JP2005308950A - Speech processors and speech processing system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech processor and a speech processing system which record contents of calls among two or more calling devices for each speaker, and is enabled to mutually share the recorded contents of calls among the calling devices. <P>SOLUTION: A speech processor PC_A receives from a calling device PH_A speech data in which two or more speakers' speech performed via a private branch exchange (PBX) is mixed. A voiceprint authentication part 20 sequentially performs voiceprint authentication of the inputted speech data in each voiceprint authentication unit, and a CPU 10 arranges and stores speech data identified by the voiceprint authentication in order on memory 40 for each utterer. Further, the data of each utterer is shared on the network via a communication interface 50. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a voice processing apparatus and a voice processing system for recognizing and processing voice commands issued by, for example, a plurality of conference attendees.

Conventionally, in a group work such as a meeting, the contents of the proceedings (audio data) are recorded on a recording medium such as a recorder, and after the meeting, the recording is performed while the speaker is confirmed by playing the recording medium. It is done to write minutes to do.
At this time, various techniques are known for recording audio data on a recording medium.

By the way, it is not an appropriate method to confirm the speaker and record the agenda contents after the conference as in the past from the viewpoint of the efficiency of the work contents and the accuracy of the speaker identification.
In other words, recording the agenda content after the meeting, not in real time, not only does it not be possible to do other work that was originally possible after the meeting, but also takes time to organize the audio data by speaker. It is not efficient because it is complicated.
In addition, it may be difficult to identify a speaker based on sound reproduced from a recording medium without error.

  The present invention has been made in view of such circumstances, and an object of the present invention is to record the content of a call between a plurality of call devices for each speaker, and to perform voice processing capable of sharing the recorded call content between the call devices. It is to provide an apparatus and a voice processing system.

  In order to achieve the above object, a first aspect of the present invention is a speech processing apparatus that classifies input speech for each speaker, and includes one or a plurality of voiceprint data and speaker data corresponding to the speaker. First and second storage means for storing them in association with each other, voice print data is extracted from the input voice every predetermined time, and the voice print data is collated with the voice print data stored in the first storage means. And voice control means for specifying speaker data, and control means for storing the input speech for each predetermined time in the second storage means for each speaker data specified by the voice print authentication means. .

  In order to achieve the above object, a second aspect of the present invention is a voice processing system that classifies voices between a plurality of communication devices connected to a circuit network for each speaker. Alternatively, a first storage unit that stores a plurality of voiceprint data in association with the speaker data corresponding to the speaker, a second storage unit, and voiceprint data is extracted from the voice every predetermined time, and the voiceprint data is extracted. For each speaker data specified by the voiceprint authentication means, voiceprint authentication means for specifying speaker data by collating data with voiceprint data stored in the first storage means, and voice for each predetermined time period. And a control means for storing in the second storage means, each of which is configured to be able to transmit and receive data by communication with each other.

  Preferably, each call device includes voice recognition means for recognizing a voice between a plurality of call devices and converting it into a character string, and the character string obtained by converting the voice by the voice recognition means is a predetermined command. In addition, the control means executes a process defined in advance by the predetermined command on the voice processing device corresponding to the speaker data specified by the voiceprint authentication means based on the voice.

  Preferably, the predetermined command is any of speaker data stored in the second storage unit, and the predetermined process is performed on the speaker data stored in the second storage unit. The corresponding voice data is transmitted to the voice processing device corresponding to the voice speaker identified by the voiceprint authentication means.

  Preferably, permitted speakers who are permitted to execute the predetermined command are limited in advance, a character string obtained by converting the speech by the speech recognition means is a predetermined command, and voiceprint authentication is performed based on the speech. When the speaker data specified by the means is the speaker data corresponding to any of the permitted speakers, the control means executes a process defined in advance by the predetermined command.

  According to the speech processing apparatus according to the first aspect of the present invention, the voiceprint authentication unit extracts voiceprint data from the input speech every predetermined time, and the voiceprint data stored in the first storage unit is stored in the voiceprint data. The speaker data is specified by collation, and the control means stores the input speech for every predetermined time in the second storage means for each speaker data specified by the voiceprint authentication means. Can automatically and accurately organize voice data for each speaker without any special operation.

  According to the speech processing system according to the second aspect of the present invention, in the speech processing device corresponding to each call device, the speech print authenticating unit extracts the speech print data from the input speech every predetermined time, and the voice print data is The voice data stored in the first storage means is collated to specify speaker data, and the control means determines the input speech for each predetermined time for each speaker data specified by the voice print authentication means. Since the input voice can be transmitted / received to / from another voice processing apparatus or from another voice processing apparatus, the voice data for each speaker can be shared among the talkers.

