US20040083090A1 - Manager for integrating language technology components - Google Patents

Abstract

A speech system includes a Language Component Manager (LCM) [116] that functions as middleware between language and technology components [117] and a high-level application [115]. In aggregate, the LCM, the language and technology component, and the high-level application form a speech system. The components of the speech system may be distributed. All data and messages in the speech system pass through the LCM. A configuration file defines relationships, such as message and data paths, for the speech system.

Description

RELATED APPLICATIONS
• This application claims priority under 35 U.S.C. § 119 based on U.S. Provisional Application No. 60/419,214 filed Oct. 17, 2002, the disclosure of which is incorporated herein by reference.[0001]
BACKGROUND OF THE INVENTION
• A. Field of the Invention[0002]
• The present invention relates generally to automated speech recognition systems and, more particularly, to the creation of speech systems from basic speech and language technology components.[0003]
• B. Description of Related Art[0004]
• Many speech and language systems include a number of discrete language technology components that are in some way tied together by the overall system. Consider a speech indexing system that is designed to receive an audio signal, convert speech in the audio signal to a text transcription, and annotate the text to include additional information derived from the speech. This speech indexing system may include a number of individually designed speech and language technology components, such as a speech recognizer component, a speaker identification component, a topic detection component, and a name extraction component.[0005]
• The speech recognizer component generates a basic text transcription of the speech. The other components operate on the transcription to generate additional information that describes the transcription. Thus, the speaker identification component provides identifications of each speaker in the transcription, the topic detection component generates key words that define topics for segments of the transcription, and the name extraction component locates and mark all of the proper nouns in the transcription.[0006]
• These basic language technology components may be reused in numerous other speech systems. Conventionally, when designing a new speech system, the designer manually integrates each of the basic language technology components into the system. This may involve individually controlling the start-up and handling of error conditions from the language technology components. Additionally, the designer may manually design structures for controlling the flow of data and control information between the language technology components. Significant resources may be spent on integrating the language technology components into the speech system.[0007]
• Thus, there is a need in the art to improve the integration of basic language technology components into a high-level speech system.[0008]
SUMMARY OF THE INVENTION
• Systems and methods consistent with the present invention include a Language Component Manager (LCM) that provides an interface between multiple language technology components and an external high-level application. Together, the high-level application, the LCM, and the language technology components create a complete speech system. The LCM marshals the interactions of the language technology components and presents a single system-level interface outside of the high-level application. The LCM can help to significantly simplify and expedite the creation of speech systems that use the language technology components.[0009]
• One aspect of the invention is directed to a method for interacting among components of a speech system that include language technology components, a middleware component, and at least one high-level application component. The method includes receiving substantially all data communications in the speech system at the middleware component and forwarding the data communications from the middleware component to a destination, one of the language technology components and the high-level application component, as determined by a configuration file. The method further includes receiving substantially all message communications in the speech system at the middleware component and forwarding the message communications from the middleware component to at least one of the language technology components and the high-level application component, as determined by the configuration file.[0010]
• A second aspect consistent with the invention is directed to a speech system that includes a high-level application component, language technology components, a configuration file, and a language component manager (LCM). The configuration file describes communication paths among the high-level application component and the language technology components. The LCM acts as an intermediary for communications between the high-level application component and the language technology components and between the language technology components. The LCM connects the high-level application component and the language technology components based on the configuration file.[0011]
• Yet another aspect consistent with the present invention is directed to a system integration device. The device includes ports configured to connect with first components that perform basic speech and language processing functions. The device also includes at least one port configured to connect with a high-level application designed to implement a speech service using the functions provided by the first components. Still further, the device includes a configuration file that includes routing information that describes communication paths among the high-level application and the first components. Communication between the first components and between the first components and the high-level application are routed through the device based on the routing information in the configuration file.[0012]
BRIEF DESCRIPTION OF THE DRAWINGS
• The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate the invention and, together with the description, explain the invention. In the drawings,[0013]
• FIG. 1 is a diagram illustrating an exemplary system in which concepts consistent with the invention may be implemented;[0014]
• FIG. 2 is a diagram illustrating functional relationships of the components shown FIG. 1;[0015]
• FIG. 3 is a diagram that defines states of a Language Component Manager (LCM) consistent with an aspect of the invention;[0016]
• FIG. 4 is a diagram illustrating communication paths through the components shown in FIG. 2;[0017]
• FIG. 5 is a diagram illustrating a message transmitted between two components in a speech system;[0018]
• FIGS. 6A and 6B are diagrams illustrating an exemplary distributed application document; and[0019]
• FIG. 7 is a diagram illustrating an exemplary speech application.[0020]
DETAILED DESCRIPTION
• The following detailed description of the invention refers to the accompanying drawings. The same reference numbers may be used in different drawings to identify the same or similar elements. Also, the following detailed description does not limit the invention. Instead, the scope of the invention is defined by the appended claims and equivalents of the claim limitations.[0021]
• A Language Component Manager (LCM) marshals the interactions (both message and data) of a group of speech and language technology components (LCs) and presents a single system-level interface to an outside application. The LCM takes on the responsibility of starting, shutting down, and restarting the LCs as necessary and it provides the high-level application with a well-defined system state. In addition, the LCM defines a framework for a fast system-abort function and it provides the LCs with a centralized logging facility.[0022]
SYSTEM OVERVIEW
• Speech-related applications and systems, as described herein, may be performed on one or more processing devices or networks of processing devices. FIG. 1 is a diagram illustrating an[0023]exemplary system100 in which concepts consistent with the invention may be implemented.System100 includes a number ofcomputing devices101 that each include a computer-readable medium, such asrandom access memory109, coupled to aprocessor108.Computing devices101 may also include a number of additional external or internal devices, such as, without limitation, a mouse, a CD-ROM, a keyboard, and/or a display.
