CA2115210C - Interactive computer system recognizing spoken commands - Google Patents

Abstract

An interactive computer system having a processor executing a target computer program, and having a speech recognizer for converting an utterance into a command signal for the target computer program. The target computer program has a series of active program states occurring over a series of time periods. At least a first active-state image is displayed for a first active state occurring during a first time period. At least one object displayed in the first active-state image is identified, and a list of one or more first active-state commands identifying functions which can be performed in the first active state of the target computer program is generated from the identified object. A first active-state vocabulary of acoustic command models for the first active state comprises the acoustic command models from a system vocabulary representing the first active-state commands. A speech recognizer measures the value of at least one feature of an utterance during each of a series of successive time intervals within the first time period to produce a series of feature signals. The measured feature signals are compared to each of the acoustic command models in the first active-state vocabulary to generate a match score for the utterance and each acoustic command model. The speech recognizer outputs a command signal corresponding to the command model from the first active-state vocabulary having the best match score.

Description

Y09-93-030 1 ~521~

INTERACTIVE COII~ K SYSTEM RECOGNIZING SPOKEN COMMANDS

Background of the Invention The invention relates to interactive computer systems in which a user provides commands to a target computer program executing on the computer system by way of an input device. The input device may be, for example, a keyboard, a mouse device, or a speech recognizer. For each input device, an input signal generated by the input device is translated into a form usable by the target computer program.

An interactive computer system in which the user can provide commands by speaking the commands may consist of a processor executing a target computer program having commands identifying functions which can be performed by the target computer program.
The computer system further includes a speech recognizer for recognizing the spoken commands and for outputting command signals corresponding -to the recognized commands. The speech recognizer recognizes a spoken command by measuring the value of at least one feature of an utterance ~uring each of a series of successive time intervals to produce a series of feature signals, comparing the measured featured signals to each of a plurality of acoustic command models to generate a match score for the utterance and each acoustic command model, and outputting a command signal corresponding to the command model having the best match score.

The set of utterance models and words represented by the utterance models which the speech recognizer can recognize is referred to as the system vocabulary. ~he system vocabulary is finite and may, for example, range from one utterance model to thousands of utterance models. Each utterance model may represent one word, or may represent a combination of two or more words spoken continuously (without a pause between the words).

The system vocabulary may contain, for example, utterance models of all of the cc ~n~ to which the target computer program is capable of responding. However, as the number of utterance models '~
~"'~'' :
:
Y09-93-030 2 ~521~
...

.
increases, the time re~uired to perform utterance recognition using the entire sys-tem vocabulary increases, and the recognition accuracy decreases.
;
Generally, a target computer program has a series of active states occurring o~er a series of time periods. For each active state, there may be a list of active state commands identifying functions which can be performed in the active state. The active state co ii~nds may be a small subset of the system vocabulary. The translation of an uttered command to a form usable by the target computer program in one state of the target computer program may be different from the translation of the same command in another state of the target computer program.

In order to improve the speed and accuracy of the speech recognizer, it is desirable to restrict the active vocabulary of utterance models which the speech recognizer can recognize in any given time period to the active state comm~n~ identifying functions which can be performed by the target computer program in that time period. To attempt to achieve this result, the speech recognizer may be provided with a finite state machine which ' duplicates the active s-tates and transitions between active states of the target computer program.

In practice, it has been found impossibl.e to build a finite state machine for the speech recognizer which exactly duplicates the active states and transitions between active states of the target computer program. The target computer program not only interacts with the user, but also interacts with data and other devices of the computer sy~tem whose states cannot be known in advance.
.~ .
For example, a command to load a file will cause a computer program to make a transition to one state if the file exists, or to a different state if the file does not exist. ~owever, the speech recognizer finite state machine must be built with some assumption r ~ that the file exists or does not exist. If a command to load a file is spoken to the computer program using the speech recognizer, then the speech recognizer finite state machine may or may not track the computer program state correctly, depending on whether ( Yo9-93-030 3 ~ 2 1 0 that file exists or does not exist. If the speech recognizer finite state machine assumes that the file exists, but in fact the file does not exist, then the speech recognizer state machine will enter a state different from the state o~ the target computer proyram. ~s a result, the target computer program can no longer receive valid input from the speech recognizer.

Summary of the Invention It is an object of the invention to provide an interactive computer system having a target computer program having a series of active program states occurring over a series of time periods, and having a speech recognizer in which the active vocabulary of commands recognized by the speech recognizer in any given time period is restricted to a list of active commands identifying functions which can be performed by the target computer program in that given time period, without having to predict in advance the states and transitions between states of the target computer program which will occur under all possible circumstances.
~, According to the invention, an interactive computer system comprises a processor executing a target computer pro~ram having a series of active program states occurring over a series of time periods. The target computer program generates active state image data signals representing an active state image for the active state of the target computer program occurring during each time period. Each active state image contains one or more objects.

The interactive computer system further comprises means for displaying at least a first active-state image for a first active state occurring during a first time period. Means are provided for identifying at least one object displayed in the first active-state image, and for generating from the identified object a list of one or more first active-state commands identifying functions which can be performed in the first active state of the target computer program.

Means are also provided for storing a system vocabulary of acoustic command models. Each acoustic command model represents one or more :
-- -- ~
yog g3 030 4 2~1~21~
. .
series of acoustic fea-ture values representing an utterance of one or more words associated with the acoustic command model. The system further includes means for ldentifying a first active-state vocabulary of acoustic command models for the first active state.
The first active-state vocabulary comprises the acoustic command models from the system vocabulary representing the first active-state commands.

. .
The interactive computer system comprises a speech recognizer for measuriny the value of at least one feature of an utterance during each of a series of successive time intervals within the first time period to produce a series of feature signals. The speech recognizer compares the measured feature signals to each of the acoustic command models in the first active-state vocabulary to generate a match score for the utterance and each acoustic command model. The speech recognizer then outputs a command signal corresponding to the command model from the first active-state vocabulary having the best match score.

The first active-state vocabulary preferably comprises substantially less than all the acoustic command models from the system vocabulary. The speech recognizer does not compare the measured feature signals for the first time period to any acoustic command model which is not in the first active-state vocabulary.

In one embodiment of the interactive computer system according to the invention, the display means displays at least a second active-state image different from the first active-state image for a second active state occurring during a second time period different from the first time period. The object identifying means identifies at least one object displayed in the second active-state image, and generates a list of one or more second active-state commands identifying functions which can be performed in the second active state of the target computer program.

The active-state vocabulary identifying means identifies a second active-state vocabulary of acoustic command models for the second active state. The second active-state vocabulary comprises the acoustic command models from the system vocabulary representing the Y09-93-030 5 21~2~0 .
second active-state commands. The second active-state vocabulary is at least partly different from the first active-state vocabulary. ~
. .
The speech recognizer measures the value of at least one feature of an utterance during each of a series of successive time intervals within the second time period to produce a series of feature ~-signals. The speech recognizer compares the measured feature signals for the second time period to each of the acoustic command models in the second active-state vocabulary to generate a match score for the utterance and each acoustic command model. The speech recognizer then outputs a command signal corresponding to the command model from the second active-state vocabulary having the best match score.

The target computer program may, for example, have only one active state occurring during each time period. The target computer program may comprise an operating system program alone, an application program and an operating system program combined, or two or more application programs and an operating system program.

At least some of the commands for an active-state identify functions which can be performed on the identified objects in the active-state image for the state.

The identified object in an active-state image may, for example, comprise one or more of a character, a word, an icon, a button, a scroll bar, a slider, a list box, a menu, a check box, a container, or a notebook.

In an alternative embodiment of the invention, the speech recognizer may output two or more command signals corresponding to the command models from the active-state vocabulary having the best match scores for a given time period.
:
The vocabulary of acoustic command models for each active state may further comprise a set of global acoustic command models representing global commands identifying functions ~h.ich can be performed in each active state of the target computer program.

Y09-93-030 6 ~52 ~ ~

The display means may comprise, for e~ample, a cathode ray tube display, a liquid crystal display, or a printer.

