JP2006011066A - Voice recognition/synthesis system, synchronous control method, synchronous control program and synchronous controller - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To synchronously control two pieces of information, i.e., voice data and a voice recognition/synthesis control command. <P>SOLUTION: In a voice recognition process, a voice input means 101 starts inputting voice data at time A0, successively divides inputted data into packets, adds identifiers and transmits them to a voice recognition means 102. The voice recognition means 102 starts receiving the packets from the voice input means 101 at time A1. Then, a control means 103 sets an identifier "AI0" and a control command "AC1", which instructs a start of the voice recognition process, at time A2 and transmits them to the voice recognition means 102. When the voice recognition means 102 receives the control command "AC1" including the identifier "AI0" at time A4, the means 102 reads the voice data, to which the specified identifier "AI0" is added, from a packet holding means 102d and starts the voice recognition process from the read packets. Thus, the voice recognition process is conducted for the appropriate voice data to be processed. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a speech recognition / synthesis system that performs speech recognition processing for analyzing input speech data or / and speech synthesis processing for generating speech data, and more particularly to speech data and control commands for speech recognition / synthesis processing. The present invention relates to a speech recognition / synthesis system to be synchronized, a synchronization control method, a synchronization control program, and a synchronization control apparatus.

  A speech application system that applies speech recognition technology that analyzes input speech data and speech synthesis technology that generates speech data includes a speech data input / output unit that inputs and outputs speech data, and speech recognition processing and / or speech synthesis processing. And a control unit that controls the voice data input / output unit and the voice recognition / synthesis processing unit.

  Conventionally, the voice application system as described above has been realized in units of program components and hardware blocks. However, with recent advances in communication network technology, an increasing number of examples are realized in which each unit constituting a voice application system is realized in units of servers.

  However, in a spoken dialogue system consisting of sparse components implemented in units of servers, data is exchanged between components via a communication network where transmission delay may occur. Necessary.

  If it is assumed that the time synchronization control mechanism is not provided, each unit cannot properly operate in a coordinated manner, and the recognition performance of the voice data is missed, the noise is falsely detected, the sound / language model is inappropriate, and the parameter is applied. Various problems such as degradation and degradation of output voice quality may occur.

  For example, audio data and other media data (images, etc.) developed on the same time axis are separated by the transmitting side terminal and transmitted through different systems, and then each media data is received by the receiving side terminal. It is a common problem in the field of multimedia data communication to be able to receive and receive synchronous organization on a single time axis.

  Patent Document 1 discloses a technique in which multimedia data developed on the same time axis is demultiplexed for each medium and transmitted separately, and the data is synchronously output at the receiving side device. Has been.

  Patent Document 2 discloses a voice application system that generates voice data and moving image data to which time synchronization control information is added from given text.

JP 2003-204492 A JP 2003-216173 A

  However, the above-described conventional technique has a problem that synchronization control cannot be performed on information existing on essentially different time axes such as voice data and voice recognition / synthesis control commands.

  That is, since it is not obvious at which timing the two pieces of information of the voice data and the voice recognition / synthesis control command should be synchronized, the synchronization control cannot be performed.

  Patent Documents 1 and 2 disclose a technique for synchronizing a plurality of media data. However, each media data subject to synchronization control is originally developed on the same time axis. Since it should be, it is obvious at what timing it should be synchronized.

  However, since there is no obvious synchronization timing between the voice data and the voice recognition / synthesis control command, the voice data and the voice recognition / synthesis control are applied even if the techniques described in the above patent documents are applied. Synchronous control with the command cannot be performed.

  In this case, if the operation is performed assuming an appropriate synchronization timing (for example, the latest speech data is regarded as the first data to be recognized at the time when the speech recognition / synthesis processing unit receives the speech recognition start command). Depending on the system environment or the like, the above problem may not occur.

  However, the method that assumes the synchronization timing as described above will fail in an environment where the transmission delay is very large or in a severe situation where control commands corresponding to audio data are issued one after another. Synchronous control between the voice data and the voice recognition / synthesis control command cannot be performed reliably.

  An object of the present invention is to solve the above-mentioned problems and to control two pieces of information of voice data and a voice recognition / synthesis control command in synchronization.

  The speech recognition / synthesis system of the present invention is a speech recognition / synthesis system (for example, speech recognition / synthesis system 100, 300, 500) that performs speech recognition processing for analyzing input speech data and / or speech synthesis processing for generating speech data. , 700, 900, 1000), and identification information (for example, identifier) for specifying the divided audio data to be processed among the respective divided audio data (for example, audio data packets) obtained by dividing the audio data into a plurality of sections. Is set to the control command according to the control command from the control command means (for example, the control means 103, 303, 504, 703, 905, the voice conversation management server 1001) and the control command means. Speech recognition processing and / or speech synthesis processing is performed on the speech segmented data specified by the identification information. Cormorant sound processing means (e.g., speech recognition means 102,302,502,503,902, sound generating means 701,903) characterized by comprising a, a.

  With the above configuration, even if the voice data transmission system and the control command transmission system are independent, the voice division data to be processed in the voice recognition process or / and the voice synthesis process can be reduced. The voice data and the control command can be controlled in synchronization with each other.

  Identification information adding means (for example, identifier giving means 101d) for adding identification information uniquely identified in the system may be provided to each piece of voice divided data obtained by dividing the input voice data into a plurality of sections.

  Voice input processing means (for example, voice input processing means 101b) that performs voice data input processing and voice divided data (for example, voice data packets) obtained by dividing the voice data input by the voice input processing means into a plurality of sections are generated. Voice data dividing means (for example, packet dividing means 101c).

  The control command means issues a control command in which the execution time of the voice recognition process or / and the voice synthesis process is set, and the voice processing means executes the execution time set in the control command according to the control command from the control command means Then, the voice recognition process or / and the voice synthesis process may be performed on the voice division data specified by the identification information set in the control command.

  A plurality of speech processing means (for example, speech recognition means 302a and 302b) and a processing result integration means (for example, result integration means 303c) for integrating the results of speech recognition processing and / or speech synthesis processing of each of the plurality of speech processing means. ) May be provided.

  The control command means is a control command for one voice processing means (for example, the first voice recognition means) of a plurality of voice processing means (for example, the first voice recognition means 502 and the second voice recognition means 503). (For example, the control command “CC1” shown in FIG. 6), and one sound processing unit sends part or all of the control command from the control command unit (for example, the control command “CC2” shown in FIG. 6) to the other You may be set as the structure which has the control command transfer means (for example, voice recognition control means 502a) which transfers to a voice processing means.

  One voice processing means uses one section or all sections (for example, a voice data packet to which the identifier “CI10” shown in FIG. 6 is added) as the processing target voice data instructed by the control command from the control command section. It may be configured to have voice data transfer means (for example, packet transmission / reception means 502e) for transferring to the voice processing means.

  The identification information adding means adds, as identification information, for example, a time stamp, a serial number, a voice interaction processing sequence number by voice recognition processing and / or voice synthesis processing, or a combination thereof to each voice divided data.

  You may provide the identification information management means (for example, control means 103) which provides the function to manage the time order of identification information. The identification information management means synchronizes the absolute time used by each component constituting the system (for example, synchronizes using the time information from the NTP server), and the identification information has a specific absolute time (for example, identification). Corresponding time (when information is given) has a function of managing the time order (for example, given order) of each identification information.

  Also, the synchronization control method of the present invention is a speech recognition / synthesis system (for example, speech recognition / synthesis system 100, 300, 500) that performs speech recognition processing for analyzing input speech data and / or speech synthesis processing for generating speech data. , 700, 900, 1000), which is a synchronization control method for each piece of voice divided data (for example, voice data packet) in which voice data is divided into a plurality of sections, each of which is divided into a plurality of sections. A control command (for example, command “AC1” in FIG. 2) in which identification information (for example, an identifier) for specifying the processing-target audio divided data among the audio divided data is set, and is set in the control command according to the control command. Audio segmentation data specified by the identification information (for example, an audio data packet to which an identifier “AI0” is added) And performing speech recognition processing and / or speech synthesis processing to.