• [0024]Computing devices101 may connect to anetwork102.Network102 may be, for example, a wide area network (WAN), such as the Internet, or a local area network (LAN). Certain ones ofcomputing devices101 may connect tonetwork102 and to each other through a router or switch103.
• In general,[0025]computing devices101 may be any type of computing platform.Computing device101 is exemplary only. Concepts consistent with the present invention can be implemented on any computing device, whether or not connected to a network.
• [0026]Processors108 can be any of a number of well-known computer processors, such as processors from Intel Corporation, of Santa Clara, Calif.Processors108 execute program instructions stored inmemory109.
• [0027]Memory109 may contain application programs and data. In particular, as shown in FIG. 1, one ofmemories109 may contain a high-level speech/language application115.Application115 may communicate with a Language Component Manager (LCM)116 implemented in a manner consistent with the present invention.LCM116 may include software that functions as middleware between high-level application115 and basic speech and language technology components (LCs)117.LCs117 may include, for example, a speech recognizer component, a speaker identification component, a topic detection component, a name extraction component, or other speech or language related components. Together, high-level application115,LCM116, andLCs117 may form a complete speech system. Different high-level applications115 may use different configurations ofLCM116 andLCs117 to create different speech systems. High-level application115 may, for example, use the services ofLCs117 to present an audio indexer speech system to end-users.
• High-[0028]level application115,LCM116, andLCs117 may inter-operate with one another in a distributed manner, as is illustrated in FIG. 1. In other implementations,application115,LCM116, andLCs117 may be implemented on a single computing device.
• FIG. 2 is a diagram illustrating functional relationships of a[0029]speech system200 that is based on high-level application115,LCM116, and LCs117 (labeled as LC1 through LCn). To high-level application115,LCM116 andLCs117 appear as a single unified service. High-level application115 does not directly communicate withLCs117. Instead, high-level application115 communicates withLCM116, which acts as an intermediary between high-level application115 andLCs117. In this manner,LCM116 presents a single system-level interface toapplication115.
• [0030]LCs117 may include speech and language processing components directed to a wide variety of basic speech technologies. As previously mentioned, these speech technologies may include speech recognition, speaker identification, topic detection, and name extraction. Techniques for implementing these technologies are known in the art and will not be further described herein.
LCM STATES
• [0031]LCM116 provides high-level application115 with a well-defined system state. The system state is a function of the system states ofLCs117. FIG. 3 is a diagram that defines the state machine for the LCM system state.LCM116 may present one of five states to high-level application115: downstate301, starting upstate302, abortingstate303, upstate304, and shutting downstate305.
• On startup of[0032]LCM116, the system state defaults to downstate301. In this state,LCs117 are not running. High-level application115 can interact withLCM116 only through predefined administrative interfaces. The administrative interfaces may allow an administrator to log intoLCM116 via, for example, a web interface or a telnet interface. With the web interface, the administrator can interact withLCM116 through a web browser. With the telnet interface, the administrator can interact withLCM116 through a command-line based interface. In general, the administrative interfaces allow the administrator to perform functions, such as viewing error logging information, adjusting error logging verbosity, view a current configuration file, change the LCM configuration, and interact with the LCM by posting messages.
• From down[0033]state301,LCM116 may receive a startup request through the administrative interfaces. In response to the startup request,LCM116starts LCs117 and enters starting upstate302.LCs117 may be started based on initialization information contained in a configuration file, called a distributed application document. When started,LCs117 may boot-up and open a communication channel withLCM116. If any ofLCs117 fail to connect within a pre-specified timeout period,LCM116 assumes a non-recoverable failure of theLC117 and changes the system state to shutting downstate305.
• When an[0034]LC117 reaches a state of operational readiness (e.g., it has loaded and processed its model data), it reports its state toLCM116.LCM116 may respond by activating data and message connections. When all ofLCs117 are ready,LCM116 enters upstate304.
• Up[0035]state304 is the primary operational state ofLCM116. In upstate304,LCM116 is responsible for managing data and message communications betweenLCs117 and high-level application115. While in upstate304,LCM116 may monitorLCs117 for failure and switch its state to shutting downstate305 in the event of a failure.
• In shutting down[0036]state305,LCM116 may discontinue data and message routing activities between the various components (i.e., LCs and high-level applications).LCM116 may then disconnect any connected high-level applications115 and subsequently instruct allLCs117 to shut down. When these actions are complete,LCM116 may enter downstate301.
• Administrators or high-[0037]level applications115 may issue abort requests toLCM116 to disconnect fromLCM116.LCM116 may respond by entering abortingstate303. In this state,LCM116 deactivates message and data routing between theLCs117 and the aborting high-level application or administrator.
• By providing a centralized system state,[0038]LCM116 can control instabilities due to random timing problems betweenLCs117. More particularly, becauseLCM116 controls all communications between components insystem200,LCM116, in the event of an error, can shut down the components in an orderly manner and, thus, avoid system race conditions.
COMPONENT COMMUNICATION
• As previously mentioned,[0039]LCs117 and high-level application115 do not communicate directly with one another. Instead, all communications inspeech system200 are passed throughLCM116, which defines a framework for inter-component communication. FIG. 4 is a diagram illustrating communication paths insystem200.
• As shown in FIG. 4, LC[0040]1 communicates with LCn throughLCM116. Similarly, high-level application115 and LC1-LCn communicate with one another throughLCM116. A benefit of this communication framework is that it minimizes port management inside the LCs and the high-level applications, such asapplication115. Specifically,LCs117 and high-level applications115 only need to communicate with one entity—LCM116.
• The LCM communication framework may provide two ways for communicating: message and data. Message communication can be conceptualized as an asynchronous remote procedure call by an initiating component to a receiving component. Message communication is generally used to initiate actions in[0041]LCs117. Data communication provides a way to pass data units between components, where these data units do not invoke actions in the receiving component. Instead, the data units are processed by the receiving component.
• Generally, the purpose of sending messages is to invoke an action in the receiver of the message. In one implementation, messages are transmitted as serialized extensible markup language (XML) elements. The end of a message may be indicated by a zero byte (i.e. 0×00, a.k.a. NULL-termination). FIG. 5 is a diagram illustrating a[0042]message501 transmitted between twoLCs117, labeled as LC1 and LC2.Message501 is a simple message that instructs LC2 to take a particular action.
• [0043]Message501 has a one-to-one relationship between components, i.e., there is one sender (LC1) and one receiver (LC2) of the message.LCM116 may also support m:n message relationships, in which there can be m senders and n receivers of one particular message, where both m and n are greater than or equal to one. A message can be sent from any one of the m senders for delivery to each of the n receivers.
• Table I, below, illustrates a number of possible system messages that relate to the control of[0044]LCM116.