The display means may display both an active-state image for an active state occurring during a tlme period, and at least a portion of one or more images for program states not occurring during the time period.

A method o~ computer lnteraction according to the invention comprises executing, on a processor, a target computer program having a series of active program states occurring over a series of time periods. The tar~et computer program generates active state image data signals representing an actlve state lmage for the active state of the target computer program occurrlng during each time period. Each active state image contalns one or more objects.
The method further comprlses dlsplaying at least a first active-state image for a first active state occurring during a first time period. At least one object displayed in the first active-state image is identified, and a list of one or more first active-state commands identifying ~unctions which can be performed in the first active state of the target computer program is generated from the identified object.

A system vocabulary of acoustic command models is stored. Each acoustic command model represents one or more series of acoustic feature values representing an utterance of one or more words associated with the acoustic command model. A first active-state vocabulary of acoustic command models for the first active state is identi~ied. The first active-state vocabulary comprises the acoustic command models from the system vocabulary representing the first active-state commands.

The value of at least one feature of an utterance is measured during each of a series of successive time intervals within the first time period to produce a series of feature signals. The measured feature signals are compared to each of the acoustic command models in the first active-state vocabulary to generate a match score for the utterance and each acoustic command model. A
command signal corresponding to the command model from the first Y09-93-030 7 ~ 2 1 0 active state vocabulary having the best match score is output.

By identifying at least one object displayed in the active-state image of the target computer program, and by generating from the identified object a list of one or more active-state commands identifying functions which can be performed in the active state of the target computer program, the active-state vocabulary of the speech recognizer can be limited to a small subset of the system vocabulary representing active-state commands, without having to predict in advance the states and transitions between states of the target computer program which will occur under all possible circumstances.

Brief Description of the Drawing Figure 1 is a block diagram of an example of an interactive computer system according to the invention.

Figure 2 shows an example of a first active-state image for a first active state of a target computer program.

Figure 3 is a block diagram of an example of a speech recognizer for an interactive computer system according to the invention.
, " .
Figure 4 shows an example of a second active-state image for a second active state of a target computer program.

Figure 5 is a block diagram of an example of an acoustic command model store for the system vocabulary of an interactive computer system according to the invention.
'~
; Figure 6 is a block diagram of an acous-tic processor for the speech ' recognizer of Figure 3.

Figure 7 schematically shows an example of an acoustic command ' model.

Figure 8 schematically shows an example of an acoustic model of a phoneme for constructing an acoustic command model.
, ~

., .

Y09-93-030 8 ~ 2 ~ 0 Figure 9 schematically shows an example of paths through the acoustic model of Figure 7.

Description of the Pre~erred Embodiments Figure 1 is a block diagram of an example of an interactive computer system according to the invention. The interactive computer system comprises a processor 10 executing a target computer program having a series of active program states occurring over a series of time periods. The target computer program generates active state image data signals representing an active state image ~or the active state of the target computer program occurring during each ti.me period. Each active state image contains one or more objects.

The processor may be, for example, a personal computer, a computer work station, or any other microcomputer, minicomputer, or main frame computer.

The target computer program may be an operating system program such as DOS, Microsoft Windows~M, OS/2~, AIX~, UNIXTM, X-WindowslM or any other operating system. The target computer program may comprise one or more application programs executing with an operating system program. Application programs include spreadsheet programs, word processing programs, database programs, educational programs, recreational programs, communication programs, and many more.

Ob;ects in an active-state image may comprise one or more of a character, a word, an icon, a button, a scroll bar, a slider, a list box, a menu, a check box, a container, a notebook, or some other items.

The interactive computer system further comprises display means 12 for displaying at least a first active-state image for a first active state occurring during a first time period. The display means may be, for example, a cathode ray tube display, a li~uid crystal display, or a printer.
:: :
Figure 2 shows an example of a hypothetical first active-state ~

image for a first active state occurring during a first time period. In this example, the active-state image includes a frame object 14 containing a title bar object 16, a menu bar object 18, a list box object 20, and a push button object 22. The menu bar object 18 includes an "items" object, an "options" object, and an "exit" object. The list box object 20 includes a vertical scroll bar object 24, and "blue", "green", "red", "orange", "black", "white", and "purple" objects. In the list box 20, only the "blue", "green", "red", "orange", and "blac~" objects are shown in Figure 2. The "white" and "purple" objects are contained in the list box and could be made visible by scrolling with the vertical scroll bar 24.

The active state image data signals may be generated by the target computer program, for example, by using operating system interrupts, function calls~ or application program interface calls.

Example I, below, illustrates C programming language source code for creating active state image data signals.

Returning to Figure 1, the interactive computer system further comprises an image object identifier 26 for identifying at least one object displayed in the first active-state image, and for generating from the identified object a list of one or more first active-state cc ~nds identifying functions which can be performed in the first active-state of the target computer program.

The image object identi~ier 26 may comprise computer program subroutines designed to intercept (hook) operating system function calls, and application program interface calls provided by one or more target computer programs, and/or may comprise computer program subroutines for using operating system interrupts, function calls, or application program interfaca calls for identifying objects displayed in the first active-state image of the target computer program. Example II, below, illustrates C programming language source code for identifying at least one object displayed in an active state image.

Table 1 shows a hypothetical example of a list of first --- 211~210 : active-state commands identifying functions which can be performed in the first active-state of the target computer program for the objects displayed in the first active-state image of Figure 2.

,, OBJECT SPOREN COMMAND FUNCTION
Frame FRAME changes the focus to the entire frame TOP BORDER identifies ~rame element to be moved BOTTOM BORDER identifies frame element to be moved LEFT BORDER identifies frame element to be moved RIGHT BORDER identifies frame element to be moved :
LEFT moves frame or frame element left RIGHT moves frame or frame element right ~ :
UP moves frame or frame element up . ~:~
DOWN moves frame or frame element down Title Bar none none . :
Menu Bar CLOSE MENU hides the menu :~:
MENU changes the focus to the menu bar SELECT selects the item at the cursor :
"ITEMS" ITEMS activates the "ITEMS" menu ~.
' "COLORS" COLORS activates the "COLORS" menu ; "NAMES" NAMES activates the "NAMES" menu "ADDRESSES" ADDRESSES activates the "ADDRESSES" menu "OPTIONS" OPTIONS activates a dialog to select options "EXIT" EXIT exits the current program state ::~
CANCEL dismisses the pop-up menu System Menu CLOSE MENU hides the menu MENU changes the focus to another menu, if any SELECT selects the item at the cursor RESTORE restores window to previous size and position MINIMIZE reduces window to smallest size ~A~TMT7.F. increases window to largest size CLOSE exits the current program state WINDOW LIST displays a list of running programs : OBJECT SPOKEN COMMAND FUNCTION
Vertical SCROLL BAR sets focus on the scroll bar Scroll Bar UP moves the list bo~ up through the subject being displayed DOWN moves the list box down through the subject being displayed TOP moves the list box to the top of the subject being displayed BOT~OM moves the lis~ box to the bottom of the subject being displayed PAGE UP moves the list box up one page through the subject being displayed PAGE DOWN moves the list bo~ down one page through the subject being displayed Push Button PRESS executes the push button .
PUSH BUTTON executes the push button "HELP" HELP executes the help facility ~:
List Box LIST BOX changes the focus to list box "BLUE" BLUE selects the named color :
"GREEN" GREEN selects the named color "RED" RED selects the named ~olor "ORANGE" O~ANGE selects the named color "BLACK" BLACK selects the named color "WHITE" WHITE selects the named color "PURPLE" PVRPLE selects the named color As shown in the example of Table 1, each object may have zero or more commands identifying functions which can be performed in the first active state of the target computer program. At least some commands identify functions which can be performed on the identified object in the active-state image for the state. For example, the command "FRAME" changes the focus to the entire frame object 14 of Figure 2. With the focus on the entire frame objPct 14, the spoken command "LEFT" operates on the frame object by moving it to the left on the display screen.

Returning again to Figure 1, the interactive computer system comprises a system acoustic command model vocabulary store 28 for - 2~1521 0 ':
storing a system vocabulary of acoustic command models. Each acoustic command model represents one or more series of acoustic feature values representing an utterance of one or more words associated with the acoustic command model.