  According to the present invention, the content of a call between a plurality of call devices is recorded for each speaker, and the recorded call content can be shared between the call devices, so that the speaker does not perform any special operation. The voice data can be automatically and accurately organized for each speaker, and the voice data for each speaker can be shared among the speakers.

First Embodiment Hereinafter, a first embodiment of a voice processing apparatus and a voice processing system according to the present invention will be described.
FIG. 1 illustrates an application example of the speech processing system according to the present embodiment. The voice processing system shown in FIG. 1 includes a plurality of call devices PH_A, PH_B, PH_C, PH_D, and PH_E, and sound processing devices PC_A, PC_B, PC_C, PC_D, and PC_E connected to the call devices, respectively. Is done.

A digital line network using a private branch exchange PBX is constructed between the communication devices. Therefore, it is possible to hold a conference call among a plurality of people of three or more people.
Also, the voice processing devices PC_A, PC_B, PC_C, PC_D, and PC_E are connected to a wireless or wired communication network, and are configured to be capable of data communication with each other.

FIG. 2 is a diagram illustrating the configuration of each call device and each voice processing device, taking the call device PH_A and the voice processing device PC_A as examples.
The REC_OUT terminal of the call device PH_A and the LINE_IN terminal of the voice processing device PC_A are connected to each other, and voice digital data is transmitted from the call device PH_A to the voice processing device PC_A.

  The voice processing device PC_A includes a CPU 10 as a control unit according to the present invention, a voiceprint authentication unit (VPA) 20 as a voiceprint authentication unit according to the present invention, and a voiceprint register (REG) 30 as a first storage unit according to the present invention. A memory (MEM) 40 as a second storage means of the present invention and a communication interface (I / F) 50 are included.

The voiceprint authentication unit 20 performs voiceprint authentication of voice data supplied from the CPU 10 every predetermined time (for example, 3 seconds) to identify a speaker. As the voiceprint authentication technique, a known technique can be applied to the voiceprint authentication unit 20.
Note that the unit time (for example, 3 seconds) of the voiceprint authentication process is preferably as short as possible from the viewpoint of the person identification accuracy if the processing capabilities of the voiceprint authentication unit 20 and the CPU 10 allow.

Specifically, first, the voiceprint authentication unit 20 generates a power spectrum (voiceprint data) of voice data supplied from the CPU 10.
In the voiceprint register 30, the speaker data (for example, the ID of the speaker) and the voiceprint data of the speaker are registered and recorded in association with each other.
The voiceprint authentication unit 20 compares the generated voiceprint data with the voiceprint data recorded in the voiceprint register 30 in order, and the possibility that both voiceprint data are approximated based on a predetermined evaluation criterion is equal to or higher than a predetermined level. In some cases, the voice data supplied from the CPU 10 is identified as the voice of the corresponding speaker.

  In general, the difference in the voiceprint of each person is determined from the difference in the volume and structure of the mouth and nostrils resulting from the face shape of the person, and the difference in the vocal cords resulting from the height and sex, so that the accuracy of identification is high. For example, even if the caller has a voice or a nose due to a cold or the like, the strength and frequency of the voiceprint waveform are not changed, and the authentication accuracy is not affected.

In the memory 40, voice data is organized and stored for each data storage area having the file name of the speaker data (for example, the speaker ID). For example, when Mr. A's speaker ID is ID_A, Mr. A's voice data (voice data) is recorded in the data storage area with the file name ID_A.
The communication interface 50 includes an interface circuit that complies with a predetermined communication standard in order to perform wireless or wired data communication with other audio processing devices (PC_B, PC_C, etc.).

The CPU 10 performs overall control of the voice processing device PC_A.
For example, access control to the voiceprint register 30 and the memory 40, input processing for voice data input from the LINE_IN terminal, timing control for voiceprint processing of the voiceprint authentication unit 20, communication control with other voice processing devices via the communication interface 50, etc. It is.
In particular, the CPU 10 uses the voice data input from the LINE_IN terminal for each unit time (for example, 3 seconds) in which the voiceprint authentication unit 20 performs voiceprint processing, and the data in the memory 40 corresponding to the speaker data specified by the voiceprint processing. Store sequentially in the storage area.

Next, processing operations performed by the voice processing device PC_A will be described.
In the description below, Mr. A uses the call device PH_A, Mr. B uses the call device PH_B, and Mr. C uses the call device PH_C. The device PC_A will be described. The processing performed in the other communication devices (PH_B, PH_C) and the other voice processing devices (PC_B, PH_C) is the same as the processing content described below.