  TABLE I
  Message Name	Description
  LCM_ABORT	While executing a system-abort operation, the LCM will send an
  	<LCM_ABORT/>to every component (LCs and high-level
  	applications), instructing them to discontinue any system-related
  	activities, reset themselves and then report their readiness.
  LCM_ABORT_DONE	Once a component has reacted appropriately to an abort request from
  	the LCM, it notifies the LCM of its return to operational readiness by
  	sending an <LCM_ABORT_DONE/>.
  LCM_CONNECT	After a high-level application opens a socket connection to the
  	LCM's port, it sends an <LCM_CONNECT name=””/>, identifying
  	itself using the attribute name.
  LCM_CONNECT_ACK	In response to an LCM_CONNECT message from a high-level
  	application, the LCM will send an LCM_CONNECT_ACK to the
  	high-level application.
  LCM_DISCONNECT	The LCM sends an <LCM_DISCONNECT/> message to a high-
  	level application to instruct the high-level application to disconnect.
  LCM_HELLO	After an LC opens a socket connection to the LCM's port, it will send
  	an <LCM_HELLO process-id=””/>, identifying itself via its process
  	ID in the attribute process-id.
  LCM_HELLO_ACK	In response to an LCM_HELLO message from an LC, the LCM will
  	send back an LCM_HELLO_ACK.
  LCM_LOG	This message is sent from LCs to the LCM providing text that the
  	respective LC wishes to add to the system log. The message syntax
  	is: <LCM_LOG level=“ . . . ” message=“ . . . ”/>, where the attri-
  	bute level specifies the logging level, and message contains
  	the logging text.
  LCM_READY	An LC sends an <LCM_READY/> message to the LCM when it has
  	reached operational readiness.
  LCM_SHUTDOWN	The LCM will send an <LCM_SHUTDOWN/> message to an LC to
  	instruct the LC to shut itself down.
  LCM_SYSTEM_ABORT	A high-level application can send an <LCM_SYSTEM_ABORT/>
  	message to the LCM to initiate a system-abort.
  LCM_SYSTEM_READY	An <LCM_SYSTEM_READY/> will be distributed to all LCs and
  	(potentially connected) high-level applications, when the entire
  	system has reached a state of operational readiness, i.e. when the
  	system state is changed to up.