  With the above configuration, even if the voice data transmission system and the control command transmission system are independent, the voice division data to be processed in the voice recognition process or / and the voice synthesis process can be reduced. The voice data and the control command can be controlled in synchronization with each other.

  Identification information uniquely identified in the system may be added to each piece of voice divided data obtained by dividing the input voice data into a plurality of sections.

  It may be configured to perform voice data input processing and generate voice divided data obtained by dividing the voice data input by the input processing into a plurality of sections.

  The identification set in the control command is issued when the execution time set in the control command is issued according to the control command after issuing the control command in which the execution time of the voice recognition process or / and the voice synthesis process is set The voice recognition process or / and the voice synthesis process may be performed on the voice division data specified by the information.

  You may be comprised so that the process result of the several speech recognition process or / and speech synthesis process which were performed by the different process means according to the control command may be integrated.

  After performing speech recognition processing or / and speech synthesis processing according to the control command, part or all of the control command is transferred to other processing means, and according to the transferred control command, speech recognition processing or / And may be configured to perform speech synthesis processing.

  One section or all sections of the audio data to be processed instructed by the control command from the control command means may be transferred to another processing means.

  As the identification information to be added to each voice division data, for example, any one of a time stamp, a serial number, a voice interaction processing sequence number by voice recognition processing and / or voice synthesis processing, or a combination thereof is used.

  Further, the synchronization control program of the present invention is a speech recognition / synthesis system (for example, speech recognition / synthesis system 100, 300, 500) that performs speech recognition processing for analyzing input speech data and / or speech synthesis processing for generating speech data. , 700, 900, 1000), a computer program (for example, a voice input unit 901, a voice recognition unit 902, a voice generation unit 903, a voice output unit) constituting a voice recognition / synthesis system. 904, issuing to the control means 905) a control command in which identification information (for example, an identifier) for specifying the audio division data to be processed among the audio division data obtained by dividing the audio data into a plurality of sections is set. And according to the control command, it is specified by the identification information set in the control command. It is intended for executing and performing the speech recognition processing and / or speech synthesis processing to voice data segment.

  With the above configuration, even if the voice data transmission system and the control command transmission system in the voice recognition / synthesis system are independent, the computer constituting the voice recognition / synthesis system can receive voice. It is possible to strictly specify the voice division data to be processed in the recognition process or / and the voice synthesis process, and to control the voice data and the control command in synchronization.

  The computer may further be configured to execute a step of adding identification information uniquely identified in the system to each of the divided audio data obtained by dividing the input audio data into a plurality of sections.

  The computer may be further configured to execute a step of performing voice data input processing and a step of generating voice division data obtained by dividing the voice data input by the input processing into a plurality of sections.

  As the identification information to be added to each voice division data, for example, any one of a time stamp, a serial number, a voice interaction processing sequence number by voice recognition processing and / or voice synthesis processing, or a combination thereof is used.

  Further, the synchronization control apparatus of the present invention includes a voice input processing means (for example, voice input processing means 101b) for performing voice data input processing, and voice division in which voice data input by the voice input processing means is divided into a plurality of sections. Audio data dividing means for generating data (for example, packet dividing means 101c) and identification for adding identification information (for example, identifier) uniquely identified in the system to each audio divided data divided by the audio data dividing means Speech processing means (for example, speech recognition means 102) for performing speech recognition processing for analyzing speech data input by the information addition means (for example, identifier giving means 101d) and / or speech synthesis processing for generating speech data. ) For which identification information for specifying the audio segmentation data to be processed is set A control command means for issuing a decree (e.g., control unit 103), characterized by comprising a.

  With the above configuration, even if the voice data transmission system and the control command transmission system are independent, the voice division data to be processed in the voice recognition process or / and the voice synthesis process can be reduced. The voice processing means can control the voice data and the control command in synchronization with each other.

  According to the present invention, even when the voice data transmission system and the control command transmission system are independent, the voice division data to be processed in the voice recognition process or / and the voice synthesis process is strictly specified. The voice data and the control command can be synchronized and controlled.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
Embodiment 1 FIG.
FIG. 1 is a block diagram showing a configuration example of a speech recognition / synthesis system 100 according to the first embodiment of the present invention.

  As shown in FIG. 1, the speech recognition / synthesis system 100 of this example includes a speech input unit 101, a speech recognition unit 102, and a control unit 103. The voice input unit 101, the voice recognition unit 102, and the control unit 103 are connected by a transmission unit 104.

  The voice input unit 101 includes a voice input control unit 101a, a voice input processing unit 101b, a packet division unit 101c, an identifier assigning unit 101d, a packet holding unit 101e, and a packet transmission unit 101f.

  The voice input unit 101 executes various processes such as a process of inputting voice data based on a control command from the control unit 103 and a process of dividing the input voice data into packets and transmitting the packets to the voice recognition unit 102. .

  The voice input control means 101a receives control commands from other components such as the control means 103, and controls the operation of the voice input means 101 as a whole based on the received control commands. In addition, the voice input control unit 101a transmits information indicating the status of voice input processing and the like to other components in response to a request from the other components.

  The voice input processing unit 101b receives voice data from the outside and performs various processes on the received voice data as necessary. For example, the audio input processing means 101b performs A / D conversion if the input audio data is analog audio, and performs decoding processing if the input audio data is data that has been subjected to some encoding. Note that the voice input processing unit 101b may have a function of performing noise suppression processing or the like.

  The packet dividing unit 101c divides the audio data into units called packets by dividing the audio data into appropriate sections and adding additional information necessary for reproducing the audio data for each section. As the auxiliary information, a serial number, a time stamp, a packet size, a quantization parameter, and the like are used.

  The identifier assigning unit 101d executes a process for assigning an identifier for uniquely identifying each packet generated by the packet dividing unit 101c.

  The packet holding unit 101e executes processing for buffering packets as necessary.

  Actually, since the number of packets that the packet holding unit 101e can hold at a time is finite, packets may be lost. When a packet is lost, there arises a problem that a processing request cannot be made for the lost packet.

  However, in many cases, the packets that need to be held are limited to some extent by the flow of the conversation, so if an appropriate buffer management algorithm is used, there is little problem in practical use.

  If there is a problem, the control means 103 may issue a control command so as to keep holding packets required from the dialogue flow. Or you may make it acquire again from the other component holding a required packet.

  The packet transmission unit 101f executes a process of transmitting the packet held in the packet holding unit 101e to other components such as the voice recognition unit 102.

  The voice recognition unit 102 includes a voice recognition control unit 102a, a voice recognition processing unit 102b, an identifier determination unit 102c, a packet holding unit 102d, and a packet reception unit 102e.

  The voice recognition unit 102 recognizes a voice in response to a control command from the control unit 103, and outputs a recognition result in response to a request from the control unit 103.

  The voice recognition control unit 102a receives a control command from another component such as the control unit 103, and controls the entire operation of the voice recognition unit 102 based on the received control command. Further, the voice recognition control unit 102a transmits information such as a recognition result held in response to a request from another component.

  The voice recognition processing unit 102b performs voice recognition processing on the voice data selected by the identifier determination unit 102c and holds the result.

  The identifier determination unit 102c selects a specific packet designated by the control command based on the identifier given to the packet, and transmits it to the voice recognition processing unit 102b.

  The packet holding unit 102d executes processing for buffering packets as necessary.

  The packet receiving unit 102e receives a packet of audio data from other components such as the audio input unit 101 and stores it in the packet holding unit 102d.

  The control means 103 includes user interface means 103a and dialogue management means 103b.

  The control unit 103 controls the voice input unit 101 and the voice recognition unit 102 to cooperate so that the entire system 100 operates as one voice dialogue system.

  The user interface unit 103a is a user interface device other than a voice, such as a display, a button, a mouse, a keyboard, and the like, and performs an interaction with the user to assist a voice conversation.

  The dialog management unit 103b manages the progress of the dialog based on information obtained from the user interface unit 103a and the voice recognition unit 102, and simultaneously transmits a control command to other components as necessary.

