JP6369022B2 - Signal analysis apparatus, signal analysis system, and program - Google Patents

Description

  The present invention relates to a signal analysis device, a signal analysis system, and a program.

Conventionally, there is a technique called a general sidelobe canceller that extracts voice in a target direction and suppresses voice outside the target direction.
Patent Document 1 discloses a voice collection system that extracts and collects a target voice of interest from a plurality of voices having different directions of arrival, and includes at least a first and a second microphone. A microphone array having a plurality of microphones spaced apart from each other by a predetermined distance is obtained, and a plurality of CSP coefficients related to the direction of arrival of speech are obtained by performing discrete Fourier transform on the speech signals received by the first and second microphones. , A plurality of voice signals are detected from a plurality of CSP coefficients, and a voice direction defined according to an angle formed by a line segment connecting the first and second microphones and an arrival direction is determined from the obtained plurality of CSP coefficients. An index is detected, and a target voice signal is extracted from a plurality of detected voice signals based on the detected voice direction index. .

JP 2010-26361 A

  It is desirable to prevent erroneous determination when identifying whether the signal acquired by the signal acquisition means is a signal from a signal generation source or a signal other than that.

According to the first aspect of the present invention, the signals generated by the first signal acquisition unit and the second signal acquisition unit are arranged at different distances from the signal generation source and acquire a signal emitted from the signal generation source. As a frequency spectrum representing the relationship between the frequency and the intensity from the signal information acquisition unit for acquiring the information on each of the information on the respective signals generated by the first signal acquisition unit and the second signal acquisition unit, A frequency spectrum calculation unit that obtains a first frequency spectrum and a second frequency spectrum, respectively, and a signal emitted from the signal generation source from the first frequency spectrum and the second frequency spectrum, A first extraction unit for extracting a first extracted spectrum that has been processed to be larger than signals originating from different directions; A process in which a signal emitted from the signal generation source is larger than a signal emitted from a direction different from the signal generation source from the first frequency spectrum and the second frequency spectrum by a process different from the output unit. A second extraction unit that extracts the second extracted spectrum, and a third extraction unit that extracts, from the first frequency spectrum and the second frequency spectrum, a suppression spectrum that suppresses a signal emitted from the signal generation source. The extraction unit subtracts the suppression spectrum from the first extraction spectrum and obtains a first enhancement spectrum by performing a flooring process, and subtracts the suppression spectrum from the second extraction spectrum. An enhancement spectrum calculation unit for obtaining two enhancement spectra, and the first enhancement spectrum and the second enhancement spectrum. A correlation calculation unit for calculating a correlation that a signal analysis apparatus comprising: a.
According to a second aspect of the present invention, the first extraction unit takes a difference between the first frequency spectrum and the second frequency spectrum, so that a signal generated from the signal generation source generates the signal generation. The signal analyzing apparatus according to claim 1, wherein the signal analyzing apparatus performs processing that is larger than a signal emitted from a direction different from the source.
According to a third aspect of the present invention, the second extraction unit adds the first frequency spectrum and the second frequency spectrum so that the signal generated from the signal generation source is the signal generation source. The signal analysis apparatus according to claim 1, wherein the signal analysis processing is performed so as to be larger than signals emitted from different directions.
According to a fourth aspect of the present invention, in the first signal acquisition unit and the second signal acquisition unit, the signal generation source, the first signal acquisition unit, and the second signal acquisition unit are substantially in a straight line. The signal analysis device according to claim 1, wherein the signal analysis device is arranged so as to be.
According to a fifth aspect of the present invention, the third extraction unit, for the first frequency spectrum and the second frequency spectrum, the signal generation source, the first signal acquisition unit, and the second signal. 5. The suppression spectrum is extracted by adjusting a time difference that occurs due to a difference in distance from an acquisition unit until the signal arrives, and obtaining each difference. 6. It is a signal analysis device described in 1.
According to a sixth aspect of the present invention, the second extraction unit, for the first frequency spectrum and the second frequency spectrum, the signal generation source, the first signal acquisition unit, and the second signal. 4. The signal analysis according to claim 3, wherein the second extracted spectrum is extracted by adding each after adjusting a time difference that occurs due to a difference in distance from an acquisition unit and arrives at the signal. Device.
The invention according to claim 7 takes into account a change in angle between the signal generation source and the first signal acquisition means and the second signal acquisition means, which is caused by the movement of the signal generation source. The signal analysis apparatus according to claim 5, wherein the time difference is determined.
The invention according to claim 8 is the signal analysis apparatus according to any one of claims 1 to 7, wherein the signal is voice.
In the invention according to claim 9, the frequency spectrum calculation unit performs Fourier transform on information about each signal generated by the first signal acquisition unit and the second signal acquisition unit. Each of the frequency spectrum and the second frequency spectrum, and the correlation calculation unit calculates the respective correlations after performing inverse Fourier transform on the first enhancement spectrum and the second enhancement spectrum. The signal analysis device according to claim 1, wherein the signal analysis device is a signal analysis device.
The invention according to claim 10 is a signal generated by a first signal acquisition unit and a second signal acquisition unit that are arranged at different distances from the signal generation source and acquire a signal emitted from the signal generation source. As a frequency spectrum representing the relationship between the frequency and the intensity from the information regarding the frequency spectrum calculating unit for obtaining the first frequency spectrum and the second frequency spectrum, respectively, and the first frequency spectrum and the second frequency spectrum. An enhanced spectrum calculation unit that obtains a first enhanced spectrum and a second enhanced spectrum as separate enhanced methods as an enhanced spectrum in which a target direction signal that is a signal generated from the signal generation source is enhanced, and the first enhancement A correlation calculating unit that calculates a correlation between the spectrum and the second enhanced spectrum. A signal analyzer to.
The invention according to claim 11 is a signal acquisition unit comprising a first signal acquisition unit and a second signal acquisition unit, which are arranged at different distances from the signal generation source and acquire a signal emitted from the signal generation source. And identifying means for identifying a target direction signal that is a signal emitted from the signal generation source and a non-target direction signal that is a signal other than the signal emitted from the signal generation source, and the identification means includes the A signal information acquisition unit that acquires information about the signals generated by the first signal acquisition unit and the second signal acquisition unit, respectively, and the first signal acquisition unit and the second signal acquisition unit The frequency spectrum for obtaining the first frequency spectrum and the second frequency spectrum as the frequency spectrum representing the relationship between the frequency and the intensity from the information on each signal. A first calculating unit and a first frequency spectrum and a second frequency spectrum, wherein a signal emitted from the signal generation source is larger than a signal emitted from a direction different from the signal generation source. A signal generated from the signal generation source is generated from the first frequency spectrum and the second frequency spectrum by a process different from the first extraction unit for extracting the extraction spectrum and the first extraction unit. A second extraction unit that extracts a second extraction spectrum that has been processed to be larger than a signal emitted from a direction different from that of the signal generation source; and the signal from the first frequency spectrum and the second frequency spectrum. A third extraction unit for extracting a suppression spectrum in which a signal emitted from the generation source is suppressed; and subtracting the suppression spectrum from the first extraction spectrum; An enhanced spectrum calculation unit that obtains a first enhanced spectrum by performing a processing, and obtains a second enhanced spectrum by subtracting the suppression spectrum from the second extracted spectrum; and the first enhanced spectrum; A signal analysis system comprising: a correlation calculation unit that calculates correlation with the second enhancement spectrum.
The invention according to claim 12 is generated by the first signal acquisition means and the second signal acquisition means, which are arranged at different distances from the signal generation source in the computer and acquire signals emitted from the signal generation source. As a frequency spectrum representing the relationship between the frequency and the intensity from the function of acquiring information related to the received signals and the information related to the signals generated by the first signal acquisition means and the second signal acquisition means, A direction in which a signal generated from the signal generation source is different from the signal generation source based on a function for obtaining the first frequency spectrum and the second frequency spectrum, and the first frequency spectrum and the second frequency spectrum. a function of extracting the first extract spectrum a more larger processing signals emitted from, extract the first extraction spectrum The different processing is the ability to, from the first frequency spectrum and the second frequency spectrum, the processing signals emitted from the signal source is greater than the signal emitted from a direction different from the corresponding signal source A function of extracting the second extracted spectrum, a function of extracting a suppression spectrum in which a signal emitted from the signal source is suppressed, from the first frequency spectrum and the second frequency spectrum; A function of subtracting the suppression spectrum from the extracted spectrum and performing a flooring process to obtain a first enhancement spectrum, and subtracting the suppression spectrum from the second extraction spectrum to obtain a second enhancement spectrum; A function of calculating a correlation between the first enhanced spectrum and the second enhanced spectrum; Is a program to realize.

