\\ && 1 \end{array} \] \end{itemize} Formatted output conveys this structure. \end{slide} \begin{slide}{} \slidetitle{\mbox{\asterlogo{}}\\ Audio System for\\ Technical Readings} The next-generation reading machine. Information is display independent. \begin{tabular}{lc} & \alltex{} $\rightarrow$ Print \\ $\mbox{\rm\begin{tabular}{|l|}\hline Electronic\\ Documents\\ \hline \end{tabular}}% _{\displaystyle\ \ \searrow}^{\displaystyle\ \ \nearrow}$ & \\ & \asterlogo{} $\rightarrow$ Audio \end{tabular} Different views of the same document. \end{slide} %%% }}} %%% {{{ %Section 2: Overview of project 2 slides, 2 minutes. %Section 2: Overview of project 2 slides \begin{slide}{} \label{slide:overview} \slidetitle{Overview} Overcome the passive nature of listening. \begin{itemize} \item Internalize document structure. \item Render structure in audio. \item Browse information structure. \end{itemize} \end{slide} \begin{slide}{} % <(architecture)> \label{slide:architecture} \slidetitle{Architecture} \begin{center} \Large \fbox{\bf Electronic documents}\\ $\Downarrow\mbox{Recognizer}\Downarrow$\\ \fbox{\bf High-level model}\\ $\Downarrow \mbox{Audio Formatting}\Downarrow$\\ \fbox{\bf Audio Rendering}\\ {\Large $+$}\\ \fbox{Rendering Rules} \\ {\Large $+$}\\ \fbox{Browsing}\\ {\Large $=$}\\ \fbox{\bf Audio Documents} \end{center} \end{slide} %%% }}} %%% {{{ %section 3: Recognition: 5 slides 5 minutes %section 3: Recognition: 4 slides \begin{slide}{} \slidetitle{Recognition Is Hard} \begin{itemize} \item A document is {\bf not} a computer program. \item Written math is ambiguous. $$\sin 2x = 2\sin x \cos x$$ \item Standard parsing techniques can fail. $$ A= \{x \vert\quad x^2 \in (0,1]\} $$ $$D=\{x \vert \quad { 1 \leq x} \land { x \vert y}\} $$ \end{itemize} \end{slide} \begin{slide}{} % <(recognition of structure )> \label{slide:recognition_of_structure} \slidetitle{Recognition of structure} High-level models: \begin{itemize} \item {\bf Tree structure} for text. \item {\bf Quasi-prefix form} for mathematics. \end{itemize} A node in the quasi-prefix form: \medskip \begin{center} \begin{tabular}{|rcl|}\hline left-sup & accent & superscript \\ &$\nwarrow$ \hfill $\uparrow$ \hfill $\nearrow$ & \\ & {\large \bf math object } & \\ & $\swarrow$ \hfill $\downarrow$ \hfill $\searrow$ & \\ left-sub & underbar & subscript\\ \hline \end{tabular} \end{center} \medskip Captures ambiguous written mathematics. \end{slide} \begin{slide}{} \label{slide:example_article} \slidetitle{Example: Article} \begin{center}\large \begin{verbatim} (article (title ) (author ) (abstract ) (section (title
)
(subsection ...) ...) (references ) ) \end{verbatim} \end{center} \end{slide} \begin{slide}{} % <(example: quasi-prefix)> \label{slide:example_quasi} \slidetitle{Example: Quasi-prefix form} $$D_x^1$$ is represented as : \begin{center} \begin{verbatim} [MATH CONTENTS: "D" CHILDREN: NIL ATTR: ([SUBSCRIPT VALUE: [MATH CONTENTS: "x" CHILDREN: NIL ATTR: NIL ]] [SUPERSCRIPT VALUE: [NUMBER CONTENTS: "1" ATTR: NIL]]) ] \end{verbatim} \end{center} \end{slide} \begin{slide}{} \label{slide:example_macro_def} \slidetitle{Extending the recognizer} \[ (A \kronecker B)^{T} = A^{T} \kronecker B^{T} \] {\small \begin{verbatim} \def\kronecker{\raisebox{1pt}{$ \:\otimes \:$}} \end{verbatim} } Recognizer extended by : \begin{center} \small \begin{verbatim} (define-text-object :macro-name "kronecker" :number-args 0 :processing-function kronecker-expand :precedence multiplication :object-name kronecker-product :super-classes (binary-operator)) \end{verbatim} \end{center} \hfil {\verb|A \kronecker B|}\hfil \centerline{{\large $\Downarrow$}} \centerline{\fbox{\bf Recognizer}} \centerline{\large $\Downarrow$} \centerline{\fbox{(kronecker-product :children( $\mbox{\tt A}'$ $\mbox{\tt B}'$))}} where\hspace*{.1in} $\mbox{\tt A} \stackrel{\mbox{\small recognizer}}{\longrightarrow} \mbox{\tt A}'$. \end{slide} %%% }}} %%% {{{ %section 4: A language for audio formatting: 4 slides, 7 minutes %section 4: A language for audio formatting: 4 slides 6 minutes \begin{slide}{} \label{slide:define_audio_space} \slidetitle{Defining the audio space} Audio renderings: $\mbox{speech} \oplus \mbox{sound}$. Formatter is a Conceptual audio device. \begin{itemize} \item Formatter parameters define audio space. \item Formatter state is a point in this space. \item Formatter has multiple components. \item Components correspond to subspaces. \item Space $=$ direct sum of subspaces. \end{itemize} Changing formatter state $\leftrightarrow$ Moving in audio space. \end{slide} \begin{slide}{} \label{slide:audio_formatting_language} \slidetitle{Audio Formatting Language} {\bf AFL} is a block structured language. {\bf AFL} variables encapsulate formatter state. Operators generate points in audio space. \begin{itemize} \item {\bf NEW-BLOCK} Introduce a local scope. \item {\bf LOCAL-SET-STATE} Assignment. \item {\bf GLOBAL-SET-STATE} Assignment. \end{itemize} \end{slide} \begin{slide}{} \label{slide:afl_blocks} \slidetitle{AFL Blocks } Create local copies of all component states. Effect of local assignments seen only within the closest enclosing block. Allows rendering rules to call one another. {\bf AFL} block is an implicit cobegin statement. \end{slide} \begin{slide}{} \label{slide:component_spaces} \slidetitle{Component spaces } Dimensions in the speech space:\\ Average-pitch, smoothness, directionality, $\ldots$ Operators: move-by, move-to, step-by, $\ldots$ Dimensions in the sound space:\\ volume, sound, $\ldots$ Operators: select-sound, switch-on, $\ldots$ \end{slide} %%% }}} %%% {{{ %section 5: Audio notation 8 slides, 11 minutes (demos) %section 5: Use of AFL in audio documents 9 slides \begin{slide}{} \label{slide:producing_audio_renderings} \slidetitle{Producing audio renderings} $$\fbox{\mbox{Recognizer }} \stackrel{\mbox{\bf AFL Rules}} {\longrightarrow} \fbox{\mbox{Audio Renderings}}$$ A typical rendering rule: \begin{center} \begin{verbatim} BEGIN-AFL-BLOCK set local state; render object END-AFL-BLOCK \end{verbatim} \end{center} Nested objects rendered relative to parent. \end{slide} \begin{slide}{} \slidetitle{Mathematical notation } Written notation: \begin{itemize} \item Linearizes implicit tree structure. \item Annotates this with attributes. \end{itemize} A parallel audio notation: Independent audio dimensions cue features. Pick a monotonic change of state function $f$ to vary audio state $P$ along these dimensions, where the ordering $P < f(P) < f^2(P) < \cdots$ is perceptible. This is sufficient to cue nested structures. \end{slide} \begin{slide}{} \slidetitle{Superscripts and subscripts} $$x^{k_1} + x^{k_2} + x^{k_3} + \cdots + x^{k_n} = 0$$ $$x_{k^1}+x_{k^2}+x_{k^3}+\cdots+x_{k^n}=0$$ $$ e^{e^{e^x}} e^{e^x} e^x$$ \end{slide} \begin{slide}{} \slidetitle{Tree structure} Move along a line in speech space spanned by six dimensions. $$\frac{a+b}{c+d}$$ $$\frac{a}{b+c+d}$$ $$a+\frac{b+c}{d+e}+x$$ Can cue