• Generally, the purpose of data communication between components, such as[0045]LCs117 and high-level applications115, is to pass data units, such as packets, of structured information between the components. In contrast to messages, data communication does not generally invoke actions in the receiving component. Instead, the receiving component consumes or processes the information contained in the packets.
• In one implementation consistent with the invention, a data pipe model is used to pass data between components. The data pipe may be a one-to-one relationship between the component that sends/produces data and the component that receives/consumes the data. A data pipe may be implemented by the producer by sending the following pieces of information to LCM[0046]116: the name of the data pipe, a data pipe command, and any arguments required by the data pipe command.LCM116 will then deliver the data pipe command and its arguments (if any) to the appropriate data-pipe consumer. A data pipe may be configured to connectmultiple LCs117 in series. In this manner, data produced by a first LC may be consumed by a second LC, which may then produce a modified version of the data which may then be consumed by a third LC.
• Table II, below, lists three exemplary data pipe commands.[0047]

  TABLE II
  	Com-			Return-value
  	mand	Argu-		(besides
  Command	ID	ments	Effect	error code)
  Clear data-	3	None	Instruct the data	None
  pipe			consumer to remove any
  			unprocessed data packets
  			that it has in its data-
  			receiving buffer
  Get number	4	None	Instruct the data	Unsigned
  of elements			consumer to return the	integer,
  in data-pipe			number of data packets	indicating the
  buffer			that are currently held in	number of data
  			the data-receiving buffer.	packets that are
  				currently in the
  				data-receiving
  				buffer.
  Send data	1	Data	Delivers a data packet to	None
  		packet	the data-pipe consumer.
  			The data-packet can be
  			rejected by the data-
  			consumer using the error
  			code BufferFull. In this
  			case, the data producer is
  			responsible for
  			retransmitting the data
  			packet at a later time.