  The dialogue management unit 103b receives and interprets control commands from other components, and controls the whole by transmitting some information as necessary.

  The transmission means 104 is a communication network for transmitting and receiving control commands and voice data packets between the components shown in FIG. In FIG. 1, the system that transmits the control command and the system that transmits the voice data packet are independent systems, but they may share the same system.

  As the identifier, a serial number, a sequence number, a time stamp, or the like is used. Hereinafter, each identifier will be described.

  The serial number is incidental information that is widely used to indicate the order relationship of packets. Since the serial number is generally expressed as a finite integer, it is unique only within a limited time.

  When a serial number is used, a mechanism for detecting overflow of the serial number is separately required. However, in practice, it is only necessary to check the magnitude relationship with the preceding number, and it is simple.

  The serial number has a unique value between one component set and another set. In the speech recognition / synthesis system 100 shown in FIG. 1, there is no problem because there is only one set of the speech input means 101 and the speech recognition means 102, but the speech recognition / synthesis system in the second embodiment to be described later. When there are a plurality of pairs of voice input means and voice recognition means such as 300 (see FIG. 3) and each of them is operating asynchronously, there is no serial number uniqueness between each pair. become.

  The sequence number is a number for identifying audio data belonging to a certain processing sequence. For example, the sequence number is assigned in a manner such as No. 1 for the first utterance and No. 2 for the second utterance. Thus, the sequence number assigner (in this case, the voice input means 101) needs to divide the voice data into some units. However, since it is relatively easy to implement, it is worth considering its use in limited applications.

  The time stamp is a method of using the time when certain audio data is processed as an identifier as it is. The time stamp is unique within an almost infinite time, and is compatible with audio data that is time-series data. However, it is necessary to align the absolute time among the constituent elements constituting the system 100. This is not necessary if each component operates on a single computer. However, when each component operates on a separate computer as in this example, for example, by using a mechanism such as NTP (Network Time Protocol) for time synchronization together, each component It is necessary to keep the absolute time in between.

  The identifier used in this example is desirably one that can uniquely identify each audio data in the entire system 100. At least, a set of components that directly exchange voice data needs to be an identifier that can guarantee sufficient uniqueness within a limited time.

  Of the above identifier examples, the most suitable identifier used in this example is a time stamp. This is because it has good compatibility with voice data that is time-series data, and it is easy for a voice dialogue system developer to grasp intuitively. Further, since the time stamp is data indicating time, it can be easily combined with the scheduling function. Therefore, in this example, a time stamp is used as an identifier.

  In addition, the system 100 of this example has a function for managing the time order of identification information. This function is provided in the control means 103, for example. With this function, the control unit 103 stores each identification information in association with the time at which each identification information was given. For this time, for example, a time used in common in a system provided from an NTP server that manages absolute time is used. What is necessary is just to make it acquire each identification information and the information regarding the time when each identification information was given from the terminal device (for example, audio | voice input means 101) which has a function which provides identification information, for example. With this configuration, the system 100 can be provided with a function for managing the time order of each piece of identification information.

Next, the operation of the speech recognition / synthesis system 100 of this example will be described.
FIG. 2 is a time chart showing an example of speech recognition processing by the speech recognition / synthesis system 100 of this example.

  In the time chart shown in FIG. 2, the vertical axis represents the passage of time from time A0 to time A14, and represents a future event as it progresses downward. In FIG. 2, solid arrows indicate the flow of control commands. A broken arrow represents the flow of audio data.

  In the voice recognition process, first, the voice input unit 101 starts to input voice data from time A0, and sequentially transmits the input voice data to the voice recognition unit 102 while dividing the packet into packets and assigning identifiers. .

  It is assumed that the identifier added to the packet first generated from the input voice data is “AI0”.

  The voice recognition unit 102 starts receiving the packet from the voice input unit 101 at time A1. However, at time A1, the speech recognition process has not yet started. For this reason, the voice recognition means 102 stores each received packet in the packet holding means 102d.

  Next, the control means 103 transmits a control command “AC1” instructing the start of the voice recognition process to the voice recognition means 102 at time A2.

  When transmitting the control command “AC1”, the control unit 103 can specify an arbitrary identifier for specifying the target data of the speech recognition process and include it in the control command “AC1”. Here, since the time when the voice data input process is started by the voice input unit 101 is the time A0, the identifier “AI0” is included in the control command “AC1”.

  When the voice recognition unit 102 receives the control command “AC1” including the identifier “AI0” at time A4, the voice recognition unit 102 reads the voice data to which the designated identifier “AI0” is added from the packet holding unit 102d, and reads the packet from the read packet. Starts speech recognition processing.

  Thereafter, when the voice recognition process is completed at time A 6, the voice recognition unit 102 transmits a voice recognition process completion notification “AC 1 ′” indicating that to the control unit 103.

  When the voice recognition processing completion notification “AC1 ′” is received, the control unit 103 temporarily stops voice input, and transmits a voice input processing stop command “AC2” to the voice input unit 101 at time A7. To do.

  When the voice input processing stop command “AC2” is received at time A8, the voice input means 101 stops sending voice data packets to the voice recognition means 102 at that time.

  Note that, from time A6 when the voice recognition means 102 completes the voice recognition processing to time A8 when the voice input means 101 stops sending voice data packets, the voice input means 101 sends a voice data packet to the voice recognition means 102. Continues to be sent. That is, the voice data packet is transmitted to the voice recognition means 102 until the time A8 after the time A6 when the voice recognition means 102 completes the voice recognition processing. For this reason, the packet holding means 102d included in the voice recognition means 102 holds voice data packets received from time A6 and transmitted from the voice input means 101 until time A8.

  Thereafter, the control means 103 transmits a command “AC3” to the voice input means 101 so as to resume voice input. When the instruction “AC3” is received at time A9, the voice input unit 101 resumes the packet transmission process to the voice recognition unit 102.

  In this example, it is assumed that the identifier added to the packet transmitted first after the restart of transmission processing at time A9 is the identifier “AI9”.

  Since the time when the voice input means 101 resumes the voice data packet transmission processing is time A, the control means 103 starts voice recognition using the packet with the identifier “AI9” as voice data to be processed. Thus, the voice recognition start command “AC4” including the identifier “AI9” is transmitted to the voice recognition means 102.

  In this example, it is assumed that the control unit 103 separately transmits a voice recognition start command “AC5” at time A10 immediately after transmitting the command “AC4”.

  In this example, even though the voice recognition start command “AC4” and the voice recognition start command “AC5” are sequentially transmitted, for example, due to the state of the transmission path, the command “AC4” is preceded. It is assumed that the command “AC5” is received by the voice recognition unit 102.

  Even in such a case, since each command includes an identifier for identifying the voice data to be processed, the voice recognition unit 102 can accurately determine from which voice data the voice recognition process should be started. Can be judged. Therefore, speech recognition processing is performed on appropriate speech data to be processed.

  As described above, in the first embodiment described above, an identifier is assigned to the voice data packet and a control command specifying an identifier for specifying the voice data to be processed is issued. Therefore, it is possible to accurately specify the audio data to be processed.

  If such a method is not used, a part of the voice data that should be the target of the recognition process is discarded, or conversely, the voice data including the voice data that should not be the target of the recognition process is subjected to the voice recognition process. There is a fear.

  In the case of the command “AC1” shown in FIG. 2, a packet having the identifier “AI0” (subsequent voice data) is designated as the target of the voice recognition process starting from time A4, but is held by the packet holding unit 102d. Any packet that arrived in the past can be specified as long as it is within the specified range.

  Alternatively, a packet that arrives after time A4 may be specified. In this case, the start of the speech recognition process is postponed until the packet arrives.

  In the example shown in FIG. 2, the input of voice data is interrupted from time A8 to A9, but old packets still remain in the packet holding means 102d inside the voice recognition means 102. For this reason, after resuming the voice data input process, the packet stored at the packet holding means 102d is apparently continuous with the packet transmitted at time A8 immediately before the interruption and the packet at time A9 transmitted immediately after resumption. And seems to be held.