The stored acoustic ccmmand models may be, for example, Markov models or other dynamic programming models. The parameters of the acoustic command models may be estimated from a known uttered training text (for example, 257 sentences) by, for example, smoothing parameters obtained by the forward-backward algorithm.
(See, for example, Jelinek. "Continuous Speech Recognition By Statistical Methods", Proceedings of the IEEE, Volume 64, No. 4, April 1976 pages 532-536.) ';"':
Preferably, each acoustic command model represents a command spoken in isolation (that is, .independent of the context of prior and subse~uent utterances). Context-independent acoustic command models can be produced, for example, either manually from models of phonemes or automatically, for example, by the method described by Lalit R. Bahl et al in U.S. Patent 4,759,068 entitled "Constructing Markov Models o~ Words from Multiple Utterances", or by any other known method of generating context-independent models.

Alternatively, context-dependent models may be produced from context-independent models by grouping utterances of a command into context-dependent categories. A context can be, for example, manually selected, or a-ttomatically selected by tagging each feature signal corresponding to a command with its context, and by grouping the feature signals according to their context to optimize a selected evaluation function. (See, for example, Lalit R. Bahl et al, "Apparatus And Method Of Groupi.ng Utterances Of A Phoneme Into Context-Dependent Catégories Based On Sound-Similarity For Automatic Speech Recognition." U.S. Patent 5,195,167.) As shown in the block diagram of Figure 1, the interactive computer system comprises an active-state command model vocabulary identifier 30 for identifying a ~irst active-state vocabulary of acoustic command models for the first active state. The first Y09-93-030 13 2 1 ~

active-state vocabulary comprises the acoustic command models from the system vocabulary 28 representing the first active-state commands from the image object identifier 26. Example III, below, illustrates C programming language source code for identifying an active-state vocabulary. Example IV, below, illustrates C
programming language source code for defining the ac-tive-state vocabulary to the speech recognizer. -Preferably, the active-state vocabulary comprises substantially less than all of the acoustic command models in the system vocabulary. For example, each active-state vocabulary may comprise 50 to 200 commands. The entire system command vocabulary may comprise 500 to 700 or more commands. The speech recognizer does ;
not compare the measured feature signals for a time period to any acoustic command model which is not in the active-state vocabulary for that time period. ~' A speech recognizer 32 measures the value of at least one feature of an utterance during each of a series of successive time intervals within the first time period to produce a series of feature signals. The speech recognizer 32 compares the measured feature signal to each of the acoustic command models in the first active-state vocabulary to generate a match score for the utterance and each acoustic command model. The speech recognizer 32 outputs a command signal corresponding to the command model from the first active-state vocabulary having the best match score.

Example V, below, illustrates C programming language source code for outputting a command signal corresponding to the command model from an active-state vocabulary having the best match score.

Figure 3 is a block diagram of an example of a speech recognizer for an interactive computer system according to the invention. In this example, the speech recognizer 32 comprises an active-state acoustic command model store 34 for storing the active-state vocabulary comprising the acoustic command models from the system vocabulary store 28 representing the active-state commands identified in active state command model vocabulary identifier 30.

Y09-93-030 14 2 ~ ~ ~ 2 ~ ~

, .
The speech recognizer 32 ~urther comprises an acoustic processor 36 for measuring the value of at least one feature of an utterance during each of a series of successive time intervals within each active-state time period to produce a series of feature signals. An acoustic match score processor 38 compares the measured feature signals from acoustic processor 36 -to each of the acoustic command models in the active-state acoustic command models store 34 to generate a match score for the utterance and each acoustic command model. An output 40 outputs one or more com~and signals corresponding to the command models from the active state vocabulary having the best match scores for a given time period.

Preferably, only one command signal corresponding to the command model from the first active-state vocabulary having the best match score is output. In this case, the one output command may be immediately executed. If two or more command signals corresponding to the command models from the active-state vocabulary having the best match scores for a given time period are output, then the recognized cc ~n~.~ may be displayed for the user to select one for execution.

The speech recogni~er may be a publicly available product such as the IBM~ Voice Type II or the IBM Speech Server Series. In products containing a fast acoustic match and a detailed acoustic match, both acoustic matches may be used in the invention.
~lternatively, since the image objec-t identifier 26 and the active state command model vocabulary identifier 30 select only a small subset of the system vocabulary in store 28 for the acoustic match, the fast acoustic match can be omitted.

In speech recognition products containing a language model, the language model can be omitted. Alternatively, all of the words in the active state vocabulary can be assigned equal language model probabilities.

In speech recognizer products having hypothesis search algorithms for generating multiple-word hypotheses, the recognition of a word is dependent in part on the recognition of successive words. Such : 2~ 1~210 Yoq-93-030 15 ' a hypothesis search algorithm need not be used with the present invention in which, preferably, each command is independent of successive commands.

Preferably, both the target computer program and the speech recognizer are executed on the same central processing unit in a time sharing manner. Alternatively, the target computer program and the speech recognizer can be executed on different central processing units, for example using a client-server architecture.

In the interactive computer system according to the invention, the display means may further display at least a second active-state image different from the first active-state image for a second active state occurring during a second time period different from the first time period.

Figure 4 shows an example of a second active-state image for a second active state of the target computer program. The second active-state image shown in Figure 4 contains a frame object 42, a title bar object 44, a system menu object 46, a vertical scroll bar object 48, a horizontal scroll bar object 50, and a container object 52~ The container object 52 contains an "editor" object, a "phone book" object, a "spreadsheet" object, a "mail" object, and a "solitaire" ob~ect.
'i The object identifying means identifies at least one object displayed in the second active-state image, and generates from the identified object a list of one or more second active-state commands identifying functions which can be performed in the second active-state of the target computer program.

Table 2 is an example of a hypothetical list of commands for each object shown in Figure ~ identifying functions which can be performed in the second active-state of the target computer program.

. YO9-93-030 16 ~ 2 1 ~
~' ' TABLE 2 .
OBJECT SPOKEN COMMAND FUNCTION
Frame FRAME changes the focus to the entire frame -~
TOP BORDER identifies frame element to be moved BOTTOM BORDER identifies frame element to be moved : LEFT BORDER identifies frame element to be moved RIGHT BORDER identifies frame element to be moved LEFT moves frame or frame element left RIGHT moves frame or frame element right UP moves frame or frame element up DOWN moves frame or frame element down Title Bar none none System Menu CLOSE MENU hides the menu MENU changes the focus to another menu, if any SELECT selects the item at the cursor RESTORE restores window to previous size and position MT~TMT7.F. reduces window to smallest si2e ~-MARTMT7.~ increases window to largest size CLOSE exits the current program state WINDOW LIST displays a list of running programs : :
Vertical SCROLL BAR sets focus on the scroll bar Scroll Bar UP moves the container up through the subject being displayed DOWN moves the container down throuyh the subject ~
, i being displayed . :
TOR moves the container to the top of the subject being displayed BOTTOM moves the container to the bottom of the ~ :
subject being displayed PAGE UP moves the container up one page through the -:
subject being displayed ~-Horizontal SCROLL BAR sets focus on the next scroll bar : scroll Bar LEFT moves the container left through the subject being displayed Y09-93-030 17 2~2~0 OBJECT SPOKEN COMMAND FUNCTION
RIGHT moves the container right through th~ subject : bein~ displayed EXTREME LEFT moves the container to the extreme left of '. the subject being displ~yed EXTREME RIGHT moves the container to the extreme right of the subject being displayed PAGE LEFT moves the container left one page through the subject being displayed PAGE RIGHT moves the container right one page through the subject being displayed Container CON~AINER changes the focus ~o the container SELECT ALL executes all programs in the container Editor EDITOR executes the Editor program Phone Book PHONE BOOK executes the Phone Book progra~
Spreadsheet SPREADSHEET executes the Spreadsheet program Mail MAIL executes the Mail program Solitaire SOLITAIRE executes the Solitaire program Comparing Figures 2 and 4, the first active-state image diffexs from the second active-state image by providing menu bar object 18, list box object 20, and push button object 22 in the first active-state image but not in the second active-state image. The horizontal scroll bar 50, and the editor, phone book, mail, spreadsheet, and solitaire objects are provided in the second active-state image, but not in the first active-state image.