According to the first aspect of the present invention, it is possible to identify whether the sound acquired by the signal acquisition means is a signal from a signal generation source or a signal other than the case where this configuration is not adopted. In addition, it is possible to provide a signal analyzer that is unlikely to cause erroneous determination.
According to the second aspect of the present invention, the first extraction spectrum can be extracted more easily than in the case where this configuration is not adopted.
According to the invention of claim 3, the second extracted spectrum can be extracted more easily than in the case where the present configuration is not adopted.
According to the fourth aspect of the present invention, erroneous determination is further less likely to occur than in the case where the present configuration is not adopted.
According to the fifth aspect of the present invention, the suppression spectrum can be obtained more easily than in the case where this configuration is not adopted.
According to the sixth aspect of the present invention, the second extracted spectrum can be obtained with higher accuracy than when this configuration is not adopted.
According to the seventh aspect of the present invention, it is possible to cope with a case where the signal generation source moves.
According to the invention of claim 8, when identifying whether it is the speech voice of the wearer wearing the signal analysis device or the speech voice of another person, compared to the case where this configuration is not adopted, It is possible to provide a signal analyzer that is unlikely to cause erroneous determination.
According to the ninth aspect of the present invention, it is easier to obtain the first frequency spectrum and the second frequency spectrum than when the present configuration is not adopted, and the correlation calculation unit calculates the correlation. Becomes easier.
According to the tenth aspect of the present invention, when the voice acquired by the signal acquisition means is a signal from the signal generation source or other signals than when the present configuration is not adopted. In addition, it is possible to provide a signal analyzer that is unlikely to cause erroneous determination.
According to the eleventh aspect of the present invention, when identifying whether the sound acquired by the signal acquisition means is a signal from a signal generation source or a signal other than that in the case where this configuration is not adopted. In addition, it is possible to provide a signal analysis system in which erroneous determination is unlikely to occur.
According to the twelfth aspect of the present invention, when identifying whether the sound acquired by the signal acquisition means is a signal from a signal generation source or a signal other than that when the present configuration is not adopted. In addition, a function that makes it difficult for erroneous determination to occur can be realized by a computer.