  If the packets held in the packet holding unit 102d are sequentially processed in the voice recognition process after the restart, the recognition accuracy deteriorates due to the discontinuity of the voice data at the time A8 and the time A9. Will end up. However, in the case of FIG. 2, since the identifier “AI9” is designated by the recognition start command “AC4” issued at the time of restart, it is affected by the discontinuity of the audio data. However, it is possible to prevent deterioration of recognition accuracy.

  Further, in the first embodiment described above, an identifier is given to the voice data packet, and a control command is issued by designating an identifier of voice data to be handled by a certain process. Even if the transmission time and the reception time of the command are complicated, the processing based on each control command can be performed on the appropriate audio data.

  The control command “AC4” and the control command “AC5” in FIG. 2 are examples when the transmission time and the reception time of a plurality of control commands are complicated.

  In the recognition processing command “AC4”, the identifier “AI9” is specified to specify the audio data to be processed, and in the recognition processing command “AC5”, the identifier “AI10” is specified to specify the audio data to be processed. Suppose.

  In this case, regardless of the order in which the commands “AC4” and “AC5” arrive, the speech recognition processing unit 102 appropriately performs speech recognition processing on the speech data to be processed specified by each command.

  On the other hand, conventionally, since different parameters are used for the command “AC4” and the command “AC5”, if the arrival order of the two is changed from the intended order, the voice data and the parameter of each command are changed. There is a possibility that the relationship becomes inappropriate and the recognition accuracy deteriorates.

  In the first embodiment described above, since the identifier is assigned to the voice data packet in the voice input unit 101, the voice data packet is closed regardless of the type or state of the other components. Data can be processed appropriately. That is, the voice input means 101 and the other components can achieve close cooperation between the control command and the voice data while maintaining independence at a high level.

Embodiment 2. FIG.
Next, a second embodiment of the present invention will be described with reference to the drawings.
FIG. 3 is a block diagram illustrating a configuration example of the speech recognition / synthesis system 300 according to the second embodiment of the present invention.

  As shown in FIG. 3, the speech recognition / synthesis system 300 of this example includes speech input means 301, a set 302 of speech recognition means 302 a to 302 n, and control means 303. The voice input unit 301, the plurality of voice recognition units 302a to 302n, and the control unit 303 are connected by a transmission unit 304, respectively.

  The voice input unit 301 has the same configuration as the voice input unit 101 described above. The set 302 of speech recognition means is composed of two or more speech recognition means. Each of the voice recognition units 302a to 302n has the same configuration as the voice recognition unit 102 described above. The transmission unit 304 has the same configuration as the transmission unit 104 described above.

  The control means 303 includes user interface means 303a, dialogue management means 303b, and result integration means 303c. The user interface unit 303a and the dialogue management unit 303b have the same configuration as the user interface unit 103a and the dialogue management unit 103b in the control unit 103 described above, respectively.

  The result integration unit 303c evaluates the recognition results respectively received from the plurality of speech recognition units 302a to 302n by some method, integrates the results, and handles them in the same manner as the speech recognition results acquired from the single speech recognition unit. To be able to. Specifically, for example, the recognition result of the speech recognition unit having high reliability is adopted for the speech recognition result of each section in the speech data to be processed (for example, the speech recognition unit 302a is reliable for address speech recognition). For speech recognition of names, the speech recognition means 302b knows in advance information such as the highest reliability), and the adopted recognition results may be connected.

  The speech recognition / synthesis system 300 of this example executes speech recognition processing on a plurality of speech recognition units 302a to 302n for the same speech data input from a single speech input unit 301 under different conditions. Let

  Then, the control unit 303 causes the result integrating unit 303c to execute a process of integrating the slightly different results derived by the voice recognition units 302a to 302n.

Next, the operation of the speech recognition / synthesis system 300 of this example will be described.
FIG. 4 is a time chart showing an example of speech recognition processing by the speech recognition / synthesis system 300 of this example.

  The voice recognition process shown in FIG. 4 shows the exchange of voice data and control commands between the constituent elements.

  The voice input unit 301 adds the identifier “BI0” to the voice data input at time B0, and transmits it to each of the voice recognition units 302a to 302c.

  Here, it is assumed that the time at which the voice data transmitted by the voice input unit 301 arrives at each of the voice recognition units 302a to 302c is different. As shown in FIG. 4, in this example, it is assumed that the voice recognition means 302c arrives at time B1, the voice recognition means 302b arrives at time B2, and the voice recognition means 302a arrives at time B3.

  Also, as shown in FIG. 4, the time at which the speech recognition start command “BC1” issued by the control unit 303 at time B4 arrives at each of the speech recognition units 302a to 302c is also different.

  At this time, the control unit 303 designates the identifier “BI0” in the command “BC1” as an identifier for specifying the processing target data. For this reason, the voice recognition processing for processing the same voice data is appropriately executed in each of the voice recognition units 302a to 302c.

  Of course, even if the control means 303 issues the voice recognition start command “BC2” at time B5 different from time B4, if the identifier “BI0” is specified by the command “BC2”, the command “BC1” is issued. The voice data that is exactly the same as that issued can be the target of voice recognition processing.

  On the other hand, the data of different sections in a certain voice data may be processed separately by each voice recognition means 302a to 302c.

  Specifically, for example, as shown in FIG. 4, the identifiers added to the audio data input at time B6, time B7, and time B8 are the identifier “BI6”, identifier “BI7”, identifier “BI8”, respectively. ”. Then, the control unit 303 sets the identifier “BI6” in the voice recognition process start command “BC3”, sets the identifier “BI7” in the command “BC4”, and sets the identifier “BI8” in “BC5”. If constituted in this way, each voice recognition means 302a-302c can be made to perform voice recognition processing for the voice data of a different voice section, respectively.

  Various methods can be considered for the recognition result integration processing in the result integration unit 303c. For example, it is possible to take a technique such as rearrangement based on the likelihood of the recognition result in each of the speech recognition units 302a to 302c and re-evaluation using the reliability. In addition, for example, a method of treating the recognition result as a pure character string, a method of evaluating by alignment with the input speech, and the like can be considered.

  As described above, in the second embodiment described above, since the voice data to be subjected to the voice recognition processing is strictly specified by the identifier, the plurality of voice recognition processing units 302a to 302c are configured in parallel. When performing voice recognition processing, it is possible to ensure that each voice recognition processing means 302a to 302c handles the same voice data as specified.

  In the second embodiment described above, the voice data to be subjected to the voice recognition process is strictly specified by the identifier. Therefore, different sections of the voice data in the voice recognition units 302a to 302c are set to the respective voices. When performing the recognition process, it is possible to accurately know the time relationship of the audio data in each processed section. Therefore, it is possible to integrate the plurality of recognition results after completely grasping the time relationship between the plurality of recognition results.

  Furthermore, in the second embodiment described above, since the voice data to be subjected to the voice recognition processing is strictly specified by the identifier, depending on the recognition result of a certain voice recognition means and the progress in the middle of the recognition processing. , Parameters for starting another speech recognition means suitable for the speech to be processed and improving the accuracy of the recognition processing (parameters suitable for the speech to be processed. Specifically, for example, parameters for names, addresses, etc. And the result integration means 303c can integrate them more accurately by examining the identifiers of the audio data handled at that time.

Embodiment 3 FIG.
Next, a third embodiment of the present invention will be described with reference to the drawings.
FIG. 5 is a block diagram illustrating a configuration example of a speech recognition / synthesis system 500 according to the third embodiment of the present invention.

  As shown in FIG. 5, the speech recognition / synthesis system 500 includes a speech input unit 501, a first speech recognition unit 502, a second speech recognition unit 503, and a control unit 504. The voice input means 501, the voice recognition means set 502, the first voice recognition means 502, the second voice recognition means 503, and the control means 504 are connected by a transmission means 505.

  The first voice recognition unit 502 includes a voice recognition control unit 502a, a voice recognition processing unit 502b, a result integration unit 502f, an identifier determination unit 502c, a packet holding unit 502d, and a packet transmission / reception unit 502e.