Since the voices of Mr. A, Mr. B and Mr. C during the call are mixed by the private branch exchange PBX shown in FIG. 1, not only the voice data of Mr. A but also the voice data in which all the voice data of the three parties are mixed. And sent from the REC_OUT terminal of the telephone device PH_A to the LINE_IN terminal of the voice processing device PC_A.
The voice processing device PC_A processes the voice data input from the LINE_IN terminal at a time interval that allows voiceprint authentication.

3A to 3G show processing operations when the voiceprint authentication unit 20 processes voice data based on instructions from the CPU 10 when Mr. A, Mr. B, and Mr. C speak in time order. FIG.
First, in FIG. 3A, an upward arrow means the start of voiceprint authentication, and a downward arrow means the end of voiceprint authentication. Here, the start and end of voiceprint authentication is 3 seconds, and the voiceprint authentication unit 20 performs voiceprint authentication processing in units of 3 seconds and generates voiceprint data.

FIG. 3B shows actual utterance timings of Mr. A, Mr. B, and Mr. C. As shown in the figure, depending on the start timing of the utterance, for example, the last part of the utterance may not reach the voiceprint authentication unit time.
In response to the utterance, the voiceprint authentication unit 20 generates voiceprint data based on the acquired utterance data every three seconds as a voiceprint authentication unit as shown in FIG. Store. In the figure, for example, the data immediately before the end of the utterance of Mr. A's utterance data is less than 3 seconds required for the voiceprint process, and therefore the voiceprint process cannot be performed.

Next, as shown in FIG. 3D, the voiceprint authentication unit 20 stores the voiceprint data stored in the buffer in the voiceprint register 30 in advance for each voiceprint authentication unit (voiceprint data for 3 seconds). Compare and collate with registered voiceprint data. That is, when the voiceprint data generated every 3 seconds and the voiceprint data stored in the voiceprint register 30 are sequentially compared and matched (or the possibility of matching is equal to or higher than a predetermined level), they match. The speaker is specified by the speaker data corresponding to the registered voiceprint data.
In the figure, the portion of the utterance data of Mr. A that is data immediately before the end of the utterance and that cannot be subjected to the voice print process is indicated by NA.

  3D corresponds to the timing stored in the buffer in FIG. 3C, but in reality, the processing time for voiceprint authentication is required. The specific timing is delayed from the timing of storing in the buffer.

FIG. 3E is a diagram showing the utterance data of Mr. A who has finished voiceprint authentication.
The CPU 10 sequentially stores the voice data input from the LINE_IN terminal for each voiceprint authentication unit in the data recording area of the file name on the memory 40 corresponding to the speaker data specified by the voiceprint authentication unit 20. For example, Mr. A's voice is stored in a file ID_A.
At that time, as shown in the figure, the voice data corresponding to the last voiceprint unauthenticated portion (NA) of Mr. A's utterance is estimated as Mr. A's voice data by checking the continuity of the voice data. Then, the audio data of the part may be stored in the file (ID_A) of the memory 40.

Next, the same applies to the case where the speaker changes and Mr. B and Mr. C speak in order. That is, as shown in FIGS. 3 (F) and 3 (G), the CPU 10 sequentially converts the voice data for each voiceprint authentication unit, for example, the voice data of Mr. B and Mr. C are called ID_B and ID_C on the memory 40, respectively. Store in the file name data storage area.
At that time, like Mr. A's voice data, there is a portion (NA) where voiceprint authentication is impossible before and after the utterance.

  As described above, according to the voice processing device according to the present embodiment, voice data obtained by mixing voices of a plurality of speakers performed via the private branch exchange PBX is input, and the voiceprint authentication unit 20 performs voiceprint authentication units. Since the voice print authentication is performed sequentially for each time, and the CPU 10 organizes and stores the input voice data on the memory 40 for each speaker data specified by the voice print authentication, the following effects are obtained.

That is, in a conference call with multiple persons, the content of the speech is organized in real time for each speaker and stored in the memory, so the user of the voice processing device records the content of the conference when creating the minutes. There is no need to specify the speaker by playing the media.
In addition, since the voice data is organized in the memory for each speaker, the minutes can be edited very easily.
Furthermore, specifying a speaker by playing a recording medium in which the contents of a conference call or the like are reproduced is complicated and sometimes difficult in terms of listening accuracy. According to the processing device, since the person is identified by voiceprint authentication, the speaker can be identified with extremely high accuracy.
Further, according to the voice processing system according to the present embodiment, since each voice processing apparatus has a communication means, it is possible to share voice data for each speaker on the network.