It is a figure which shows the structural example of the audio | voice analysis system by this embodiment. (A)-(b) is a figure which shows the structural example of the terminal device in this embodiment. It is a figure which shows the positional relationship of a wearer and another person's mouth (speaking part), and a microphone. It is the figure which showed the function structural example of the audio | voice analysis part in this embodiment. It is a flowchart which shows operation | movement of the terminal device in this embodiment. It is a figure explaining the process of the signal processing in an audio | voice analysis part. It is the figure which showed an example of the audio | voice signal acquired by the audio | voice information acquisition means. (A) to (b) show what kind of spectrum is generated between the voice uttered by the wearer and the voice of the other person when the difference between the first frequency spectrum and the second frequency spectrum is taken. It is a figure explaining whether it is extracted. Here, in order to make the explanation easy to understand, not the frequency spectrum but an audio signal before Fourier transform will be explained. Therefore, in (a) to (b), the horizontal axis represents time (s), and the vertical axis represents the intensity of the audio signal. In (a) to (b), when the first frequency spectrum and the second frequency spectrum are added, what kind of spectrum is extracted between the voice uttered from the wearer's voice and the voice of the other person. It is a figure explaining how. Here, in order to make the explanation easy to understand, not the frequency spectrum but an audio signal before Fourier transform will be explained. Therefore, in (a) to (b), the horizontal axis represents time (s), and the vertical axis represents the intensity of the audio signal. (A)-(b) adjusts the time difference until the voice arrives for the first frequency spectrum and the second frequency spectrum due to the difference in distance between the wearer's utterance part and the first microphone and the second microphone. Then, it is a figure explaining what kind of spectrum is extracted by the voice uttered from the wearer's utterance part and the voice of the other person when each difference is obtained. Here, in order to make the explanation easy to understand, not the frequency spectrum but an audio signal before Fourier transform will be explained. Therefore, in (a) to (b), the horizontal axis represents time (s), and the vertical axis represents the intensity of the audio signal. It is an example of the 1st target direction emphasis spectrum about a wearer's speech voice, and the 2nd target direction emphasis spectrum. It is an example of the 1st target direction emphasis spectrum about the speech sound of others, and the 2nd target direction emphasis spectrum. It is an example of the 1st target direction emphasis spectrum and the 2nd target direction emphasis spectrum when both a wearer's speech sound and the speech sound of others are acquired with the microphone. It is the figure which showed the result of having performed the process demonstrated after FIG. 4 with respect to the audio | voice signal shown in FIG. It is a figure which shows the condition where the several wearer who each mounted | wore with the terminal device of this embodiment is talking. It is a figure which shows the example of the speech information of each terminal device in the conversation condition of FIG.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Here, speech is exemplified as a signal, and the speech analysis system is described as an example of the signal analysis system.
<System configuration example>
FIG. 1 is a diagram illustrating a configuration example of a speech analysis system according to the present embodiment.
As shown in FIG. 1, the speech analysis system 1 of this embodiment includes a terminal device 10 and a host device 20. The terminal device 10 and the host device 20 are connected via a wireless communication line. As a type of the wireless communication line, a line using an existing method such as Wi-Fi (Wireless Fidelity), Bluetooth (registered trademark), ZigBee, or UWB (Ultra Wideband) may be used. Further, in the illustrated example, only one terminal device 10 is described, but the terminal device 10 is worn and used by each user, and actually the terminal devices 10 for the number of users. Is prepared. Hereinafter, a user wearing the terminal device 10 is referred to as a wearer.

  The terminal device 10 has a plurality of microphones (first microphone 11 (first signal acquisition means) and second microphone 12 (second signal acquisition means) as voice acquisition means (signal acquisition means) for acquiring voice. And an amplifier (the first amplifier 13 and the second amplifier 14) The terminal device 10 also includes a voice analysis unit 15 that analyzes the acquired voice, and data transmission for transmitting the analysis result to the host device 20. And a power supply unit 17.

  The 1st microphone 11 and the 2nd microphone 12 are arranged in the position where the distance from a wearer's mouth (voice part and signal generation source) differs. Here, the first microphone 11 is disposed at a position (for example, about 10 cm) close to the mouth (speaking part) of the wearer, and the second microphone 12 is a position (for example, about 35 cm) far from the mouth (speaking part) of the wearer. ). As the types of microphones used as the first microphone 11 and the second microphone 12 of the present embodiment, various existing types such as a dynamic type and a condenser type may be used. In particular, a non-directional MEMS (Micro Electro Mechanical Systems) type microphone is preferable.

  The first amplifier 13 and the second amplifier 14 amplify electrical signals (audio signals) that are output according to the audio acquired by the first microphone 11 and the second microphone 12, respectively. As the amplifier used as the first amplifier 13 and the second amplifier 14 of the present embodiment, an existing operational amplifier or the like may be used.

  The voice analysis unit 15 analyzes the voice signal output from the first amplifier 13 and the second amplifier 14. Then, it is identified whether the voice acquired by the first microphone 11 and the second microphone 12 is a voice uttered by the wearer wearing the terminal device 10 or a voice uttered by another person. Details of specific processing for voice identification will be described later. As will be described in detail later, the voice analysis unit 15 is a voice other than a target direction signal that is a voice (signal) emitted from the wearer's mouth (speech part, signal generation source) and a voice emitted from the wearer's mouth. It functions as a discriminating means (signal analyzing device) for discriminating a non-target direction signal.

  The data transmission unit 16 transmits the acquired data including the analysis result by the voice analysis unit 15 and the terminal ID to the host device 20 via the wireless communication line. As information to be transmitted to the host device 20, in addition to the above analysis results, for example, the acquisition time of the sound by the first microphone 11 and the second microphone 12 and the acquired sound according to the contents of the processing performed in the host device 20 Information such as sound pressure may be included. Further, the terminal device 10 may be provided with a data storage unit for storing the analysis result by the voice analysis unit 15 and the stored data for a certain period may be transmitted collectively. It may be transmitted via a wired line.

  The power supply unit 17 supplies power to the first microphone 11, the second microphone 12, the first amplifier 13, the second amplifier 14, the voice analysis unit 15, and the data transmission unit 16. As the power source, for example, an existing power source such as a dry battery or a rechargeable battery is used. Moreover, the power supply part 17 contains well-known circuits, such as a voltage conversion circuit and a charge control circuit, as needed.

  The host device 20 outputs a data reception unit 21 that receives data transmitted from the terminal device 10, a data storage unit 22 that stores the received data, a data analysis unit 23 that analyzes the stored data, and an analysis result. Output unit 24. The host device 20 is realized by an information processing device such as a personal computer. Further, as described above, a plurality of terminal devices 10 are used in the present embodiment, and the host device 20 receives data from each of the plurality of terminal devices 10.

The data receiving unit 21 corresponds to the above-described wireless line, receives data from each terminal device 10, and sends it to the data storage unit 22.
The data storage unit 22 is realized by a storage device such as a magnetic disk device of a personal computer, for example, and stores received data acquired from the data receiving unit 21 for each speaker. Here, the speaker is identified by collating the terminal ID transmitted from the terminal device 10 with the name of the speaker registered in advance in the host device 20 and the terminal ID. Further, the wearer state may be transmitted from the terminal device 10 instead of the terminal ID.