  The voice recognition control means 502a, the voice recognition processing means 502b, the identifier discrimination means 502c, and the packet holding means 502d are respectively the voice recognition control means 102a, the voice recognition processing means 102b, the identifier discrimination means 102c, and the packet. The same as the holding means 102d.

  The packet transmitting / receiving unit 102e executes processing for receiving the voice data packet from the voice input unit 501, processing for transmitting the voice data packet to the second voice recognition unit 503, and the like.

  The result integration unit 502f executes processing for integrating the recognition result of the first speech recognition unit 502 and the recognition result of the second speech recognition unit 502, and the like.

  The first voice recognition unit 502 receives the voice data from the voice input unit 501, executes voice recognition processing for recognizing the received voice data according to the control of the control unit 504, and transmits the recognition result. .

  In this example, the first speech recognition unit 502 calls the second speech recognition unit 503 at an arbitrary timing of the speech recognition process, receives the processing result of the speech recognition process by the speech recognition unit 503, and integrates the result integration unit 502f. After integrating it with its own recognition result, it is used as the final result.

  The voice data used by the second voice recognition means 503 is the second voice recognition means 503 by transferring the voice data read from the packet holding means 502d inside the first voice recognition means 502 by the packet transmitting / receiving means 502e. Is input. Note that voice data may be directly transmitted from the voice input unit 501 to the second voice recognition unit 503.

  The voice input unit 501, the control unit 504, and the transmission unit 505 are configured in the same manner as the voice input unit 101, the control unit 103, and the transmission unit 104, respectively. The first voice recognition unit 502 is configured by adding a result integration unit 502f to the above-described voice recognition unit 102 and further changing the packet transmission unit 102e to a packet transmission / reception unit 502e. Further, the second voice recognition unit 502 has the same configuration as the voice recognition unit 102 described above.

It is also possible to sandwich third, fourth,... Voice recognition means between the first voice recognition means 502 and the second voice recognition means 503. Further, each voice recognition means 502, 503
Instead of the above, the set 302 of the plurality of speech recognition units 302a to 302n in the second embodiment described above can be used.

Next, the operation of the speech recognition / synthesis system 500 of this example will be described.
FIG. 6 is a time chart showing an example of speech recognition processing by the speech recognition / synthesis system 500 of this example.

  In this example, in the course of performing a certain recognition process by the first voice recognition unit 502, the voice recognition process is executed under different conditions on part or all of the voice data to be processed. In this case, the first speech recognition unit 502 may transfer the speech data to be subjected to speech recognition processing under different conditions to the second speech recognition unit 503.

  FIG. 6 shows an example when the first voice recognition unit 502 causes the second voice recognition unit 503 to process the same voice data. FIG. 6 shows two examples of giving voice data to the second voice recognition means 503.

  In the first example (example shown at time C0 to time C8), the second voice recognition unit 503 directly receives voice data from the voice input unit 501. That is, the voice input unit 501 sequentially transmits the input voice data with an identifier added to the first voice recognition unit 502 and the second voice recognition unit 503.

  At time C0, the control unit 504 issues a command “CC1” including the identifier “CI1” so that the first speech recognition unit 502 recognizes the voice data of the identifier “CI1”.

  At the time C2 at which the command “CC1” is received, the first speech recognition unit 502 starts speech recognition processing and also recognizes speech data with the same identifier “CI1” from the second speech recognition unit 503. The command “CC2” including the identifier “CI1” is issued.

  The second voice recognition unit 503 directly receives the voice data with the identifier “CI1” from the voice input unit 501 and performs voice recognition processing. Then, when the recognition process is completed, the second voice recognition unit 503 transmits a recognition process completion notification “CC2 ′” indicating that to the first voice recognition unit 502.

  Upon receiving the recognition process completion notification “CC2 ′”, the first speech recognition unit 502 integrates its own recognition result with the recognition result of the second speech recognition unit 503 in the result integration unit 502f, and performs final recognition. A recognition process completion notification “CC1 ′” indicating the completion of the process is transmitted to the control means 504.

  As described above, in the first example, the first voice recognition unit 502 starts the voice recognition process in response to receiving the command “CC1” from the control unit 504, and also performs the second voice recognition. A command “CC2” is issued to the means 503. The command “CC1” and the command “CC3” may be commands having the same contents as long as they include the identifier “CI1”, and some of the contents are different (for example, information regarding processing time and processing result response destination). It may be. In this case, the first voice recognition unit 502 may create the command “CC2” by changing the content of the command “CC1”. The same applies to the recognition process completion notifications “CC2 ′” and “CC1 ′”.

  In the second example (example shown at time C9 to time C18), the second speech recognition unit 503 receives speech data to be recognized from the first speech recognition unit 502. That is, the first voice recognition unit 502 receives the voice data to which the identifier is added from the voice input unit 501 and sequentially transfers it to the second voice recognition unit 503.

  The control unit 504 issues a command “CC3” including the identifier “CI10” so that the first speech recognition unit 502 recognizes the speech data of the identifier “CI10” at time C9.

  At the time C11 at which the command “CC3” is received, the first voice recognition unit 502 starts voice recognition processing and transfers the voice data of the identifier “CI10” to the second voice recognition unit 503, and then At time C12, a command “CC4” is issued to start the recognition process for the voice data having the same identifier “CI10” also to the second voice recognition unit 503.

  The second voice recognition unit 503 receives the voice data packet to which the identifier “CI10” is added from the first voice recognition unit 502 and performs voice recognition processing according to the command “CC4”.

  Then, when the voice recognition process is completed, the second voice recognition unit 503 transmits a recognition process completion notification “CC4 ′” indicating that to the first voice recognition unit.

  Upon receiving the recognition process completion notification “CC4 ′”, the first speech recognition unit 502 integrates its own recognition result and the recognition result of the second speech recognition unit 503 in the result integration unit 502f, and performs final recognition. A recognition process completion notification “CC3 ′” indicating the completion of the process is transmitted to the control means 504.

  Note that it is desirable to determine from which component the second speech recognition means 503 receives the speech data depending on the state of the transmission path between the components.

  For example, if the transmission path from the voice input unit 501 to the voice recognition unit is relatively congested while the transmission path between the voice recognition units is relatively free, the voice data is stored in the voice recognition unit. It is better to communicate between them.

  As described above, in the above-described third embodiment, since a certain voice recognition unit calls another voice recognition unit to execute the voice recognition process, it is apparently involved in other components. The speech recognition processing performance can be improved without any. At this time, the voice data to be processed by the second voice recognition unit 503 can be strictly specified by using the control command specifying the identifier.

Embodiment 4 FIG.
Next, a fourth embodiment of the present invention will be described with reference to the drawings.
FIG. 7 is a block diagram showing a configuration example of a speech recognition / synthesis system 700 according to the fourth exemplary embodiment of the present invention.

  As shown in FIG. 7, the speech recognition / synthesis system 700 includes a speech generation unit 701, a speech output unit 702, and a control unit 703. The voice generation unit 701, the voice output unit 702, and the control unit 703 are connected by a transmission unit 704.

  The voice generation unit 701 includes a voice generation control unit 701a, a voice generation processing unit 701b, a packet division unit 701c, an identifier assigning unit 701d, a packet holding unit 701e, and a packet transmission unit 701f.

  The voice generation unit 701 receives a control command from the control unit 703, generates voices, cuts them into packets, assigns identifiers thereto, and transmits them to the voice output unit 702.

  The voice generation control unit 701a receives a control command from another component such as the control unit 703, and controls the entire operation of the voice generation unit 701 based on the received control command. Also, the voice generation control unit 701a transmits information such as the status of voice generation processing in response to a request from another component.

  The voice generation processing unit 701b generates voice data based on control commands from other components. Specifically, processing for synthesizing a speech waveform from a certain character string using speech synthesis technology, processing for reading a speech waveform file designated by a control command, and the like are executed.

  The identifier assigning unit 701d operates in the same manner as the identifier assigning unit 101d included in the speech recognition / synthesis system 100 described above, but further has a function of determining an identifier to be assigned to a packet by a control command from another component. .