Second Embodiment Hereinafter, a second embodiment of the speech processing apparatus and speech processing system according to the present invention will be described.
The speech processing apparatus according to the present embodiment is characterized by recognizing a command by a voice from the outside and performing a predetermined process corresponding to the recognized command.
The system configuration of the speech processing system according to the present embodiment is the same as the configuration shown in FIG. FIG. 4 shows an example of the configuration of the speech processing apparatus according to the present embodiment configured in the speech processing system shown in FIG. 1, using the speech processing apparatus PC_A as an example.

As shown in FIG. 4, the speech processing apparatus PC_A according to the present embodiment is different from the speech processing apparatus PC_A according to the first embodiment shown in FIG. 2 in that a speech recognition unit 60 is added. .
The voice recognition unit 60 converts voice data input from the LINE_IN terminal into character string data using a voice recognition technique, and supplies it to the CPU 10.

The speech recognition unit 60 includes an acoustic model and a recognition dictionary. The acoustic model includes a basic sound unit used for speech recognition, that is, a unit of small human pronunciation (phoneme) such as a consonant or a vowel. Acoustic features are described. This acoustic feature is statistical acoustic feature information of phonemes obtained from the voices of many speakers.
In the recognition dictionary, character string data for voice recognition is described.

Specifically, the voice recognition unit 60 performs the following processing.
(1) Calculation of acoustic features The input voice data is subjected to spectrum analysis, and acoustic features (acoustic features) are extracted.
(2) Output of speech recognition result The feature amount of the input speech data is collated with the feature amount described in the acoustic model, and the character string data closest to the input speech is recognized as the speech recognition result. Output. At that time, the character string data whose acoustic feature is closest to the acoustic feature of the input speech is searched from the character string data described in the recognition dictionary, and is output as a speech recognition result.
In addition, in order to improve the voice recognition rate by the voice recognition unit 60, it is possible to recognize a registrant's voice in a particularly accurate manner by registering a word to be recognized in advance with a voice of a specific person (a specific speaker voice). recognition).

The CPU 10 according to the present embodiment performs processing to be executed in response to a plurality of registered commands (character string data), and the CPU 10 has character string data that is a result of speech recognition processing by the speech recognition unit 60 as a result of a plurality of commands. If it is one of them, the process corresponding to the command is executed.
For example, the operation in the case where “name of Mr. B” is registered in the CPU 10 as a command and “transmission processing of Mr. B's voice data” is defined in advance as the processing content corresponding to the command will be described.

In the speech processing apparatus PC_A to which speech data corresponding to “name of Mr. B” (for example, Yamada) is input from the LINE_IN terminal, speech recognition processing by the speech recognition unit 60 is first performed. As a result, the voice recognition unit 60 recognizes that the input voice data is “Mr. B's name”, and sends “Mr. B's name” as character string data to the CPU 10.
In the CPU 10, “Mr. B's name” is registered as a plurality of commands, and as a corresponding processing content, “ID_B (speaker data) file on the memory 40 is transmitted to the input voice speaker. It is assumed that the processing content “Yes” is defined.

  In that case, the CPU 10 instructs the voiceprint authentication unit 20 to perform voiceprint authentication based on the input voice data. Then, the voiceprint authentication unit 20 specifies that the speaker data is ID_B, and the CPU 10 transmits the voice data in the file ID_B on the memory 40 to the voice processing device PC_B via the communication interface 50.

As described above, according to the voice processing system according to the present embodiment, voice data for each speaker recorded in a conference call or the like by a plurality of people can be acquired via a network by a simple command by voice. It becomes possible.
Therefore, the voice data stored in the remote voice processing apparatus can be acquired without performing a complicated operation on a personal computer or the like, which is highly convenient.

Third Embodiment Hereinafter, a voice processing device and a voice processing system according to a third embodiment will be described.
According to the voice processing system according to the second embodiment described above, a plurality of voice processing apparatuses are connected to a network, and voice data stored in a remote voice processing apparatus can be acquired via the network. It may be a security problem to allow any person connected to the network to access the audio data.
Therefore, in the speech processing system according to the present embodiment, each speech processing device permits the processing of commands by speech via the corresponding call device only when the speech is the speech of a previously registered speaker. Do it.

The audio processing device according to the present embodiment has the same configuration as the audio processing device according to the second embodiment shown in FIG. 4, but the processing performed by the CPU 10 is different.
Hereinafter, the same example as that of the description of the second embodiment, that is, “Mr. B's name” is registered in the CPU 10 as a command. The operation of the speech processing apparatus according to this embodiment when “is defined in advance” will be described.