  The data analysis unit 23 is realized by a program-controlled CPU of a personal computer, for example, and analyzes data stored in the data storage unit 22. The specific analysis content and analysis method can take various contents and methods depending on the purpose and use of the system of the present embodiment. For example, the frequency of dialogue between wearers of the terminal device 10 and the tendency of each wearer's dialogue partner may be analyzed, or the relationship between the dialogues may be inferred from information on individual utterance length and sound pressure in the dialogue. Done.

  The output unit 24 outputs the analysis result from the data analysis unit 23 or performs output based on the analysis result. The means for outputting the analysis result or the like can take various means such as display display, print output by a printer, voice output, and the like according to the purpose and use of the system and the content and format of the analysis result.

<Configuration example of terminal device>
2A and 2B are diagrams illustrating a configuration example of the terminal device 10. Of these, FIG. 2 (a) is a view of the state when the wearer actually wears the terminal device 10 as seen from the front of the wearer, and FIG. 2 (b) is a view of this state from the side. FIG.
As described above, the terminal device 10 is used by being attached to each user. In order for the user to be able to wear the terminal device 10 according to the present embodiment, as shown in FIGS. 2A to 2B, the device main body 30 and a strap 40 connected to the device main body 30 are provided. The configuration is as follows. In the configuration shown in the figure, the user puts the neck through the strap 40 and hangs the apparatus main body 30 from the neck.

  The apparatus main body 30 is a thin rectangular parallelepiped case 31 made of metal, resin, or the like, and a circuit and a power source for realizing at least the first amplifier 13, the second amplifier 14, the voice analysis unit 15, the data transmission unit 16, and the power supply unit 17. The power supply (battery) of the unit 17 is accommodated. The case 31 may be provided with a pocket for inserting an ID card or the like displaying ID information such as the name and affiliation of the wearer. Further, such ID information or the like may be printed on the surface of the case 31 itself, or a sticker describing the ID information or the like may be attached.

  The strap 40 is provided with a first microphone 11 and a second microphone 12 (hereinafter referred to as microphones 11 and 12 when the first microphone 11 and the second microphone 12 are not distinguished). The microphones 11 and 12 are connected to the first amplifier 13 and the second amplifier 14 accommodated in the apparatus main body 30 by cables (electric wires or the like) passing through the inside of the strap 40. As the material of the strap 40, various existing materials such as leather, synthetic leather, cotton and other natural fibers and synthetic fibers such as resin, metal, and the like may be used. Moreover, the coating process using a silicon resin, a fluororesin, etc. may be given.

  The strap 40 has a cylindrical structure, and the microphones 11 and 12 are housed inside the strap 40. By providing the microphones 11 and 12 inside the strap 40, the microphones 11 and 12 are prevented from being damaged and dirty, and the conversation person is prevented from being aware of the presence of the microphones 11 and 12. The second microphone 12 disposed at a position far from the wearer's mouth (speaking part) may be provided in the apparatus main body 30. In the present embodiment, a case where the second microphone 12 is provided on the strap 40 will be described as an example.

  Referring to FIGS. 2A to 2B, the first microphone 11 is positioned away from the end of the strap 40 connected to the device main body 30 (for example, a position about 20 cm to 30 cm from the connection site). Is provided. Accordingly, the first microphone 11 is positioned at the wearer's neck (for example, a position corresponding to the clavicle) in a state where the wearer hangs the strap 40 on the neck and the apparatus main body 30 is lowered, and the wearer's mouth (voice) From about 10 cm to 20 cm.

  The second microphone 12 is provided at an end of the strap 40 connected to the device main body 30 (for example, a position within 10 cm from the connection site). Accordingly, the second microphone 12 is disposed at a position about 30 cm to 40 cm away from the wearer's mouth (speaking part) in a state where the wearer hangs the strap 40 around the neck and lowers the apparatus main body 30. . Even when the second microphone 12 is provided in the apparatus main body 30, the distance from the wearer's mouth (speaking part) to the second microphone 12 is approximately the same.

  In the present embodiment, as shown in FIG. 2 (b), the first microphone 11 and the second microphone 12 are arranged so that the first microphone 11, the second microphone 12 and the mouth (speaking part) of the wearer are substantially in a straight line. Arrange to become. In addition, here, the direction of the wearer's mouth (speaking part) with respect to the first microphone 11 and the second microphone 12 is referred to as a “target direction”, and the other direction is referred to as a “non-target direction”. .

  In addition, the terminal device 10 of this embodiment is not limited to the structure shown to Fig.2 (a)-(b). For example, in the microphones 11 and 12, the distance of the sound wave arrival path from the second microphone 12 to the wearer's mouth (speech part) is the distance of the sound wave arrival path from the first microphone 11 to the wearer's mouth (speech part). The positional relationship between the first microphone 11 and the second microphone 12 may be specified so as to be several times. Further, the microphones 11 and 12 are not limited to the configuration provided on the strap 40 as described above, and may be attached to the wearer by various methods. For example, each of the first microphone 11 and the second microphone 12 may be configured to be individually fixed to clothes using pins or the like. Alternatively, a dedicated mounting tool designed so that the positional relationship between the first microphone 11 and the second microphone 12 is fixed at a desired position may be prepared and mounted.

  Moreover, as shown to Fig.2 (a)-(b), the apparatus main body 30 is comprised not only as the structure connected to the strap 40 and being hung up from a wearer's neck but as an apparatus which is easy to carry. It only has to be done. For example, instead of the strap as in the present embodiment, it may be configured to be attached to clothes or a body with a clip or a belt, or may be configured to be simply carried in a pocket or the like. In addition, a function of receiving, amplifying, and analyzing a voice signal from the microphones 11 and 12 may be realized in a mobile phone or other existing portable electronic information terminal.

  Furthermore, the microphones 11 and 12 and the apparatus main body 30 (or the voice analysis unit 15) may be connected by wireless communication instead of being connected by wire. The first amplifier 13, the second amplifier 14, the voice analysis unit 15, the data transmission unit 16, and the power supply unit 17 are housed in a single case 31 in the above configuration example, but are configured as a plurality of individuals. May be. For example, the power supply unit 17 may be used by being connected to an external power supply without being stored in the case 31.