deeply nested structures. \[ 1+ {x \over {\scriptstyle 1+ {\scriptstyle x \over {\scriptstyle 1 + {\scriptstyle x \over {\scriptstyle 1 + {\scriptstyle x \over {\scriptstyle 1+ {\scriptstyle x \over 1+{\scriptstyle \atop \ddots }}}}}}}}}} \] \end{slide} \begin{slide}{} \slidetitle{Complex notation} Features can be combined:\\ Tree-structure within superscript. $$x+y^{2\over k+1}$$ Superscript within tree-structure. \[ \sqrt{e^{e^{e^x}} + \sqrt{e^{e^x} + \sqrt{e^x}}} \] This system of audio notation scales very well. \end{slide} \begin{slide}{} % <(stereo effects)> \label{slide:stereo_effects} \slidetitle{Stereo Effects} Directionality is an audio dimension.\\ Moving along it produces stereo. Let L be the line spanned by the dimensions left-volume and right-volume. {\small \begin{verbatim} compute-step-size(length(row),L); BEGIN-AFL-BLOCK set speech state to left speaker; LOOP for element in row do render element ; step by 1 along L ENDLOOP END-AFL-BLOCK \end{verbatim} } $$A=\pmatrix{a_{1 1}&a_{1 2}&\ldots&a_{1 n}\cr a_{2 1}&a_{2 2}&\ldots&a_{2 n}\cr \vdots&\vdots&\ddots&\vdots\cr a_{m 1}&a_{m 2}&\ldots&a_{m n}\cr}$$ \end{slide} \begin{slide}{} \slidetitle{Classification of audio cues} Speech and non-speech audio cues. Also classified as: \begin{itemize} \item {\bf Fleeting cue} Does not last. \item {\bf Persistent cue} Persists across time. \end{itemize} Fleeting cues identify objects. Persistent cues convey environment. \end{slide} \begin{slide}{} % <( reading styles)> \label{slide:reading_styles} \slidetitle{Rendering Styles} Rendering rules with same name (for different objects) form a style.\\ List of active styles: \begin{center} (special-patterns descriptive simple default) \end{center} \begin{displaytext} Alpha-equality commutes with substitution. If $t \alphaeq t'$ and $s \alphaeq s'$, then $$\subst{t}{s}{x} \alphaeq \subst{t'}{s'}{x}$$. \end{displaytext} Active rule overrides active style. Activate rules to customize styles. \end{slide} %%% }}} %%% {{{ Variable substitution: 1 slide 4 minutes %Demo slides made separately using lslides.sty. % file: b-exam-var-subst.tex %Demo slides inserted here. %Skip 1 slides \addtocounter{slide}{1} %%% }}} %%% {{{ %section 6: Conclusion 3 slides, 3 minutes %Still need to write a good conclusion slide? %section 6: Conclusion 3 slides \begin{slide}{} \label{slide:performance} \slidetitle{Performance} Audio formats both complete and partial text books, technical reports and {\bf AMS} Bulletins in real time. Text books include: \begin{itemize} \item Nonlinear Optimization: S. Vavasis. \item Matrix Computation: Golub, Van Loan. \item Analysis of Algorithms: Kozen. \item Discrete Mathematics: Gries, Schneider. \item Commutative Ring Theory: Kaplansky. \end{itemize} \end{slide} \begin{slide}{} \slidetitle{Contributions} \begin{itemize} \item Enable immediate availability of technical documents in audio. \item Internalize written mathematics. \item Render such structure orally. \item Enable active listening. \end{itemize} \end{slide} \begin{slide}{} \slidetitle{Next Step} \begin{itemize} \item Start mass-producing audio documents on cassette. \item Port \aster{} to other platforms. \end{itemize} Research: \begin{itemize} \item An editing tool for technical documents. \end{itemize} \end{slide} %%% }}} %%% {{{ end of file %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% local variables: %% folded-file: t %% end: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%% }}}