In the speech processing apparatus PC_A to which speech data corresponding to “name of Mr. B” (for example, Yamada) is input from the LINE_IN terminal, speech recognition processing by the speech recognition unit 60 is first performed. As a result, the voice recognition unit 60 recognizes that the input voice data is “Mr. B's name”, and sends “Mr. B's name” as character string data to the CPU 10.
In the CPU 10, “Mr. B's name” is registered as a plurality of commands, and as a corresponding processing content, “ID_B (speaker data) file on the memory 40 is transmitted to the input voice speaker. It is assumed that the processing content “Yes” is defined.

In that case, the CPU 10 instructs the voiceprint authentication unit 20 to perform voiceprint authentication based on the input voice data, and as a result, the voiceprint authentication unit 20 specifies that the speaker data is ID_B.
The CPU 10 has a list of speaker data that permits command processing in advance, and confirms whether the speaker data specified by the voiceprint authentication unit 20 is included in the list.
As a result, when the speaker data specified by the voiceprint authentication unit 20 is included in the list, the CPU 10 is in the file ID_B on the memory 40 as in the case of the speech processing apparatus according to the second embodiment. The audio data is transmitted to the audio processing device PC_B via the communication interface 50.
On the contrary, when the speaker data specified by the voiceprint authentication unit 20 is not included in the list, the CPU 10 does not perform any processing.

As described above, according to the voice processing system according to the present embodiment, when voice data for each speaker recorded in a conference call or the like by a plurality of people is acquired via a network by a voice command, a voiceprint is obtained. Only when the speaker specified by the authentication is registered in advance, the process corresponding to the command is performed. When the speaker is not registered, the process corresponding to the command is not performed.
Therefore, safety can be improved when voice data is shared on the network. For example, access to each speaker data in a conference call can be limited to people associated with the conference agenda.

The contents of the embodiment are not limited to the contents described above, and various modifications can be made.
For example, it goes without saying that the list of permitted speaker data can be easily changed according to the importance (safety level) of the command.

1 is a configuration of an embodiment of a voice processing system according to the present invention. It is a structure of one Embodiment of the audio processing apparatus which concerns on this invention. It is a figure for demonstrating operation | movement of the audio processing apparatus which concerns on embodiment. It is a structure of one Embodiment of the audio processing apparatus which concerns on this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... CPU, 20 ... Voiceprint authentication part, 30 ... Voiceprint register, 40 ... Memory, 50 ... Communication interface, 60 ... Voice recognition part, PBX ... Private branch exchange, PH_A, PH_B, PH_C, PH_D, PH_E ... Communication apparatus, PC_A, PC_B, PC_C, PC_D, PC_E... Voice processing device.

Claims (5)

A speech processing device that classifies input speech for each speaker,
First storage means for storing one or more voiceprint data in association with speaker data corresponding to the speaker;
A second storage means;
Voice print data is extracted from the input voice every predetermined time, the voice print data is compared with the voice print data stored in the first storage means, and voice print authentication means for specifying speaker data;
And a control unit that stores the input speech for each predetermined time in the second storage unit for each speaker data specified by the voiceprint authentication unit. A voice processing system that classifies voice between a plurality of communication devices connected to a line network for each speaker,
Each communication device
First storage means for storing one or more voiceprint data in association with speaker data corresponding to the speaker;
A second storage means;
Voice print data is extracted from the voice every predetermined time, the voice print data is compared with the voice print data stored in the first storage means, and voice print authentication means for specifying speaker data;
Control means for storing the sound for each predetermined time in the second storage means for each speaker data specified by the voiceprint authentication means, respectively,
Each voice processing device is a voice processing system configured to be able to transmit and receive data by communication with each other. Each call device includes voice recognition means for recognizing voice between a plurality of call devices and converting it into a character string,
When the character string obtained by converting the voice by the voice recognition means is a predetermined command, the control means,
The voice processing system according to claim 2, wherein a process specified in advance by the predetermined command is executed on a voice processing device corresponding to the speaker data specified by the voiceprint authentication unit based on the voice. The predetermined command is any of speaker data stored in the second storage means,
In the predetermined processing, voice data corresponding to the speaker data stored in the second storage means is transmitted to a voice processing device corresponding to the voice speaker specified by the voiceprint authentication means. The voice processing system according to claim 3. The permitted speakers allowed to execute the predetermined command are limited in advance,
The character string obtained by converting the voice by the voice recognition unit is a predetermined command, and the speaker data specified by the voiceprint authentication unit based on the voice is speaker data corresponding to one of the permitted speakers. If there is
The voice processing system according to claim 3, wherein the control means executes a process specified in advance by the predetermined command.

Images