<Identification of speakers (self and others) based on non-linguistic information of acquired speech>
The system of the present embodiment distinguishes between the voice of the wearer of the terminal apparatus 10 and the voice of the other person using the voice information acquired by the two microphones 11 and 12 provided in the terminal apparatus 10. To do. In other words, the present embodiment identifies self / other distinction (self / other identification) regarding the speaker of the acquired voice. Further, in the present embodiment, the speaker is identified based on non-linguistic information, not linguistic information obtained by using morphological analysis or dictionary information in the acquired speech information. In other words, the voice speaker is identified not from the utterance content specified by the linguistic information but from the utterance situation specified by the non-linguistic information.

FIG. 3 is a diagram illustrating a positional relationship between the mouths (speaking parts) of the wearer and the other person and the microphones 11 and 12.
In the relationship shown in FIG. 3, let La1 be the distance between the sound source a that is the mouth (speaking part) of the wearer and the first microphone 11, and let La2 be the distance between the sound source a and the second microphone 12. Further, the distance between the sound source b, which is the mouth (speaking part) of the other person, and the first microphone 11 is Lb1, and the distance between the sound source b and the second microphone 12 is Lb2. In this case, the following relationship holds.
La1 <La2 (La2≈1.5 × La1 to 4 × La1)
Lb1≈Lb2

At this time, the sound pressure is attenuated according to the distance between the microphones 11 and 12 and the sound source.
Therefore, whether the voice is the voice of the wearer or the voice of the other using the difference in sound pressure between the first microphone 11 and the second microphone 12 generated according to the difference in distance between La1 and La2 and Lb1 and Lb2. It is conceivable to determine the self-other identification.
However, with this method, when the wearer's speech is inflected, or when the wearer performs an operation such as arm-arming or sitting at the desk, the expected sound pressure cannot be obtained, Misidentification may be caused about self-other identification. In addition, an erroneous determination may occur depending on the volume of the surrounding voice, such as when another person speaks nearby.

  In addition, when self-other identification is performed by a method using a general sidelobe canceller, if the sound in the target direction deviates from the set direction, the effect of suppressing the sound in the non-target direction will be reduced, and the other party's voice will be ignored regardless of the wearer's voice. Are often determined to be uttered voices. In addition, if there is an inflection in the voice of the wearer, it is often determined that the voice of the wearer is the voice of the other person where the wearer's voice is small.

<Explanation of speech analysis unit>
Therefore, in this embodiment, this problem is suppressed by configuring the voice analysis unit 15 as follows.
FIG. 4 is a diagram illustrating a functional configuration example of the voice analysis unit 15 in the present embodiment.
As illustrated, the voice analysis unit 15 includes a voice information acquisition unit 151, a frequency spectrum calculation unit 152, a first target direction signal spectrum extraction unit 153, a second target direction signal spectrum extraction unit 154, and a non-target A direction signal spectrum extraction unit 155, a target direction enhancement spectrum calculation unit 156, a correlation calculation unit 157, and an own / other identification unit 158 are provided.

  The audio information acquisition unit 151 is an example of a signal information acquisition unit, and acquires information related to audio generated by the first microphone 11 and the second microphone 12. In the present embodiment, the audio information acquisition unit 151 acquires the audio signal output from the microphones 11 and 12 and amplified through the first amplifier 13 and the second amplifier 14 as information related to audio.

  The frequency spectrum calculation unit 152 performs Fourier transform on information about each sound generated by the first microphone 11 and the second microphone 12, and uses the first frequency spectrum and the second frequency spectrum as frequency spectra representing the relationship between the frequency and the intensity. Each frequency spectrum is obtained. That is, the Fourier transform is performed on the audio signals from the first microphone 11 and the second microphone 12, and the frequency spectrum is obtained for each.

  The first target direction signal spectrum extraction unit 153 is an example of a first extraction unit, and the voice uttered from the wearer's utterance part by taking the difference between the first frequency spectrum and the second frequency spectrum. A first target direction signal spectrum to be expressed (first extracted spectrum) is extracted. The first target direction signal spectrum is obtained by processing such that a signal emitted from the wearer's utterance region is larger than a signal emitted from a direction different from the wearer's utterance region.

The second target direction signal spectrum extraction unit 154 is an example of a second extraction unit, and represents a voice uttered from the wearer's utterance part by adding the first frequency spectrum and the second frequency spectrum. A second target direction signal spectrum (second extracted spectrum) is extracted. This second target direction signal spectrum is also obtained by processing in which the signal emitted from the wearer's utterance part becomes larger than the signal emitted from a direction different from the utterance part of the wearer.
The first target direction signal spectrum and the second target direction signal spectrum will be described later.

  The non-target direction signal spectrum extraction unit 155 is an example of a third extraction unit, and converts information related to the sound generated by the first microphone 11 and the second microphone 12 from the first frequency spectrum and the second frequency spectrum. A non-target direction signal spectrum (suppression spectrum) that is included and represents a signal other than a voice uttered from the uttered part of the wearer is extracted. The non-target direction signal spectrum is obtained by suppressing a signal emitted from the wearer's utterance part. A specific method for extracting the non-target direction signal spectrum will be described later.

  The target direction enhancement spectrum calculation unit 156 is an example of the enhancement spectrum calculation unit, and subtracts the non-target direction signal spectrum from the first target direction signal spectrum (spectral subtraction process). Further, by performing flooring processing, a first target direction enhancement spectrum is obtained as a target direction enhancement spectrum that emphasizes a target direction signal that is a voice uttered from the wearer's voice. Further, the target direction enhancement spectrum calculation unit 156 obtains the second target direction enhancement spectrum as the target direction enhancement spectrum by subtracting the non-target direction signal spectrum from the second target direction signal spectrum (spectral subtraction process). The spectral subtraction processing, flooring processing, and the like performed here will be described later.