DISTRIBUTED APPLICATION DOCUMENT
• For a particular high-[0048]level application115, the choice of whichLCs117 to use in the application and the paths for the flow of messages and data betweenLCs117 and the high-level application115 is controlled by a configuration document (or file) referred to herein as a distributed application document (DAD).
• In one implementation, the DAD is an XML document that describes in detail the configuration of an LCM-managed system.[0049]LCM116 parses the DAD in determining the configuration ofspeech system200. A particular system configuration may define the specification ofLCs117, the specification of high-level application115, the message and data communication connections between all the components, and error logging settings. As such, the DAD can be viewed as a complete system specification.
• FIGS. 6A and 6B illustrate an[0050]exemplary DAD600. For ease of explanation, the DAD shown in FIG. 6 is a simplified version of a typical DAD.
• [0051]DAD600 includes a name tag601 that includes the name of the particular high-level application. In this example, the application is “Audio Indexer.”
• Run-[0052]time section602 may include, during run-time, a list of all connectedLCs117 and high-level applications115. These run-time connections may be dynamically changed byLCM116 as the connected components change. In this example, run-time section602 also includes an auto-start tag that indicates thatDAD600 is to be automatically loaded byLCM116 whenLCM116 is initially started.
• The system specification section, located between the tags “<SYSTEM_SPECIFICATION>” (FIG. 6A) and “</SYSTEM_SPECIFICATION>” (FIG. 6B), includes a complete description of the configuration of the components in[0053]speech system200, including the message and data paths. As shown, the “SYSTEM_SPECIFICATION” section includes a “CONNECTION_LISTENERS” section, a “LOGGING_TARGETS” section, a “MESSAGES” section, a “DATA_PIPES” section (FIG. 6B), a “REMOTE_APPLICATIONS” section (FIG. 6B), and a “LANGUAGE_COMPONENTS” section (FIG. 6B). Each of these sections is illustrated in FIG. 6 by corresponding XML tags in which brackets (“< . . . >”) denote the beginning of a section and brackets with a slash (“</ . . . >”) denote the end of a section.
• The CONNECTION_LISTENERS section includes a number of CONNECTION_LISTENER tags that define the low-level connections between the components, including the port of[0054]LCM116 with which the components connect. Additionally, these tags may also define whether each port is a port dedicated to a data producer, data consumer, or message.
• The LOGGING_TARGETS section defines error logging settings for the system, such as target file names. Additionally, the level of detail to which errors are logged can be set in this section through the verbosity setting. Valid values for the verbosity setting according to one implementation consistent with the invention are shown in Table III.[0055]

  TABLE III
  Logging Level	Description
  None	No logging information.
  Error	Indicates a severe exception condition in the LC. This condition often leads to
  	component failure and hence to system shutdown.
  Warning	Unexpected exception-like condition. The LC can avoid failure but this is still a severe
  	exception.
  Info	Provides high-level component state information that might be of general interest.
  Detail	Provide very fine-grain detailed information about the component's system state. This
  	type of information is typically useful for component debugging.