The active-state vocabulary identifying means further identifies a second active-,state vocabulary of acoustic command models for the second active state. The second active state vocabulary comprises the acoustic command models from the system vocabulary representing the second active-state commands. The second active-state vocabulary is at least partly different from the first active-state vocabulary.

Comparing Tables 1 and 2, the first active-state vocabulary comprises the spoken commands listed in Table 1. The second active-state vocabulary comprises the spoken co ~n~s listed in Table 2. In this example, the first active-state vocabulary is at least partly different from the second active-state vocabulary as Y09-93-030 18 211~ 21 D

shown therein.
~; .
The speech recogni~er measures the value of at least one feature of an utterance during each of a series of successive time intervals within the second time period to produce a series of feature signals. The speech recognizer compares the measured feature signals for the second time period to each of the acoustic command models in the second active-state vocabulary to generate a match score for the utterance and each acoustic command model. The speech recognizer OtltputS a command signal corresponding to the command model from the second active-state vocabulary having the best match score.

Preferably, the target computer program has only one active state occurring during each time period.

Figure 5 is a block diagram of an example of the acoustic command model vocabulary store 28 of Figure 1. The system vocabulary may comprise, for example, a set of global acoustic command models representing global co An~ identifying functions which can be performed in every active state of the target computer program.

Table 3 lists some examples of global cc And~ represented by globsl acousti~ command models.

Y09-93-030 19 21~21~

GLOBAL COMMANDS
SPOKEN COMMAND FUNCTION
MICROPHONE OFF turns the microphone off ENTER sends "ENT~R" keystroke to l~eyboard input buffer LEFT sends "LEFT ARROW" keystroke to keyboard input buffer RIGHT sends "RIGHT ARROW" keystroke to keyboard input buffer PASTE inserts contents of clipboard into application with the focus WINDOW LIST displays a list of running programs EDITOR executes the editor program DESK top makes the des~ top window active ;:~
The system vocabulary may further comprise object type acoustic command models associated with different types of objects. For example, as shown in Tables 1 and 2, frame object type acoustic commands include "frame", "top border", "bottom border", "left border", "right border", "left", "right", "up", and "down".
Vertical scroll bar object type acoustic co ~n~ include "scroll bar", "up", "down", "top", "bottom", "page up", "page down". Push button object type acoustic command models include "press" and "push button".

Finally, the system vocabulary includes application-specific acoustic command models representing application-specific objects.
In the examples of Tables 1 and 2, application-specific objects include the words "items", "colors", "names", "addresses", "phone book", "spreadsheet", "mail" and "solitaire".

The display means 12 of Figure 1 may display both an active-state image for an active state occurring during a time period, and at least a portion of one or more images for program states not occurring during the time period.

One example of the acoustic processor 36 of Figure 3 is shown in Figure 6. The acoustic processor comprises a microphone 54 for -' 2~1 ~21 0 generating an analog electrical signal corresponding to the utterance. The analog electrical signal from microphone 54 is converted to a digital electrical signal by analog to digital converter 56. For this purpose, the analog signal may be sampled, for example, at a rate of twenty kilohertz by the analog to digital converter 56.

A window generator 58 obtains, for example, a twenty millisecond duration sample of the digi-tal signal from analog to digital converter 56 every ten milliseconds (one centisecond). Each twenty millisecond sample of the digital signal is analyzed by spectrum analyzer 60 in order to obtain the amplitude of the digital signal sample in each of, for example, twenty frequency bands.
Preferably, spectrum analy~er 60 also generates a twenty-first dimension signal representing the total amplitude or total power of the twenty millisecond digital signal sample. The spectrum ~:
analyzer 60 may be, for example, a fast Fourier transform ;~
processor. Alternatively, it may be a bank of twenty band pass filters. ;~

The twenty one dimension vector signals produced by spectrum analyzer 60 may be adapted to remove background noise by an adaptive noise cancellation processor 62. Noise cancellation processor 62 subtracts a noise vector N(t) from the feature vector F(t) input into the noise cancellation processor to produce an output feature vector F'(t). The noise cancellation processor 62 adapts to changing noise levels by periodically updating the noise vector N(t) whenever the prior feature vector F(t-1) is identified ;
as noise or silence. The noise vector N(t) is updated according to the formula N( t) = N( t~ k[F( t-l ) -Fp( t-l ) ] (1) '"
where N(t) is the noise vector at time t, N(t-l) is the noise vector at time (t-1), k i~ a fixed parameter of the adaptive noise cancellation model, F(t 1) is the feature vector input into the noise cancellation processor 62 at time (t-1) and which represents -noise or silence, and Fp(t-1) is one silence or noise prototype vector, ~rom store 64, closest to feature vector F(t-l).

The prior feature vector F(t-1) is recognized as noise or silence if either (a) the total energy of the vector is below a threshold, or (b) the closest prototype vector in adaptation prototype vector store 66 to the feature vector is a prototype representing noise or silence. For the purpose of the analysis of the total energy of the feature vector, the threshold may be, for example, the fifth percentile of all feature vectors (corresponding to both speech and silence) produced in the two seconds prior to the feature vector being evaluated.

After noise cancellation, the feature vector F'(t) is normalized to adjust for variations in the loudness of the input speech by short term mean normalization processor 68. Normalization processor 68 normalizes the twenty-one dimension feature vector F'(t) to produce a twenty dimension normalized feature vector X(t). The twenty-first dimension of the feature vector F'(t), representing the total amplitude or total power, is discarded. Each component i of the normalized feature vector X(t) at time t may, for example, be given by the equation ..
Xi(t)=F'/l(t)-Z(t) (2) in the logarithmic domain, where F'i(t) is the i-th component of the unnormalized vector at time t, and where Z(t) is a weighted mean of the components of F'(t) and Z(t - 1) according to Equations 3 and 4:
: ,~
Z(t)=0.9Z(t~ O.lM(t) ~3) and where ':

---'' 2~21~

M(t)= 1 ~ F/ (t) (4) :, The normalized twenty dimension feature vector X(t) may be further processed by an adaptive labeler 70 to adapt to variations in pronunciation of speech sounds. An adapted twenty dimension feature vector X'(t) is generated by subtracting a twenty dimension adaptation vector A(t) from the twenty dimension feature vector .
X(t) provided to the input of the adaptive labeler 70. The ~' adaptation vector A(t) at time t may, for example, be given by the formula ,', ~ .

A ( t) = A ( t-l ) +k [X( t-l ) -Xp ( t-l ) ] : ' where k is a fixed parameter of the adaptive labeling model, X(t~
is the normalized twenty dimension vector input to the adaptive labeler 70 at time (t-l), Xp(t-1) is the adaptation prototype vector (from adaptation prototype store 66) closest to the twenty dimension feature vector X(t-l) at time (t-l), and A(t-1) is the adaptation vector at time (t-1).

The twenty dimension adapted feature vector signal X'(t) from the adaptive labeler 70 is preferably provided to an auditory model 72.
Auditory model 72 may, for example, provide a model of how the human auditory system perceives sound signals. An example of an auditory model is described in U.S. Patent 4,980,918 to Bahl et al entitled "Speech Recognition System with Efficient Storage and Rapid Assembly of Phonological Graphs".

Preferably, according to the present invention, for each frequency band i of the adapted feature vector signal X'(t) at time t, the auditory model 72 calculates a new parameter Ei(t) according to Equations 6 and 7:

E~(t)=~1+K2(X/~(t))(NI ( t-1)) (6) where 2~2~0 . ..
Ni(t)=K3xNi(t-1)-Ei(t-1) (7) and where K1, K2 and K3 are fixed parameters of the auditory model.