  The correlation calculation unit 157 calculates the correlation by performing inverse Fourier transform on the first target direction enhancement spectrum and the second target direction enhancement spectrum.

  Based on the correlation information calculated by the correlation calculation unit 157, the self-other identification unit 158 determines whether the voice acquired by the microphones 11 and 12 is the voice of the wearer or other person other than the wearer. Identify whether the voice is speech.

<Operation example of terminal device>
FIG. 5 is a flowchart showing the operation of the terminal device 10 in the present embodiment. FIG. 6 is a diagram illustrating a signal processing process in the voice analysis unit 15. Hereinafter, the operation of the terminal device 10 will be described with reference to FIGS. 2 and 4 to 6.
As shown in FIG. 5, when the microphones 11 and 12 of the terminal device 10 acquire sound, an electric signal (audio signal) corresponding to the acquired sound is transmitted from each microphone 11 and 12 to the first amplifier 13 and the second amplifier 14. (Step 101). When the first amplifier 13 and the second amplifier 14 acquire the audio signal from the microphones 11 and 12, the first amplifier 13 and the second amplifier 14 amplify the signal and send it to the audio analysis unit 15 (step 102).

  The audio information acquisition unit 151 of the audio analysis unit 15 acquires the amplified audio signal (step 103).

FIG. 7 is a diagram illustrating an example of an audio signal acquired by the audio information acquisition unit 151. In FIG. 7, the horizontal axis represents time (s), and the vertical axis represents the intensity of the audio signal.
In the illustrated example, there is a self-speaking section in which the wearer speaks between 0 s and about 5 s. There is an other person utterance section where another person other than the wearer speaks between about 5 s and about 10 s, and both the wearer and the other person speak between about 10 s and about 15 s. There is another utterance section.

  The voice information acquisition unit 151 calculates the average sound pressure of the acquired voices of the microphones 11 and 12 for each predetermined time unit (for example, tens of seconds to hundreds of seconds). Obtain (step 104). Then, it is determined whether or not the average sound pressure is greater than or equal to a certain threshold (whether or not there is a gain) (step 105).

  If there is a gain of the average sound pressure in each of the microphones 11 and 12 obtained in step 105 (Yes in step 105), the audio information acquisition unit 151 determines that there is an utterance voice (utterance has been performed).

  Then, the frequency spectrum calculation unit 152 performs a Fourier transform on the audio signals from the first microphone 11 and the second microphone 12, and obtains the frequency spectrum for each (step 106). In FIG. 8, this is illustrated as two “FFT (Fast Fourier Transform)”. Thereby, the first frequency spectrum is obtained based on the audio signal from the first microphone 11, and the second frequency spectrum is obtained based on the audio signal from the second microphone 12. The audio signal is converted from the time domain to the frequency domain by the frequency spectrum calculation unit 152.

  Next, the first target direction signal spectrum extraction unit 153 takes the difference between the first frequency spectrum and the second frequency spectrum, thereby expressing the first target direction signal spectrum representing the voice uttered from the wearer's speaking part. Is extracted (step 107).

  FIGS. 8A and 8B show what kind of spectrum is extracted between the voice of the wearer and the voice of the other when the difference between the first frequency spectrum and the second frequency spectrum is taken. It is a figure explaining what is done. Here, in order to make the explanation easy to understand, not the frequency spectrum but an audio signal before Fourier transform will be explained. Therefore, in FIGS. 8A to 8B, the horizontal axis represents time (s), and the vertical axis represents the intensity of the audio signal.

  FIG. 8A shows an audio signal when the speech sound of the wearer is acquired by each of the microphones 11 and 12. Here, since the second microphone 12 is located farther from the wearer's utterance site than the first microphone 11, the audio signal reaches the second microphone 12 later in time than the first microphone 11. Therefore, when the difference between these audio signals is taken, both these audio signals remain as they are.

On the other hand, FIG. 8B shows a voice signal when another person's uttered voice is acquired by each of the microphones 11 and 12.
Here, the other person's voice or the like arrives at approximately the same time in the first microphone 11 and the second microphone 12 because the distances until they reach the first microphone 11 and the second microphone 12 are substantially the same. Therefore, when the difference between these audio signals is taken, they cancel each other and the amplitude becomes small.

Thus, when the difference between the first frequency spectrum and the second frequency spectrum is taken, the frequency spectrum of the voice uttered from the wearer's voice part remains as it is, while the other signals have smaller amplitudes. . That is, in the first target direction signal spectrum obtained by taking the difference between the first frequency spectrum and the second frequency spectrum, the one representing the sound uttered from the wearer's utterance portion occupies a larger portion. That is, it is possible to extract the voice uttered from the uttered part of the wearer.
Hereinafter, the first target direction signal spectrum obtained in this way may be referred to as “α1”.

Next, the second target direction signal spectrum extraction unit 154 adds the first frequency spectrum and the second frequency spectrum to generate a second target direction signal spectrum representing the voice uttered from the wearer's speaking part. Extract (step 108).
At this time, the second frequency spectrum is preferably passed through the delay circuit indicated by “delay” in FIG. At this time, the time adjusted by the delay circuit is set to be the time difference until the voice produced by the distance difference between the utterance part of the wearer and the first microphone 11 and the second microphone 12 arrives. In other words, the voice from the utterance part of the wearer can be treated as having arrived at almost the same time regardless of the distance difference between the utterance part and the first microphone 11 and the second microphone 12.

  9A and 9B, when the first frequency spectrum and the second frequency spectrum are added, what kind of spectrum is extracted between the voice of the wearer and the voice of the other person. FIG. Here, in order to make the explanation easy to understand, not the frequency spectrum but an audio signal before Fourier transform will be explained. Therefore, in FIGS. 9A to 9B, the horizontal axis represents time (s), and the vertical axis represents the intensity of the audio signal.

  FIG. 9A illustrates the sound signal when the wearer's sound signal is acquired by each of the microphones 11 and 12. Here, the audio signal from the second microphone 12 is adjusted by the above-described delay circuit, and apparently arrives at substantially the same time as the audio signal from the first microphone 11. In FIG. 9A, this time is shown as “Δ”. In this case, when the first frequency spectrum and the second frequency spectrum are added, the amplitude is amplified and becomes larger.