• The MESSAGES section defines the possible flow of messages between the components in[0056]speech system200. The MESSAGES section may include three subsections: “PUBLIC_IN”, “PUBLIC_OUT”, and “PRIVATE.” PUBLIC_IN refers to messages generated by high-level application115 and transmitted toLCM116. PUBLIC_OUT refers to messages fromLCM116 and transmitted to high-level applications115. PRIVATE refers to messages transmitted betweenLCs117 viaLCM116.
• Exemplary message definitions are illustrated for the PUBLIC_OUT subsection of MESSAGES. Messages in PUBLIC_IN and PRIVATE may be implemented similarly. Each[0057]entry604 in the PUBLIC_OUT sub-section includes a message name (“message name” field), the name of the LC from which the message originates (“from” field), and the name of the destination high-level application (“to”) field. In this example, the messages are all destined for the application “decode.” The messages named “passage” and “gap” originate from the LCs identified as “spk_segment.” The message named “word” originates from the LC named “nbscore.” In operation, the LC named “nbscore” may transmit a message to the LCM that includes the label “word.”LCM116, based onDAD600, knows that this message should be forwarded to the high-level application “decode.”
• The DATA_PIPES section defines the possible flow of data between the components in[0058]system200. As with the MESSAGES section, the DATA_PIPES section may include the three sub-sections PUBLIC_IN, PUBLIC_OUT, and PRIVATE. These sections similarly define data pipes between high-level applications115 andLCM116, betweenLCM116 and the high-level applications, and between LCs, respectively.
• Two exemplary data pipe entries are shown in[0059]DAD600 under the PRIVATE sub-section of DATA_PIPES. The first data pipe entry is named “phn2seq.” This data pipe is received byLCM116 from the LC “filter_component” and forwarded to the LC “speaker_seg.” The second data pipe entry is named “seg2fw.” Data transmitted through this data pipe is received byLCM116 from the LC “speaker_seg” and forwarded to the LC “decoder_fw.”
• The “REMOTE_APPLICATIONS” section defines high-[0060]level applications115 that connect toLCM116. One remote application, named “feed_audio” is illustrated as being defined in this section.
• The “LANGUAGE_COMPONENTS” section defines[0061]LCs117 that may communicate withLCM116. OneLC117, named “decoder_phn” is defined in this section.
EXEMPLARY APPLICATION
• FIG. 7 is a diagram illustrating components in an exemplary speech application. The speech application shown in FIG. 7 may be a dialog manager that recognizes speech received over a telephone connection and takes action based on the instructions from the telephone caller.[0062]
• High-[0063]level application701 may include a dialog manager application that responds to speech-based commands from the caller and generates appropriate responses for the caller. The speech processing used to recognize the commands in the speech is primarily performed by LCs703-707. LCs703-707 include DTMF (dual-tone multi-frequency)recognizer component703, voice-over-IP (VoIP)component704,speech segmenter component705, text-to-speech component706, andspeech recognizer component707.
• [0064]DTMF component703 is configured to recognize key presses on a touch-tone telephone based on the corresponding DTMF signal generated by the telephone.VoIP component704 receives the user's telephone call. In this example, the telephone call is received as a packetized signal in a VoIP connection.Speech segmenter component705 segments speech signals into useful segments.Speech segmenter component705 may, for example, segment input speech signals into audio segments corresponding to individual words or sentences. Text-to-speech component706 may generate synthesized speech signals based on input text documents.Speech recognizer component707 performs basic speech recognition of input speech signals.
• [0065]LCM702 acts as an intermediary between high-level application701 and LCs703-707.LCM702 reads a DAD document to determine the message and data interactions between high-level application701 and LCs703-707. For example, the DAD may dictate that speech received atVoIP component704 is to be transmitted via a data pipe tospeech segmenter component705. The output ofspeech segmenter component705 may be forwarded tospeech recognizer component707, the output of which may then be forwarded to high-level application701. Similarly, high-level application701 may respond to callers by generating text that it transmits, viaLCM702, to text-to-speech component706. The synthesized text output from text-to-speech component706 may then be transmitted toVoIP component704 for transmission to the caller. In each of these instances of data transfer, the data is transmitted through and routed byLCM702.
CONCLUSION
• An LCM, as described above, functions as middleware between one or more language and technology components and one or more high-level applications. In aggregate, the LCM, the language and technology component(s), and the high-level application(s) form a speech system. The components of the speech system may be distributed. All data and messages in the speech system pass through the LCM. A configuration file defines relationships, such as message and data paths, for the speech system.[0066]
• The foregoing description of preferred embodiments of the invention provides illustration and description, but is not intended to be exhaustive or to limit the invention to the precise form disclosed. Modifications and variations are possible in light of the above teachings or may be acquired from practice of the invention.[0067]
• Certain portions of the invention have been described as software that performs one or more functions. The software may more generally be implemented as any type of logic. This logic may include hardware, such as application specific integrated circuit or a field programmable gate array, software, or a combination of hardware and software.[0068]
• No element, act, or instruction used in the description of the present application should be construed as critical or essential to the invention unless explicitly described as such. Also, as used herein, the article “a” is intended to include one or more items. Where only one item is intended, the term “one” or similar language is used.[0069]
• The scope of the invention is defined by the claims and their equivalents.[0070]

Claims (29)

What is claimed:

1. A method for interacting among a plurality of components of a speech system, the plurality of components including language technology components, a middleware component, and at least one high-level application component, the method comprising:

receiving substantially all data communications in the speech system at the middleware component;

forwarding the data communications from the middleware component to a destination one of the language technology components and the high-level application component, as determined by a configuration file;

receiving substantially all message communications in the speech system at the middleware component; and

forwarding the message communications from the middleware component to at least one of the language technology components and the high-level application component, as determined by the configuration file.

2. The method ofclaim 1, wherein the speech and language technology components include at least one of a speech recognizer component, a speaker identification component, a topic detection component, and a name extraction component.

3. The method ofclaim 1, wherein the language technology components, the middleware component, and the at least one high-level application component are software modules that arc distributed over a plurality of computing devices.

4. The method ofclaim 1, further comprising:

logging errors encountered by the language technology components at the middleware component.

5. The method ofclaim 1, wherein the middleware component presents a single system state to the high-level application component, the single system state being based on individual states of the language technology components.