  The packet dividing unit 701c, the packet holding unit 701e, and the packet transmitting unit 701f are configured in the same manner as the packetizing unit 101c, the packet holding unit 101e, and the packet transmitting unit 101f included in the speech recognition / synthesis system 100 described above, respectively.

  The audio output unit 702 includes an audio output control unit 702a, an audio output unit 702b, an identifier determination unit 702c, a packet holding unit 702d, and a packet reception unit 702e.

  The voice output unit 702 performs processing for outputting a voice in response to a control command from the control unit 703, processing in response to a request from the control unit 703, and processing for transmitting the processing result to the control unit 703. .

  The control unit 703 and the transmission unit 704 are configured similarly to the control unit 103 and the transmission unit 104 included in the speech recognition / synthesis system 100 described above.

Next, the operation of the speech recognition / synthesis system 700 of this example will be described.
FIG. 8 is a time chart showing an example of speech synthesis processing by the speech recognition / synthesis system 700 of this example.

  In the time chart of FIG. 8, two voice data generated by the voice generation unit 701 based on an instruction from the control unit 703 are transmitted to the user (user terminal) at the time intended by the control unit 703 through the voice output unit 702. An example of the process output toward is shown.

  In FIG. 8, the solid arrows in the vertical direction indicate the passage of time in each component, and the future events as they go down. Also, a solid line arrow in the left-right direction represents the flow of control commands, and a broken line arrow represents the flow of audio data. “User” means a user terminal such as a personal computer or a portable information terminal.

  The control unit 703 issues a command “DC1” at time D0 so as to perform voice generation processing on the voice generation unit 701 and transmit the voice data generated as a result to the voice output unit 702. At this time, the control unit 703 instructs the audio generation unit 701 to add the identifier “DI0” to the first packet of the generated audio data.

  When receiving the command “DC1”, the voice generation unit 701 creates the generated voice by the voice generation processing unit 701b and divides it into packets. The voice generation unit 701 sequentially assigns an identifier “DI1”, an identifier “DI2”,... From the identifier “DI0” to each divided packet. Here, it is assumed that the identifier given to the last packet is the identifier “DI4”.

  When the generation of the audio data corresponding to the command “DC1” is completed, the audio generation unit 701 transmits a generation completion notification “DC1 ′” indicating that to the control unit 703.

  Next, at time D6, the control unit 703 outputs an audio data output command “DC3” to the audio output unit 702 so that audio data to which the identifiers “DI0” to “DI4” are added is sequentially output from the time D11. Is issued.

  Upon receiving the audio data output command “DC3”, the audio output unit 702 receives the audio data to which the identifier (identifier “DI0” to identifier “DI4”) is added from the audio generation unit 701, and packets are received by time D11. The data is held in the holding unit 702d, and thereafter, audio data to which the identifiers “DI0” to “DI4” are added is sequentially output from time D11.

  In this example, as shown in FIG. 8, the control means 703 outputs a control command “DC2” so that another audio data is output following the audio data generated and output according to the control command “DC1”. Shall be issued.

  Specifically, similarly to the control command “DC1”, the control unit 703 issues a command “DC2” at time D3 to the voice generation unit 701, and the first packet of the generated voice data includes An instruction to assign the identifier “DI5” is given.

  When the instruction “DC2” is received, the voice generation unit 701 creates the generated voice by the voice generation processing unit 701b and divides it into packets. The voice generation unit 701 sequentially assigns the identifier “DI6”, the identifier “DI7”,... To the divided packets. Here, it is assumed that the identifier given to the last packet is the identifier “DI10”.

  When the generation of the audio data corresponding to the command “DC2” is completed, the audio generation unit 701 transmits a generation completion notification “DC2 ′” indicating that to the control unit 703.

  Next, at time D8, the control unit 703 issues an audio data output command “DC4” to the audio output unit 702 so as to sequentially output audio data to which the identifiers “DI5” to “DI10” are added. To do. At this time, the audio data output command “DC4” is used to add audio data (identifier “DI0” to identifier “DI4”) that is output before the audio data to be output as the actual start time of audio data output. The time D12 calculated by adding the length of the output time of the sound data to the start time D11 of the sound data).

  The length of the audio data output time may be acquired in advance by some method. Specifically, for example, depending on the definition of the identifier, it can be calculated from the identifier itself. Further, for example, it may be acquired by making an inquiry to the voice generation unit 701. Further, for example, at the time of response “DC1 ′” to the first control command “DC1”, information indicating the output time length of the corresponding audio data may be transmitted at the same time.

  Also, as a method different from the method shown in FIG. 8, immediately after the output of the audio data with the identifier “DI4” is completed, the identifier “DI5” to the identifier “DI10” are output to the audio output unit 702. It is also conceivable to output voice continuously by issuing a command to output the command.

  As described above, in the above-described fourth embodiment, when instructing to actually output audio data, not only the output time but also the identifier of the audio data to be processed is specified. Therefore, it can be ensured that appropriate audio data is output at an appropriate time.

Embodiment 5. FIG.
Next, a fifth embodiment of the present invention will be described with reference to the drawings.
FIG. 9 is a block diagram showing a configuration example of a speech recognition / synthesis system 900 according to the fifth embodiment of the present invention.

  As shown in FIG. 9, the speech recognition / synthesis system 900 includes speech input means 901, speech recognition means 902, speech generation means 903, speech output means 904, and control means 905. The voice input unit 901, the voice recognition unit 902, the voice generation unit 903, the voice output unit 904, and the control unit 905 are connected by a transmission unit 906.

  In addition, it is good also as a structure which is not provided with 1 or 2 or more components among each component except the control means 905. FIG.

  The voice input unit 901, the voice recognition unit 902, the control unit 905, and the transmission unit 906 are the same as the voice input unit 101, the voice recognition unit 102, the control unit 103, and the transmission unit 104 provided in the voice recognition / synthesis system 100 described above, respectively. Configured.

  The voice generation unit 903 and the voice output unit 904 are configured in the same manner as the voice generation unit 701 and the voice output unit 702 provided in the voice recognition / synthesis system 700 described above, respectively. Therefore, the details of the individual operations in each component shown in FIG. 9 are omitted.

  The operation of the speech recognition / synthesis system 900 of this example is a combination of the speech recognition process in the first embodiment and the speech synthesis process in the fourth embodiment.

  Therefore, in the fifth embodiment, both the effects described in the first embodiment and the fourth embodiment described above can be obtained.

  Furthermore, in the fifth embodiment, the barge-in function can be realized by combining the voice input unit 901 and the voice output unit 904. Therefore, compared to the conventional technique, the section of the input voice data to be discarded can be reduced.

  Note that the speech recognition / synthesis system 900 shown in FIG. 9 may be further combined with the configuration described in the second embodiment or the third embodiment. With such a configuration, the effects described in the second embodiment and the third embodiment described above can be enjoyed at the same time.

  As described above, in each of the above-described embodiments, since the processing is performed by adding an identifier uniquely identified in the system to the voice data packet, the voice data and voice recognition control or voice synthesis control are processed. Information that is essentially independent of the control command can be appropriately synchronously controlled, and each component constituting the speech recognition / synthesis system can be appropriately coordinated.

  That is, it is not a configuration as disclosed in Patent Document 1 or Patent Document 2 in which the data transmission source sets the beginning of each data to be synchronized as a reference point, and the voice data packet is uniquely assigned in the system. Since it is configured to perform processing by adding an identified identifier, any component in the system can freely set any synchronization timing without being limited to any component that can set synchronization timing. Thus, flexible processing can be performed.

  Further, in each of the above-described embodiments, an identifier that is uniquely identified in the system is added to the voice data packet, and a control command in which the identifier is set is configured so that any component can be freely synchronized. By setting the timing, it is possible to synchronize essentially independent information between the speech data and the control command for speech recognition control or speech synthesis control. Thus, individual components can operate more intelligently.