On the other hand, FIG. 9B illustrates an audio signal when the voice of another person is acquired by each of the microphones 11 and 12.
Here, the other person's voice or the like arrives at approximately the same time in the first microphone 11 and the second microphone 12 because the distances until they reach the first microphone 11 and the second microphone 12 are substantially the same. However, since the audio signal from the second microphone 12 is adjusted by the delay circuit described above, a time Δ shift occurs. Therefore, when the first frequency spectrum and the second frequency spectrum are added, both of these audio signals remain as they are.

When the first frequency spectrum and the second frequency spectrum are added in this way, the frequency spectrum for the wearer's speech is amplified, while the frequency spectrum for the other's speech remains. There is no amplification. That is, the second target direction signal spectrum obtained by adding the first frequency spectrum and the second frequency spectrum occupies a larger portion that represents the voice uttered from the uttered part of the wearer. That is, it is possible to extract the voice uttered from the uttered part of the wearer.
Hereinafter, the second target direction signal spectrum obtained in this way may be referred to as “α2”.

Next, the target direction enhancement spectrum calculation unit 156 extracts a non-target direction signal spectrum from the first frequency spectrum and the second frequency spectrum (step 109).
More specifically, the non-target direction signal spectrum extraction unit 155 generates the first frequency spectrum and the second frequency spectrum due to the distance difference between the wearer's utterance site and the first microphone 11 and the second microphone 12. After adjusting the time difference until the voice arrives, the non-target direction signal spectrum is extracted by obtaining each difference.

  FIGS. 10A to 10B show the first frequency spectrum and the second frequency spectrum until the voice arrives due to the difference in the distance between the wearer's utterance part and the first microphone 11 and the second microphone 12. It is a figure explaining what kind of spectrum is extracted by a wearer's utterance voice and another person's utterance voice, when each time difference is adjusted and each difference is calculated. Here, in order to make the explanation easy to understand, not the frequency spectrum but an audio signal before Fourier transform will be explained. Therefore, in FIGS. 10A to 10B, the horizontal axis represents time (s), and the vertical axis represents the intensity of the audio signal.

  FIG. 10A illustrates the sound signal when the wearer's sound signal is acquired by each of the microphones 11 and 12. Here, the audio signal from the second microphone 12 is adjusted by the above-described delay circuit, and apparently arrives at substantially the same time as the audio signal from the first microphone 11. In FIG. 10A, this time is illustrated as “Δ”. In this case, when the difference between the first frequency spectrum and the second frequency spectrum is obtained, they are canceled out and the amplitude is reduced.

On the other hand, FIG. 10B illustrates an audio signal when another person's speech is acquired by each of the microphones 11 and 12.
Here, since the distance until the other person's voice reaches the first microphone 11 and the second microphone 12 is substantially the same, the first microphone 11 and the second microphone 12 arrive at substantially the same time. However, since the audio signal from the second microphone 12 is adjusted by the delay circuit described above, a time Δ shift occurs. Therefore, when the difference between the first frequency spectrum and the second frequency spectrum is obtained, both of these audio signals remain as they are.

Therefore, in this case, the frequency spectrum of the wearer's uttered voice has a small amplitude, while the frequency spectrum of the other person's uttered voice remains. That is, in this case, a signal representing a signal other than the voice uttered from the uttered part of the wearer occupies a larger part. That is, a signal other than the voice uttered from the wearer's utterance part can be extracted.
Hereinafter, the non-target direction signal spectrum obtained in this way may be referred to as “β”.

  Next, the target direction emphasis spectrum calculation unit 156 subtracts the non-target direction signal spectrum β from the first target direction signal spectrum α1 and further performs flooring processing to thereby generate a target voice that is uttered from the wearer's voice part. A first target direction enhancement spectrum is obtained as a target direction enhancement spectrum in which the direction signal is enhanced (step 110).

  That is, here, the frequency spectrum of “α1−β” is first obtained (spectral subtraction process). At this time, about the speech voice of the wearer, since the difference between the result of FIG. 8A and the result of FIG. 10A is conceptually taken, α1−β> 0. On the other hand, the other person's speech is conceptually the difference between the result of FIG. 8B and the result of FIG. 10B, and α1−β <0.

  The flooring process is a process of replacing small values included in actual data with appropriate numerical values without using them as they are. As an example of the replacement value, for example, a numerical value such as “0” or “0.1” is used. However, in this embodiment, it is replaced with “0” in this embodiment.

In the present embodiment, this flooring process is applied when α1-β <0. Then, the frequency spectrum of “α1-β” becomes the value of “α1-β” as it is for the utterance voice of the wearer, and becomes 0 for the utterance voice of the other person. Eventually, this frequency spectrum becomes a target direction enhancement spectrum in which a target direction signal which is a voice uttered from a wearer's utterance part is emphasized.
Hereinafter, the first target direction enhancement spectrum obtained in this way may be referred to as “SS1”.

  Further, the target direction enhancement spectrum calculation unit 156 obtains the second target direction enhancement spectrum as the target direction enhancement spectrum by subtracting the non-target direction signal spectrum β from the second target direction signal spectrum α2 (step 111).

  That is, here, the frequency spectrum of “α2−β” is obtained (spectral subtraction process). At this time, the voice uttered from the uttered part of the wearer conceptually takes the difference between the result of FIG. 9A and the result of FIG. 10A. For other signals, conceptually, the difference between the results in FIG. 9B and the results in FIG. 10B is taken. Here, since the flooring process is not performed, both values are “α2−β”.

However, when the difference between the result of FIG. 10A from the result of FIG. 9A and the difference of the result of FIG. 10B from the result of FIG. 9B is compared, the former has a larger amplitude. Become. Eventually, this frequency spectrum also becomes a target direction enhancement spectrum in which a target direction signal that is a voice uttered from the wearer's utterance portion is emphasized.
Hereinafter, the second target direction enhancement spectrum thus obtained may be referred to as “SS2”.