6. The method ofclaim 5, wherein the single system state is selected from:

a starting up state that indicates that at least one of the language technology components is initializing,

a down state that indicates that at least one of the language technology components is not available, and

an up state that indicates that all of the language technology components are available.

7. The method ofclaim 6, wherein the single system state is further selected from:

an aborting state that indicates that the middleware component is disconnecting from the high-level application component, and

a shutting down state in which the middleware component discontinues communications with the language technology components.

8. The method ofclaim 1, wherein the configuration file is an extensible markup language (XML) document.

9. The method ofclaim 1, wherein the data communications are implemented as data pipes.

10. A speech system comprising:

a high-level application component;

a plurality of language technology components;

at least one configuration file that describes communication paths among the high-level application component and the plurality of language technology components; and

a language component manager (LCM) configured to act as an intermediary for communications between the high-level application component and the language technology components and between ones of the language technology components, the LCM connecting the high-level application component and the language technology components based on the configuration file.

11. The speech system ofclaim 10, wherein the configuration file is a complete system specification for the speech system.

12. The speech system ofclaim 10, wherein the high-level application component provides speech services to end-users.

13. The speech system ofclaim 10, wherein the LCM presents a single interface to the high-level application component.

14. The speech system ofclaim 10, wherein the speech system performs speech recognition.

15. The speech system ofclaim 10, wherein the language technology components include at least one of a speech recognizer component, a speaker identification component, a topic detection component, and a name extraction component.

16. The speech system ofclaim 10, wherein the language technology components, the LCM, and the high-level application component are software modules that are distributed over a plurality of computing devices.

17. The speech system ofclaim 10, wherein the LCM presents a single system state to the high-level application component, the single system state being based on individual states of the language technology components.

18. The speech system ofclaim 17, wherein the single system state is selected from:

a starting up state that indicates that at least one of the language technology components is initializing,

a down state that indicates that at least one of the language technology components are not available, and

an up state that indicates the speech system is available.

19. The speech system ofclaim 10, wherein the configuration file is an extensible markup language (XML) document.

20. A system integration device comprising:

ports configured to connect with a plurality of components configured to perform speech and language processing functions;

at least one port configured to connect with a high-level application designed to implement a speech service using the functions provided by the components; and

a configuration file that includes routing information that describes communication paths among the high-level application and the plurality of first components, communications between the first components and between the first components and the high-level application being routed through the system integration device based on the routing information in the configuration file.

21. The device ofclaim 20, wherein the configuration file is a complete system specification for a speech system.

22. The device ofclaim 20, wherein the high-level application provides speech services to end-users.

23. The device ofclaim 20, wherein the components include at least one of a speech recognizer component, a speaker identification component, a topic detection component, and a name extraction component.

24. The device ofclaim 20, wherein the system integration device presents a single system state to the high-level application, the single system state being based on individual states of the components.

25. The device ofclaim 24, wherein the single system state is selected from:

a starting up state that indicates that at least one of the components is initializing,

a down state that indicates that at least one of the components is not available, and

an up state that indicates the plurality of components are available.

26. The device ofclaim 20, wherein the configuration file is an extensible markup language (XML) document.

27. A device for interacting among a plurality of components of a speech system, the plurality of components including one or more language technology components and at least one high-level application component, the device comprising:

means for describing message and data connections between the one or more language technology components and the at least one high-level application component;

means for receiving substantially all data communications in the speech system from the one or more language technology components and the at least one high-level application component;

means for forwarding the received data communications to a designated one of the one or more language technology components and the at least one high-level application component based on the means for describing;

means for receiving substantially all message communications in the speech system from the one or more language technology components and the at least one high-level application component; and

means for forwarding the message communications to at least one of the one or more language technology components and the at least one high-level application component based on the means for describing.

28. A computer-readable medium containing programming instructions for execution by a processor, the computer-readable medium comprising:

instructions for causing the processor to connect with a plurality of components configured to perform speech and language processing functions;

instructions for causing the processor to connect with a high-level application designed to implement a speech service using the functions provided by the components; and

instructions that define a configuration file that includes routing information that describes communication paths among the high-level application and the plurality of components, the communications between the components and between the components and the high-level application being routed through the processor based on the routing information in the configuration file.

29. The computer-readable medium ofclaim 28, wherein the configuration file is a complete system specification for a speech system.

Images