For each centisecond time interval, the output of the auditory model 72 is a modified twenty dimension feature vector signal. This feature vector is augmented by a twenty-first dimension having a value equal to the square root of the sum of the s~uares of the values of the other twenty dimensions. ' For each centisecond time interval, a concatenator 74 preferably concatenates nine twenty-one dimension feature vectors representing the one current centisecond time interval, the four preceding centisecond time intervals, and the four following centisecond time intervals to form a single spliced vector of 189 dimensions. Each 189 dimension spliced vector is preferably multiplied in a rotator 76 by a rotation matrix to rotate the spliced vector and to reduce the spliced vector to fifty dimensions.
~ :~
The rotation matrix used in rotator 76 may be obtained, for ~;
example, by classifying into M classes a set of 189 dimension spliced vectors obtained during a training session. The covariance matrix for all of the spliced vectors in the training set is multiplied by the inverse of the within-class covariance matrix for all of the spliced vectors in all M classes. The first fifty eigenvectors of the resulting matrix form the rotation matrix.
(See, for example, "Vector Quantization Procedure For Speech Recognition Systems Using Discrete Parameter Phoneme-Based Markov Word Models" by L. R. Bahl, et al, IBM Technical Disclosure Bulletin, Volume 32, No. 7, December 1989, pages 320 and 321.) "~

Window generator 58, spectrum analyzer 60, adaptive noise cancellation processor 62, short term mean normali~ation processor 68, adaptive labeler 70, auditory model 72, concatenator 74, and ~;
rotator 76, may be suitably programmed special purpose or general ;~
purpose digital signal processors. Prototype stores 64 and 66 may be electronic computer memory of the types discus~ed above.
..~
:,"-".':' ':' .~.,'' : . .

:
Y09-93-030 24 211~21~

The proto-type vectors in prototype store 54 may be obtained, for example, by clustering feature vector signals from a training set ~i into a plurality of clusters, and then calculating the mean and standard deviation for each cluster to form the parameter values of the prototype vector. When the training script comprises a series of word-segment models (forming a model of a series of words), and each word-segment model comprises a series of elementary models having specified locations in the word-segment models, the feature vector signals may be clustered by specifying that each cluster corresponds to a single elementary model in a single location in a single word-segment model. Such a method is described in more detail in U.S. Patent No. 5,276,766, issued January 4, 1994, ' entitled "Fast Algorithm for Deriving Acoustic Prototypes for Automatic Speech Recognition."
: ' Alternatively, all acoustic feature vectors generated by the utterance of a training text and which correspond to a given elementary model may be clustered by K-means Euclidean clustering or K-means Gaussian clustering, or both. Such a method is described, for example, by Bahl et al in U.S. Patent 5,182,773 entitled "Speaker-Independent Label Coding Apparatus".

Figure 7 schematically shows a hypothetical example of an acoustic command model. The hypothetical model shown in Figure 7 has a starting state Sl, an ending state S4, and a plurality of paths from the starting state Sl to the ending state S4.
.
Figure 8 schematically shows a hypothetical example of an acoustic Markov model of a phoneme. In this example, the acoustic phoneme model comprises three occurrences of transition Tl, four -~
' occurrences of transition T2, and three occurrences of transition T3. The transitions shown in dotted lines are null transitions.
.
~ Each solid-line transition in the acoustic models of Figures 7 and -~
8 has at least one model output comprising an acoustic feature value. Each model output has an output probability. Each null -transition has no output. Each solid line transition and each dotted line transition from a state has a probability of occurrence when the model is in that state.

211~21~

Figure 9 shows a hypothetical example of paths through the acoustic model o~ Figure 7. The match score for an utterance and an acoustic command model is the sutn of the probabilities of the measured features of the utterance for all paths through the acoustic command model. For each path, the probability of the measured ~eatures of the utterance is e~ual to the product of the probabilities of the transitions along the path times the probabilities of the measured features at each transition along the path.

Preferably, the interactive computer system according to the invention may be made by suitably programming a general purpose digital computer system. More specifically, the processor 10, the image object identifier 26, and the active-state command model vocabulary identifier 30 may be made by suitably programming a general purpose digital processor. The system acoustic command model vocabulary store 28 and the active-state acoustic command models store 34 may be electronic computer memory. The display means 12 may comprise a video display such as a cathode ray tube, a liquid crystal display, or a printer.

As mentioned above, the target computer program may be one or more application programs and an operating system program. For example, the target computer program may be IBM OS/2 version 2.0, and Presentation Manager~.

IBM OS/2 version 2.0 operating system and Presenta-tion Manager have application program interface calls in various languages, including the C programming language, the assembly programming language, and the REXX programming language. The complete collection of application program interface calls is part of the OS/2 2.0 Technical Library. The syntax for the application program interface calls in a language is compatible with how standard calls operate in the language. The name of a particular application program interface call may be different for different languages.
Also, some aspects of the application program inter~ace in one language may not be supported from another language.
. , .
For the C programming language, the application program interface ~-!

: ~11521~

:
consists of many library calls. C programming language source code be compiled with the IBM C Set/2~M compiler.

Examples I to V illustrate C programming language source code for OS/2 and Presentation Manager for (a) creating and displaying an image, (b) reading the active ~tate image to identify at least one object displayed in the active state image, (c) creating the vocabulary from the active state image, (d) defining the vocabulary to the speech recognizer, and (e) outputting a command signal corresponding to the command model from an active-state vocabulary having the best match score.

Exampl~ I

Example I illustrates C programming language source code for creating the hypothetical first active-state image shown in Figure .. ~
There is a concept of a "standard window" in OS/2 and Presentation Manager. A standard wi.ndow is a combination of several commonly-used windows. In Figure 2, the frame window, title bar, system menu and mellu bar can be considered to be part of a standard window. The standard window is created with the following C ' programming language source code using the OS/2 application program inter~ace call Wi~CreateStdWi~dow(). The comments ~ollowing the double slashes (//) describe the operation of the source code.

#define INCL WIN // Required to get Presentation // Manager definitions. '~
#include <os2~11> // Requlred to get Presentation // Manager definitions.
// Prototype definition for window // procedu~e.
MRESVLT ~hnL~Y SampleProc( HWND lwnd, ULONG ulMsg, MPARAM mpl, MPARAM mp2 );
HWND hwndFrame- // This is a variable to hold a "handle"
// to a frame window. A window handle is // unique for ea~h window.
HWND hwndClient- // This is a variable to hold a "handle"
// to a client window.
ULONG ulFlags; // This is a variable for the frame data // to be used at creation.
HAB hAB- // A ~resentation Manager anchor block // handle...not important for this // example. It's a handle which is // received during initialization and // used when termlnating.
HMQ hMQ; // A message queue. Presentation Manager -' 211~21~

;:
., .
// uses this to send messages to the // application windows.
// All applications must make this call // to initialize Presentation Manager.
hA8 = WinInitialize(0);
// Create a message queue for ;~ // Presentation Manager to use. The ; // seconc parameter means to take the // defau_t size of message queue.
; hMQ = WinCreateMsgQueue( hAB, 0 j;
// Reyister the class of our client // window. This s~ecifies a function : // which Presentatlon Manager will use // to send messages of events that the // window would like to know about. Some // messages are WM_SIZE which tells the // window that its size is ~hi~ng;ng, // WM CREATE which tells a window that it // is being created, and WM_BUTTONlDOWN
i // which tells when a mouse button has ?- // been clicked in the window. -~
// The arguments for WinRegisterClass():
// hAB - the handle received from // WinInitialize(). ~
' // "Generic" - the name of our window class. This ~-// string will be used to create a window // of our type.
// SampleProc - the name of our window procedure as // defined with the above prototype.
// OL - class style.. none ~' ~' // OL - Amount of special storage reserved // for application's use.. none.
WinRegisterClass( hAB, "Generic", SampleProc, OL, OL 3: :~
// Set up the frame creation data to ~; // specify some of the specific windows i~
// desired. ,~
ulFlags = FCF_TITLEBAR ¦ FCF_SYSMENU I FCF_BORDER;
// The arguments for WinCreateStdWindow(): ;;
! / j HWND_DESXTOP - the parent window. Make the frame to // be the child of the Presentation // Manager desk top.
// OL - frame style.. none // ulFlags - frame creation fla~s // "Generic" - our previously reglstered window // procedure.
// "Title" - title to be in title bar.
// OL - client window style.. none // NULLHANDLE - implies that frame resources, such as // the menu bar description are ~
// compiled into the resultant EXE using -'i // the resource compiler that is part of // the OS/Z Toolkit for its application ' // program inter~ace.
; // 10 - ID of the resources in the EXE.
// &hwndClient - pass the address of the client window // handle so that the application program // interface can copy back the newly // created client handle.