  Further, in each of the above-described embodiments, an identifier uniquely identified in the system is added to the voice data packet, and a control command in which the identifier is set is executed, so that each voice recognition / speech synthesis process is performed. Therefore, it is possible to easily and strictly specify the voice data to be controlled, and to prevent deterioration of voice recognition accuracy and synthesized voice quality.

  Further, in each of the above-described embodiments, since the section of the voice data to be subjected to the voice recognition process is strictly specified by the identifier, it is possible to avoid the lack of the input voice, particularly the beginning part of the utterance. And a reduction in voice recognition accuracy can be prevented. Moreover, if noise mixing is minimized, erroneous detection of the speech section can be suppressed.

  When the voice data transmission system and the command transmission system are independent as in each of the embodiments described above, at which timing the voice data section corresponding to the command instructing the start of the voice recognition process is It is uncertain whether it will arrive at the means. In each of the above-described embodiments, it is possible to specify an appropriate voice data section that is a processing target without knowing the arrival timing of the voice data, and to appropriately perform recognition processing.

  Further, in each of the above-described embodiments, even in a situation where a command for instructing the start of voice recognition processing is frequently issued, the voice data that is the target of each command can be identified without confusion. Since the voice data section to be subjected to the voice recognition process can be strictly specified, it can be ensured that the voice recognition process using the parameter setting suitable for the voice to be processed is appropriately performed.

  That is, as in the above-described embodiments, since the audio data and the control command pass through different transmission paths, the order relationship is not guaranteed at all. In each of the above-described embodiments, even if the voice data arrives at the voice recognition component in an order different from the order in which the component that issued the control command is intended, the appropriate voice data section to be processed is ensured. And the recognition process can be appropriately performed. Therefore, for example, when the utterance of the first and last name and the utterance of the telephone number are made consecutively, the parameter setting for the telephone number is used for the former, and the parameter setting for the first and last name is used for the latter. This is prevented.

  In each of the above-described embodiments, if the management of identifiers to be added is devised (for example, the combination of voice data and added identifiers or the input voice data is stored and held), any arbitrary It can be set as the structure which matches the audio | voice data of another arbitrary time with the control command issued at the time. With this configuration, even voice data input at an arbitrary time in the past or in the future from the time at which the control command is issued can be commanded by the control command. As described above, since it is possible to perform a command by the control command for voice data input at an arbitrary time in the past or in the future from the time when the control command is issued, The apparent idle time can be minimized. In practice, it is necessary to implement some contrivance (buffering of past data and scheduling of future control) so that the designated audio data can be reliably obtained.

  In each of the above-described embodiments, only the example in which the time for executing the voice output is set as the control command (for example, the control command “DC3”) has been described. However, the time for executing the voice recognition processing and the voice synthesis processing is described. You may make it set to a control command. In this case, in accordance with the control command, at the time set in the control command, the voice recognition process and the voice synthesis process are executed for the voice data packet set in the control command.

  Although not specifically mentioned in each of the above-described embodiments, each process executed in the system 100, 300, 500, 700, 900 is a control program (synchronous control program) installed in the system 100 or the like. ) Is executed. This control program is, for example, a synchronization control program that causes a speech recognition / synthesis system that performs speech recognition processing to analyze input speech data or / and speech synthesis processing to generate speech data to perform synchronization control. A step of issuing a control command in which identification information for specifying the audio division data to be processed among the audio division data obtained by dividing the audio data into a plurality of sections is set to a computer constituting the synthesis system; A program for executing a step of performing speech recognition processing and / or speech synthesis processing on speech divided data specified by identification information set in the control command according to the command.

Next, specific examples of the present invention will be described.
The example described below corresponds to the fifth embodiment described above.

  FIG. 10 is an explanatory diagram showing a speech recognition / synthesis system 1000 according to the present embodiment. As shown in FIG. 10, the speech recognition / synthesis system 1000 includes a speech dialogue management server 1001, an input terminal device 1002, a speech input / output server 1003, a speech recognition server 1004, and a speech synthesis server 1005.

  The voice interaction management server 1001, the input terminal device 1002, the voice input / output server 1003, the voice recognition server 1004, and the voice synthesis server 1005 are each connected by a computer network 1006.

  Any number of computers 1002, 1003, 1004, and 1005 other than the voice interaction management server 1001 can be connected. One device can also serve as two or more arbitrary devices. For example, a single computer can serve as both the voice interaction management server 1001 and the voice input / output server 1003.

  Alternatively, one computer can serve as all the components. Furthermore, as in the third embodiment described above, a single speech recognition server 1004 or speech synthesis server 1005 may serve as a proxy server that calls another speech recognition server 1004 or speech synthesis server 1005. Good.

  The voice interaction management server 1001 is a control device that controls the entire system, and is a computer on which a program having a function corresponding to the control unit 905 (see FIG. 9) operates. It also has a function of performing load distribution processing by selecting one from a plurality of speech recognition servers or speech synthesis servers and mediating to a speech input / output server.

  The input / output terminal device 1002 is an input / output device directly used by a user, and can include a display device, a keyboard, a mouse, a touch panel, and the like in addition to a voice input / output function.

  Specifically, as the input / output terminal device 1002, a PC (personal computer), a telephone (fixed phone, a mobile phone), a PDA (Personal Digital Assistants), a network-compatible home appliance, or the like is used.

  The input / output terminal device 1002 also serves as a part of the functions of the control unit 905, the voice input unit 901, and the voice output unit 904 provided in the voice recognition / synthesis system 900.

  The voice input / output server 1003 is a server device having a function of dividing voice data input by the input / output terminal device 1002 into voice packets, assigning identifiers, and transmitting them to each component.

  Conversely, the voice input / output server 1003 also has a function of combining packets received from other components and sending them to the voice output device 1002.

  That is, the voice input / output server 1003 has a part of each function of the voice input unit 901 and the voice output unit 904 provided in the voice recognition / synthesis system 900. In this embodiment, a time stamp is used as an identifier of audio data.

  The voice recognition server 1004 is a server device that performs voice recognition processing on voice data obtained from the voice input / output server 1003 and executes processing for transmitting the result to the voice dialogue management server 1001. The voice recognition server 1004 corresponds to the voice recognition unit 902 included in the voice recognition / synthesis system 900.

  The voice synthesis server 1005 is a server device having a function of synthesizing voice data in accordance with an instruction from the voice dialogue management server 1001 and transmitting the result to the voice input / output server 1003.

  Note that the speech synthesis process need not be performed each time, and the synthesized result may be cached in advance and the cached synthesized speech may be used. In addition to the synthesized speech, the speech data may be generated by reproducing an arbitrary waveform file.

  The voice synthesis server 1005 corresponds to the voice generation unit 903 provided in the voice recognition / synthesis system 900.

  The computer network 1006 is, for example, a generally used LAN, but a network with a large transmission delay, such as a wireless network, a telephone line network, or a WAN, can also be used.

  By combining the input / output terminal device 1002 and the voice input / output server 1003, the functions of the voice input unit 901 and the voice output unit 904 provided in the voice recognition / synthesis system 900 are realized.

  The components of the speech recognition / synthesis system 1000 according to the present embodiment are divided as described above, so that various existing input / output terminal devices can be used in the interactive system 1000. This is because the voice input / output server 1003 absorbs the difference in voice input / output by the output terminal device.

  Therefore, in FIG. 10, each input / output terminal device is connected to a different voice input / output server. However, the example shown in FIG. 10 does not always indicate that the input / output terminal device and the voice input / output server have a one-to-one relationship. If there are a plurality of input / output terminal devices corresponding to a certain voice input / output server, a plurality of types of input / output terminal devices may be connected to one voice input / output server.

  Note that the operation of each unit in the speech recognition / synthesis system 1000 is the same as the operation of each unit in the speech recognition / synthesis system 900 in the fifth embodiment described above, and a detailed description thereof will be omitted.

  According to the present invention, in addition to a voice dialogue system such as an automatic voice response device, it is applied to various uses such as a voice remote controller, a voice internet browsing device, a voice user interface for a disabled person, or a voice dialogue function of a robot. Useful for.