.

21~l~21 ~

// hwndFrame = WinCreateStdWindow( HWND_DESKTOP, OL, &ulFlags, "Generic' "Title", OL, NULLHANDLE, 10, &hwndClient );
// Size and position the frame on the // screen, and make it visible with // WinSetWindowPos().
// The arguments for WinSetWindowPos():
// hwndFrame - handle to our frame for which we want // to set the size and position // HWND TOP - set the frame above all other frames .
// so that it can be seen and used.
// 10, 20 - the desired position (x, ~
// 300, 500 - the desired size (width, height).
// SWP_..... - flags telling Presentation Manager to // process the size, move the window, ;;
// and show it.
// '.WinSetWindowPos( hwndFrame, HWND TOP, 30d, 500' SWP_SIZE I SWP_MOVE i SWP_SHOW );
// Presentation Manager is a message based system and // during the create call, a WM CREATE message is sent to // the above-registered window procedure. The other child "~
// windows are created while processing this message. This // is depicted below:
MRESULT ~x~ Y SampleProc( HWND hwndClient, ULONG ulMsg, MPARAM mpl, MPARAM mp2 ); ~'~
HWND hwndList; ;~
~WND hwndButton-switch( ulMsg ) {
~, ! case WM_CREATE:
// We are processing the WM_CREATE
// message for our client window which is // just being created. The passed window // handle, hwndClient, wil be returned // via the last parameter -n the // WinCreateStdWindow() ca l.
// Now create the child list box.
// The arguments for WinCreateWindow():
// hwndClient - set the parent to be // the client window.
// WC_LISTBOX - window class. This is // a list box.
// "" - no titls text // associated with the // list box.
// WS_... - window styles...make a :

'~ 21~21~

., // visible pushbutton.
// O, 0 - in_tial coordinates at // wh_ch to place window.
// 50, 30 - in-tial si~e of // window.
// hwndClient - set the owner to be // the client window.
// HWND_TOP - place this window // above all others.
// ID_BUTTON - window id.
// NULT - no control data // NULL - no presentation // parameters hwndList = WinCreateWindow( hwndClient, WC_LISTBOX, ~
WS_VISIBLE I LS_MULTIPLESEL, ;
; 50, 30 hwndClient, HWND_TOP
ID_LISTBOX, NULL
' NULL );
// The arguments for WinCreateWindow() // are the same as above, with the // exceptions that there are different // window styles for the button class, ;~
// there is a different class name, the // ID is different, and the button ' ~-// has meaningful text.
hwndButton = WinCreateWindow( hwndClient, ~-WC_BUTTON, "Help", WS_VISIBLE I BS_P~H~UllON, '~
O, 70, 100, 250 hwndClient, HWND_TOP, ID_BUTTON, NULL, NULL );
// Finished processing the message. -~
// Return control to Presentation // Manager.
break;

return( FALSE ); ;

Example II

~: Example II illustrates C programming language source code for ;; : reading an active-state image.
'~:
Presentation Manager provides an application program interface call - 2~ 1~2~
.~

., for any application to put a "hook" into the queues of messages which are passed back and forth between windows. A hook is installed with a call back function which gets called with every message which is sent. Call back functions for hooks must reside in a presentation manager dynamic link library. The required procedure is to load the dynamic link library which contains the call back function and then load the hook.
HMODVLE hm; // A handle for a loaded dynamic link library // a frame window. A window handle is unique // This is the function prototype for the // callback. It follows the syntax for a // SendMsgHook as described in the IBM
// Presentation Manager Programming Reference, // Volume III.
VOID ~X~EI~l~Y CallbackProc( HAB hAB, PSMHSTRUCT pSmh, BOOL bTask );
// To load the dynamic link library with the callback // function use DosLoadModule().
// The arguments for DosLoadModule() are the following:
// NULL - no buffer to return error information // O - length of buffer :
// "MYDLL" - name of DLL to load // &hm - address where to return the module ' // handle DosLoadModule( NULL, O, "MYDLL", &hm );
// Now set the hook. The arguments for WinSetHook() are as // follows:
// hAB - anchor block handle received from // Presentation Manager initialization.
// NULLHANDLE - hook the Presentation Manager system // ~ueue // HK SEND - install a hook for sent messages ! // CallbackProc - callback procedure from the loaded // dynamic link library // hm - handle to the loaded module WinSetHook( hAB, hMQ, HR SENDMSG, (PFN)CallbackProc, hm );
// With the hook installed the call back routine will get // called every time a message is sent n Presentation // Manager. One message that contains -nformation that a // new lmage (window) lS active is WM S TFOCUS. It can // be processed a follows to get the frame window which is // active.
VOID EXPENTRY CallbackProc( HA3 hAB, PSMHSTRUCT pSmh, BOOL bTask ) // Declaring some variables.
HWND hwndWithFocus;

' ~ 21~2~

HWND hwndFrame-HWND hwndParent~
HWND hwndDesktop;
{if (pSmh->msg == WM_SETFOCUS) // The call back has been called // with a WM_SETFOCUS Message.
// Unpack the message's second // parameter. This tells if the // message is for a window // receiving or losing focus.
if (SHORTlFROMMP(pSmh->mp2)j // This window is receiving the // focus.
hwndWithFocus = pSmh->hwnd; , // This may be a child window of // an actual image becoming // active. Get the absolute // parent which is a frame. Look // until we've reached the // Presentation Manager desk top // which is the root of all // visible windows. '~
// Get the desk top handle as a // comparison for the limit.
hwndDesktop = WinQueryDesktopWindow( hAB, NULLHANDLE );
hwndParent = hwndWithFocus;
// Loop to find the last parent // in the window chain.
while( hwndParent l= hwndDesktop ) hwndFrame = hwndParent;
// Query for the next parent.
hwndParent = WinQueryWindow( hwndFrame, QW_PARENT );

//
// At this point hwndFrame is the frame for the active // imagel } }

Exa~ple III ~

Example III illustrates C programming language source code for -:.
identifying the list of active-state commands from the active-state image.

The procedure for creating the list of active-state cc ~n~s from the image is as follows. (1) Create a list of all the windows which are children (direct or indirect) of the active frame found :~
:' :
21152:L~
Y09~93-030 32 : rj above. (2) Identify all windows i~ the list by their window class.
(3) For windows from window classes which display text to the user, query all the window text (hidden and visible). (4) Combine a global list of words with a standard list of words for each window type and with the words which were queried from the application in step (3).

Step (4) mer~ly involves combinin~ multiple arrays of words into one array of words. Therefore, source code for Step (4) is not illustrated.
, ~
~! // Step (1) Create a list of all the windows which are // children (direct or indirect) of the // active frame found above.
// Assume that we won't have more than 100 child ~! // windows, .~
. HWND AllWindows[100]; // Declare an array to hold the // window handles.
',~ int index = O; // Index at which to put windows ~ // into the AllWindows [] array.
? HWND hwndFrame; // Assume to be initialized to ; // the active window in the // CallbackProc() as outlined // above.
// Use a recursive function to get all children.
// Call it initially with the frame:
// FindChildren( hwndFrame );
~' VOID FindChildren~ HWND hwndParent ) ; HENUM hwndList;
HWND hwndChild;
// Put this window on the list. Increment the index // to point to the next available slot in the array.
AllWindows [ index ] = hwndChild;
index = index + 1; ' // Initiate an enumeration of the immediate child // windows. An enumeration handle, hwndList is // returned. It is used to sequentially accéss all // the child windows.
hwndList = Win~eginEnumWindows( hwndParent );
// Loop through all the children until the enumeration // returns a O window handle which means that there // are no more windows.
while( hwndChild = WinGetNextWindow( hwndList ) ) // For each window call this function again to get all // the children of THIS window.
FindChildren( hwndChild ); ;, . .

2 1 ~

// End the enumeration.
WinEndEnumWindows( hwndList );
~/ Step (2) Identify all windows in the list by their // window class.
// For each window in the list, get its type.
int i; // counting index -CHAR szBuffer[200]; // buffer to get class name int BufSize = sizeof(szBuffer)-HWND hwnd;
{for (i = 0; i < index; i~ ) hwnd = AllWindows [ i ~;
// This next function returns the class name as a // string in the buffer which is passed as an // argument.
WinQueryClassName( hwnd, BufSize, s~Buffer );
// Here are some class names defined in Presentation // Manager as generic windows. The actual strings are // enclosed in quotes, following C programming // language strlng conventions.
~ " a frame window //"-2" a button //"=4" a menu //"=~" a list box //"=P" a scroll bar // Step (3) For windows from window classes which // d_splay text to the user, query all the // w:ndow text (hidden and visible).
// In this coce sample it is shown how to read text // displayed by an application.
// - Assume that no text is longer than 200 bytes // for this example.
// - Assume that pBuffer is pointinq to a buffer of // shared memorY wlich has been given to the // process in which the window resides.
// - Assume that classname has been filled with the // class name of the object as described in (2) // above. ;~
CHAR classname [100];
CHAR *pBuffer;
int BufSize = 201;
int ListboxCount;
int i;
// Get application text for list boxes and buttons.
if (strcmp( classname, "#3" ) == 0) // This is a button. Get its text.
WinQueryWindowText( hwndButton, BufSi~e, pBuffer );
if (strcmp( classname, "#7" ) == 0) // This is a list box. Loop through all of the items -// to get all the text. Interfacing with the list box -, 2~ 1521~
, Y09-93-030 34 :
// requires the Presentation Manager application // program interface call WinSendMsg(). It always has // the same 4 parameters window handle // - message ;; // - message-specific parameter or O
// - message-specific parameter or O
ListboxCount = WinSendMsg( hwndListbox, LM_QU~K~ CO~NT, O, O );
// Here's the loop.
for (i = O; i < ListboxCount; i+-~ ) :~' // Use Presentation Manager application program // interface packing macros for the last 2 parameters.
// The first is made of two numbers.
//
; // MPFROM2SHORT( index of item, buffer size ) // The second is a pointer to the buffer.
// MPFROMP( buffer ) ; WinSendMsg( hwndListbox, ,~ LM_QU~;KYl'~ Xl', ' MPFROM2SHORT( i, BufSize ), MPFROMP( pBuffer ) );
// The text for one item is in the buffer now. It // should be copied to be saved somewhere.
}
Example IV

Example IV illustrates C programming language source code for defining the active-state vocabulary to the speech recognizer.

An application program interface for the speech recognizer is used to set it up for recognition. A possible application program interface which can be used is the Speech Manager application program interface that comes with the IBM Speech Server Series product. Source code for a similar application program interface Will be shown below. ;
#include "smapi.h" // Speech Manager application // program interface header file SmArg Args [9]- // Local variable - array of // arguments used to initialize the // speech system.
int iNumArgs;

// Initiali~e the speech system. No parameters are used. ~-~
SmOpen( O, NULL );
// Set up the arguments to be used to make a connection.

,, ;
-;

2~52:~
~ YO9-93-030 35 ~.
// The second parameter in the SmSetArg() function is the // name of the argument. The third parameter is the value.
; //
// Initialize for recognition.
SmSetArg( Args [0], SmNrecognize, TRUE );
// This is the user ID.
SmSetArg( Args [3], SmNuserId, "User" );
// This is the user's trained statistics.
SmSetArg( Args [4], SmNenrollId, "Enroll ID" );
// This is the domain of text to be used.
SmSetArg( Args [5], SmNtask, "Office System" );
// This is a previously created window // which will be used by the speech // recogni~er to communlcate wlth this // appllcation.
SmSetArg( Args [6], SmNwindowHandle, hwndCommunication );
// This is an ID to identify messages // which come from the speech recognizer.
SmSetArg( Args [7], SmNconnectionId, 27 );
// This is the application name.
SmSetArg( Args [8], SmNapplicationName, "Patent Application" );
// Make a connection to the speech recognizer. The last // parameter to this function tells the speech recognizer // to make this call asynchronously.
SmConnect( 9, Args, SmAsynchronous );
// Now there is a connection with the speech recognizer.
// The vocabulary created above can now be defined, // enabled, and used for recognition.
// To define a vocabulary, SmDefineVocab() is used.
// During the define, the speech recognizer looks among a // large pool of words to find a speech model for the word.
// If no speech model exists, one would have to be added // before the word can be used. For those that do exist, a // table is made, including only these, to be used for ; // recognition.
. ~:
// The arguments for SmDefineVocab():
// "ActiveiVocabulary" - name to be associated with the // vocabulary // 3S - number of words in the // vocabulary S
// pWords - a pointer to an array of the // words in a form speclfied by // the application program // interface.
// SmAsynchronous - make the call asynchronously SmDefineVocab( "Active Vocabulary", 35, pWords, SmAsynchronous );
// To enable the vocabulary for recognition, the // application program interface call, // SmEnableVocab() is used.
// The arguments for SmEnableVocab(): -//
// "Active Vocabulary" - name of the vocabulary to // enable 2~1521~

// SmAsynchronous - make the call asynchronously SmEnableVocab( "Active Vocabulary", SmAsynchronous );
// The system is now ready for recognition. To begin // recognizing, the microphone is turned on using // SmMicOn(), and a word is requested using // SmRecognizeWextWord(). Both calls are made // asynchronously here.
SmMicOn( SmAsynchronous )-SmRecognizeNextWord( SmAsynchronous ), Example V

Example V illustrates C programming language source code for outputting a command signal corresponding to the command model from an active-state vocabulary having the best match score.

To begin, a list of commands and command-object associations is manually defined as described above. Each command, with the exception of the global commands, is associated with an object.

Assume the word "RIGHT" from Table 1 is recognized. From the list of command-object associations, the target for the command is known. This target is designated hwndTarget in the example.

HWND hwndTarget;

The action defined by "RIGHT" for this target is to move the target to the right by a previously-defined increment, for example 10 picture elements (pels).
~:~
#de~ine INCREMENT_RIGHT 10 The command is performed on the target using the OS/2 Presentation Manager application program interface call named WinSetWindowPos().
The current window position must be queried first so that the new position can be determined.

SWP 5Wp; // Presentation Manager structure for // window position // Get the initial window position.
// hwndTarget - target window or object ~ - 21~521~

. , ',' // &swp - address where the target's window features will be returned WinQueryWindowPos( hwndTarget, &swp );
// Execute the command, "RIGHT."
// hwndTarget - target window or object // NULLHANDLE - unneeded parameter // swp.x ~ INCREMENT RIGHT
new x-coordinate for window // swp.y - use the same y-coordinate // O, O, - unneeded parameters ~' // SWP_MOVE - tell the window to move ~- WinSetWindowPos( hwndTarget, '', NULLHANDLE, ;, swp.x + INc~;Mn~l_RIGHT, swp .y, ,'~ 0, 0, SWP_MOVE );
.,, Instead, assume the word, "ORANGE," is recognized. From the list ~ of command-object associations, the target for the command is ; known. This is hwndTarget in the example.