  Further, according to the present invention, it is also possible to apply the present invention to strictly handling time-series data other than voice such as moving images and stock prices.

It is a block diagram which shows the structural example of the speech recognition / synthesis system in the 1st Embodiment of this invention. It is a time chart which shows an example of operation | movement of the speech recognition / synthesis system in the 1st Embodiment of this invention. It is a block diagram which shows the structural example of the speech recognition / synthesis system in the 2nd Embodiment of this invention. It is a time chart which shows an example of operation | movement of the speech recognition / synthesis system in the 2nd Embodiment of this invention. It is a block diagram which shows the structural example of the speech recognition / synthesis system in the 3rd Embodiment of this invention. It is a time chart which shows an example of operation | movement of the speech recognition / synthesis system in the 3rd Embodiment of this invention. It is a block diagram which shows the structural example of the speech recognition / synthesis system in the 4th Embodiment of this invention. It is a time chart which shows an example of operation | movement of the speech recognition / synthesis system in the 4th Embodiment of this invention. It is a block diagram which shows the structural example of the speech recognition / synthesis system in the 5th Embodiment of this invention. It is a block diagram which shows the structure of the speech recognition / synthesis system in the Example of this invention.

Explanation of symbols

100, 300, 500, 700, 900, 1000 Speech recognition / synthesis system 101, 301, 501 Speech input means 102, 302, 302a, 302b, 302n Speech recognition means 103, 303, 504, 703 Control means 104, 304, 505 , 704 Transmission means 101a Voice input control means 101b Voice input processing means 101c, 701c Packet division means 101d, 701d Identifier assignment means 101e, 102d, 502d, 701e, 702d Packet holding means 101f, 701f Packet transmission means 102a, 502a Voice recognition control Means 102b, 502b Voice recognition processing means 102c, 502c, 702c, 904b Identifier discrimination means 102e, 502e, 702e Packet receiving means 103a, 303a User interface 103b, 303b Dialog management means 303c, 502f Result integration means 502 First speech recognition means 503 Second speech recognition means 701 Speech generation means 702 Speech output means 701a Speech generation control means 701b Speech generation processing means 702a Speech output Control unit 702b Speech output processing unit 1001 Spoken dialogue management server 1002 Input / output terminal device 1003 Speech input / output server 1004 Speech recognition server 1005 Speech synthesis server

Claims (23)

A speech recognition / synthesis system for performing speech recognition processing for analyzing input speech data and / or speech synthesis processing for generating speech data,
Control command means for issuing a control command in which identification information for specifying the voice divided data to be processed is specified among the respective voice divided data obtained by dividing the voice data into a plurality of sections;
Voice processing means for performing voice recognition processing or / and voice synthesis processing on voice divided data specified by the identification information set in the control command in accordance with the control command from the control command means A speech recognition / synthesis system characterized by this. The speech recognition / synthesis system according to claim 1, further comprising identification information adding means for adding identification information uniquely identified in the system to each divided voice data obtained by dividing the inputted voice data into a plurality of sections. Voice input processing means for performing voice data input processing;
The voice recognition / synthesis system according to claim 1, further comprising: voice data dividing means for generating voice divided data obtained by dividing the voice data input by the voice input processing means into a plurality of sections. The control command means issues a control command that sets the execution time of the voice recognition process or / and the voice synthesis process,
In accordance with the control command from the control command unit, the voice processing unit converts the voice division data specified by the identification information set in the control command when the execution time set in the control command is reached. The speech recognition / synthesis system according to any one of claims 1 to 3, wherein speech recognition processing and / or speech synthesis processing is performed on the speech recognition processing. With a plurality of voice processing means,
The speech recognition / synthesis system according to any one of claims 1 to 4, further comprising a processing result integration unit that integrates processing results of the speech recognition processing and / or speech synthesis processing of each of the plurality of speech processing units. The control command means issues a control command to one of the plurality of voice processing means,
6. The voice recognition / synthesis system according to claim 5, wherein the one voice processing means has a control command transfer means for transferring a part or all of the control command from the control command means to another voice processing means. 7. The voice processing means according to claim 6, further comprising: voice data transfer means for transferring one or all sections of processing target voice data instructed by a control command from the control command means to another voice processing means. Recognition / synthesis system. The identification information adding means adds, as identification information, a time stamp, a serial number, a voice interaction processing sequence number by voice recognition processing or / and voice synthesis processing, or a combination thereof to each piece of voice divided data. The speech recognition / synthesis system according to claim 7. The speech recognition / synthesis system according to any one of claims 1 to 8, further comprising an identification information management unit that provides a function of managing the time order of identification information. The identification information management means manages the time order of each identification information by synchronizing the absolute time respectively used by each component constituting the system and making the identification information correspond to a specific absolute time. 9. The speech recognition / synthesis system according to 9. A synchronization control method in a speech recognition / synthesis system for performing speech recognition processing for analyzing input speech data and / or speech synthesis processing for generating speech data,
Issuing a control command that sets identification information for specifying the audio division data to be processed among the audio division data obtained by dividing the audio data into a plurality of sections,
According to the control command, a voice recognition process or / and a voice synthesis process are performed on the voice divided data specified by the identification information set in the control command. The synchronization control method according to claim 11, wherein identification information uniquely identified in the system is added to each voice division data obtained by dividing the inputted voice data into a plurality of sections. Perform audio data input processing,
The synchronization control method according to claim 11 or 12, wherein voice divided data is generated by dividing the voice data input by the input process into a plurality of sections. Issue a control command that sets the execution time of speech recognition processing and / or speech synthesis processing,
In accordance with the control command, when the execution time set in the control command is reached, voice recognition processing or / and speech synthesis processing is performed on the voice divided data specified by the identification information set in the control command. The synchronization control method according to any one of claims 11 to 13. The synchronous control method according to any one of claims 11 to 14, wherein the processing results of a plurality of speech recognition processes and / or speech synthesis processes performed by different processing means according to a control command are integrated. After performing speech recognition processing or / and speech synthesis processing according to the control command, a part or all of the control command is transferred to other processing means,
The synchronous control method according to claim 15, wherein speech recognition processing and / or speech synthesis processing is performed by the other processing means in accordance with the transferred control command. The synchronous control method according to claim 16, wherein one or all sections of the audio data to be processed instructed by the control command from the control command means are transferred to another processing means. The time stamp, serial number, speech recognition processing sequence number and / or speech synthesis processing sequence number by speech synthesis processing, or a combination thereof is used as the identification information added to each voice division data. The synchronization control method according to any one of 17. A synchronization control program that causes a speech recognition / synthesis system that performs speech recognition processing to analyze input speech data and / or speech synthesis processing to generate speech data to perform synchronization control,
In a computer constituting the speech recognition / synthesis system,
Issuing a control command that sets identification information for identifying the audio division data to be processed among the audio division data obtained by dividing the audio data into a plurality of sections;
A synchronization control program for executing, in accordance with the control command, a step of performing speech recognition processing and / or speech synthesis processing on the speech divided data specified by the identification information set in the control command. On the computer,
The synchronization control program according to claim 19, further comprising the step of adding identification information uniquely identified in the system to each piece of voice divided data obtained by dividing the inputted voice data into a plurality of sections. On the computer,
Furthermore, a step of performing voice data input processing;
21. The synchronization control program according to claim 19 or 20, further comprising: generating audio division data obtained by dividing the audio data input by the input processing into a plurality of sections. The time stamp, serial number, speech recognition processing sequence number and / or speech synthesis processing sequence number by speech synthesis processing, or a combination thereof is used as identification information added to each voice division data. 21. The synchronization control program according to any one of 21. Voice input processing means for performing voice data input processing;
Voice data dividing means for generating voice divided data obtained by dividing the voice data input by the voice input processing means into a plurality of sections;
Identification information adding means for adding identification information uniquely identified in the system to each voice divided data divided by the voice data dividing means;
Identification information for specifying speech-division data to be processed for speech processing means for performing speech recognition processing for analyzing speech data input by the speech input processing means and / or speech synthesis processing for generating speech data And a control command means for issuing a control command in which the control command is set.

Images