~ , HWND hwndTarget;
.: :
' The action defined by "ORANGE" for this target is to select the f entry in the listbox. The command is performed on the target by eending a message, LM_SELECTITEM, to the listbox u~ing the OS/2 Presentation Manager application interface call named WinSendMsg().
First the index of the item has to be found.
~ i SHORT sItem; // item index for querying // Find the recognized word in the list.
// hwndTarget - target window or object // LM_SEARCHSTRING - message being sent // MPFROM2SHORT() - Presentation Manager packing macro // LSS_PREFIX - asls for the item index which // begins with the string in the next // parameter // LIT_FIRST - ask for the first item that // matches // MPFROMP() - Presentation Manager packing macro // pListboxWord - the recognized word "ORANGE"
: //
sItem = (SHORT)WinSendMsg( hwndTarget LM_SEARCHSTRING
~' MPFROM2SHORT( LSS PREFIX
' LIT FIRST 5 ,~ MPFROMP( pListboxWord ) j ; // Select the recognized word.
// hwndTarget - target window or object // LM_SELECTITEM - message being sent // sItem - the item in the list to act upon // TRUE - select the item ':
.~-2 ~ 2 1 0 //
WinSendMsg( hwndTarget, LM SELECTITEM, MPFROMSHORT( sItem ), UPFROULONG( TFUE ) ) '"''~

il ' ~

Claims (25)

1. An interactive computer system comprising:
a processor executing a target computer program having a series of active program states occurring over a series of time periods, said target computer program generating active state image data signals representing an active state image for the active state of the target computer program occurring during each time period, each active state image containing one or more objects;
means for displaying at least a first active-state image for a first active state occurring during a first time period;
means for identifying at least one object displayed in the first active-state image, and for generating from the identified object a list of one or more first active-state commands identifying functions which can be performed in the first active state of the target computer program;
means for storing a system vocabulary of acoustic command models, each acoustic command model representing one or more series of acoustic feature values representing an utterance of one or more words associated with the acoustic command model;
means for identifying a first active-state vocabulary of acoustic command models for the first active state, the first active-state vocabulary comprising the acoustic command models from the system vocabulary representing the first active-state commands;
and a speech recognizer for measuring the value of at least one feature of an utterance during each of a series of successive time intervals within the first time period to produce a series of feature signals, said speech recognizer comparing the measured feature signals to each of the acoustic command models in the first active-state vocabulary to generate a match score for the utterance and each acoustic command model, and said speech recognizer outputting a command signal corresponding to the command model from the first active-state vocabulary having the best match score.

2. An interactive computer system as claimed in Claim 1, characterized in that:
the first active-state vocabulary comprises substantially less than all of the acoustic command models from the system vocabulary;
and the speech recognizer does not compare the measured feature signals for the first time period to any acoustic command model which is not in the first active-state vocabulary.

3. An interactive computer system as claimed in Claim 2, characterized in that:
the display means displays at least a second active-state image different from the first active-state image for a second active state occurring during a second time period different from the first time period;
the object identifying means identifies at least one object displayed in the second active-state image, and generates from the identified object a list of one or more second active-state commands identifying functions which can be performed in the second active state of the target computer program;
the active-state vocabulary identifying means identifies a second active-state vocabulary of acoustic command models for the second active state, the second active-state vocabulary comprising the acoustic command models from the system vocabulary representing the second active-state commands, the second active-state vocabulary being at least partly different from the first active-state vocabulary; and the speech recognizer measures the value of at least one feature of an utterance during each of a series of successive time intervals within the second time period to produce a series of feature signals, said speech recognizer comparing the measured feature signals for the second time period to each of the acoustic command models in the second active-state vocabulary to generate a match score for the utterance and each acoustic command model, and said speech recognizer outputting a command signal corresponding to the command model from the second active-state vocabulary having the best match score.

4. An interactive computer system as claimed in Claim 3, characterized in that the target computer program has only one active state occurring during each time period.

5. An interactive computer system as claimed in Claim 4, characterized in that the target computer program comprises an operating system program.

6. An interactive computer system as claimed in Claim 5, characterized in that the target computer program comprises an application program and an operating system program.

7. An interactive computer system as claimed in Claim 6, characterized in that the target computer program comprises two or more application programs and an operating system program.

8. An interactive computer system as claimed in Claim 6, characterized in that at least some commands for an active-state identify functions which can be performed on the identified objects in the active-state image for the state.

9. An interactive computer system as claimed in Claim 8, characterized in that the identified object in an active-state image comprises one or more of a character, a word, an icon, a button, a scroll bar, a slider, a list box, a menu, a check box, a container, or a notebook.

10. An interactive computer system as claimed in Claim 9, characterized in that the speech recognizer outputs two or more command signals corresponding to the command models from the active-state vocabulary having the best match scores for a given time period.

11. An interactive computer system as claimed in Claim 10, characterized in that the vocabulary of acoustic command models for each active state further comprises a set of global acoustic command models representing global commands identifying functions which can be performed in each active state of the target computer program.

12. An interactive computer system as claimed in Claim 11, characterized in that the display means comprises a display.

13. An interactive computer system as claimed in Claim 11, characterized in that the display means displays both an active-state image for an active state occurring during a time period, and at least a portion of one or more images for program states not occurring during the time period.

14. A method of computer interaction comprising:
executing, on a processor, a target computer program having a series of active program states occurring over a series of time periods, said target computer program generating active state image data signals representing an active state image for the active state of the target computer program occurring during each time period, each active state image containing one or more objects;
displaying at least a first active-state image for a first active state occurring during a first time period;
identifying at least one object displayed in the first active-state image, and generating from the identified object a list of one or more first active-state commands identifying functions which can be performed in the first active state of the target computer program;
storing a system vocabulary of acoustic command models, each acoustic command model representing one or more series of acoustic feature values representing an utterance of one or more words associated with the acoustic command model;
identifying a first active-state vocabulary of acoustic command models for the first active state, the first active-state vocabulary comprising the acoustic command models from the system vocabulary representing the first active-state commands;
measuring the value of at least one feature of an utterance during each of a series of successive time intervals within the first time period to produce a series of feature signals;
comparing the measured feature signals to each of the acoustic command models in the first active-state vocabulary to generate a match score for the utterance and each acoustic command model; and outputting a command signal corresponding to the command model from the first active-state vocabulary having the best match score.

15. A method of computer interaction as claimed in Claim 14, characterized in that:

the first active-state vocabulary comprises substantially less than all of the acoustic command models from the system vocabulary;
and the step of comparing does not compare the measured feature signals for the first time period to any acoustic command model which is not in the first active-state vocabulary.

16. A method of computer interaction as claimed in Claim 15, further comprising the steps of:
displaying at least a second active-state image different from the first active-state image for a second active state occurring during a second time period different from the first time period;
identifying at least one object displayed in the second active-state image, and generating from the identified object a list of one or more second active-state commands identifying functions which can be performed in the second active state of the target computer program;
identifying a second active-state vocabulary of acoustic command models for the second active state, the second active-state vocabulary comprising the acoustic command models from the system vocabulary representing the second active-state commands, the second active-state vocabulary being at least partly different from the first active-state vocabulary;
measuring the value of at least one feature of an utterance during each of a series of successive time intervals within the second time period to produce a series of feature signals;
comparing the measured feature signals for the second time period to each of the acoustic command models in the second active-state vocabulary to generate a match score for the utterance and each acoustic command model; and outputting a command signal corresponding to the command model from the second active-state vocabulary having the best match score.

17. A method of computer interaction as claimed in Claim 16, characterized in that the target computer program has only one active state occurring during each time period.

18. A method of computer interaction as claimed in Claim 17, characterized in that the target computer program comprises an operating system program.

19. A method of computer interaction as claimed in Claim 18, characterized in that the target computer program comprises an application program and an operating system program.

20. A method of computer interaction as claimed in Claim 19, characterized in that the target computer program comprises two or more application programs and an operating system program.

21. A method of computer interaction as claimed in Claim 19, characterized in that at least some commands for an active-state identify functions which can be performed on the identified objects in the active-state image for the state.

22. A method of computer interaction as claimed in Claim 21, characterized in that the identified object in an active-state image comprises one or more of a character, a word, an icon, a button, a scroll bar, a slider, a list box, a menu, a check box, a container, or a notebook.

23. A method of computer interaction as claimed in Claim 22, characterized in that the step of outputting a command signal comprises outputting two or more command signals corresponding to the command models from the active-state vocabulary having the best match scores for a given time period.

24. A method of computer interaction as claimed in Claim 23, characterized in that the vocabulary of acoustic command models for each active state further comprises a set of global acoustic command models representing global commands identifying functions which can be performed in each active state of the target computer program.

25. A method of computer interaction as claimed in Claim 24, further comprising the step of displaying both an active-state image for an active state occurring during a time period, and at least a portion of one or more images for program states not occurring during the time period.