  As described above, in the present embodiment, by performing spectral subtraction processing in the frequency spectrum, the target direction signal representing the voice uttered from the wearer's uttered part is emphasized, and the non-target direction signal representing the voice of the other person Two target direction emphasizing spectra with suppressed are generated. By performing such processing on the frequency spectrum, the target direction signal can be more emphasized and the non-target direction signal can be suppressed more than the voice signal is processed as it is. This uses the spurt nature that even when multiple sounds are mixed, the signal energy exists only sparsely in the frequency domain.

  Next, the correlation calculation unit 157 performs inverse Fourier transform on the first target direction enhancement spectrum SS1 and the second target direction enhancement spectrum SS2 (step 112). In FIG. 6, this is illustrated as two “IFFT (Inverse Fast Fourier Transform)”. As a result, the first target direction enhancement spectrum SS1 and the second target direction enhancement spectrum SS2 return to the audio signal. As a result, the first target direction enhancement spectrum SS1 and the second target direction enhancement spectrum SS2 are converted from the frequency domain to the time domain.

  Then, the correlation calculation unit 157 calculates the correlation between the first target direction enhancement spectrum and the second target direction enhancement spectrum (step 113). Here, this calculation is performed after performing the inverse Fourier transform, but may be performed without performing the inverse Fourier transform.

  When the correlation calculation unit 157 calculates that the correlation between the two is high compared to a predetermined threshold (Yes in step 113), the self-other identification unit 158 is acquired by the microphones 11 and 12. It is determined that the received voice is the voice of the wearer (step 114). If the correlation calculation unit 157 calculates that the correlation between the two is low (No in step 113), the own / other identification unit 158 determines that the speech is not the wearer's speech but the speech of the other person. (Step 115). If there is no gain of the average sound pressure in each of the microphones 11 and 12 obtained in step 104 (No in step 105), the own / other identification unit 158 determines that there is no uttered voice (no utterance is performed) ( Step 116). The threshold for determining the level of correlation may be determined by the correlation calculation unit 157, or the correlation values calculated for the first target direction enhancement spectrum and the second target direction enhancement spectrum are presented to the analyst once. Then, the analyst may determine the threshold value.

FIGS. 11 to 13 are specific examples showing the first target direction enhancement spectrum SS1 and the second target direction enhancement spectrum SS2.
Among these, FIG. 11 is an example of the first target direction enhancement spectrum SS1 and the second target direction enhancement spectrum SS2 for the speech of the wearer. As shown in the figure, it is understood that the correlation between the two is high.

  FIG. 12 is an example of the first target direction enhancement spectrum SS1 and the second target direction enhancement spectrum SS2 for the speech of another person. Here, the amplitude of the first target direction enhancement spectrum SS1 is 0 because the flooring process is performed. Therefore, it can be seen that the correlation between the two is low.

  Further, FIG. 13 is an example of the first target direction enhancement spectrum SS1 and the second target direction enhancement spectrum SS2 when both the voice of the wearer and the voice of the other person are acquired by the microphones 11 and 12. . Also at this time, the correlation between the two is high as in the case of FIG.

  The following items are summarized as shown in Table 1 below.

  In order to obtain the correlation in the correlation calculation unit 157, an existing method can be used. For example, a cross-correlation coefficient is obtained for the first target direction enhancement spectrum SS1 and the second target direction enhancement spectrum SS2, and the correlation between the two is calculated based on the value of the correlation coefficient. Alternatively, a method using a cross correlation function may be used. However, the accuracy of self-other identification is further improved when the cross-correlation function is used rather than the cross-correlation coefficient.

FIG. 14 is a diagram showing a result of performing the processing described in FIG. 4 and subsequent drawings on the audio signal shown in FIG. In FIG. 14, the horizontal axis represents time (s), and the vertical axis represents the standardized value of the cross-correlation coefficient.
In the self-speaking section where the wearer utters between 0 s and about 5 s, the value of the cross-correlation coefficient is high. In this section, the voice acquired by the microphones 11 and 12 is the voice of the wearer. Can be judged. Further, the cross-correlation coefficient is low in the other person utterance section between about 5 s and about 10 s, and in this section, it can be determined that the voice acquired by the microphones 11 and 12 is the voice of the wearer. Further, in the subsequent self-other speech period of about 10 s to about 15 s, the value of the cross-correlation coefficient is high, and in this period, the voices acquired by the microphones 11 and 12 include the voice of the wearer. Can be determined.

  Thereafter, the voice analysis unit 15 transmits the information obtained by the processing of Step 104 to Step 116 to the host device 20 as an analysis result via the data transmission unit 16. As the analysis result, for example, presence / absence of utterance, wearer's information (terminal ID), and voices acquired by the microphones 11 and 12 are voices of the wearer or voices of others other than the wearer. This is self-other identification information, etc., that is information identifying the above. At this time, the length of the utterance time for each speaker (the wearer himself or another person), the value of the average sound pressure gain, and other additional information may be transmitted to the host device 20 together with the analysis result.

  In the present embodiment, two target direction enhancement spectra are generated by emphasizing the target direction signal representing the voice uttered from the uttered part of the wearer and suppressing the non-target direction signal representing the voice of the other person (first). Target direction enhancement spectrum, second target direction enhancement spectrum). Based on these correlations, the voices acquired by the microphones 11 and 12 are identified. According to this method, it is determined whether the voice is the wearer's speech or the other's speech based on the similarity of the waveform rather than the strength of the speech. That is, even when there is an inflection in the wearer's spoken voice, this determination can be performed with higher accuracy. Therefore, when using the sound pressure described with reference to FIGS. 3 to 5 and the method using the general sidelobe canceller, the accuracy of self-other identification is further improved. And it becomes easier to distinguish the self-speaking section from other sections. That is, even when the wearer's utterance voice has an inflection or the like, the self-speaking section can be continuously specified. On the other hand, in the case of using the sound pressure described in FIGS. 3 to 5 and the method using the general sidelobe canceller, it is often determined that the other person's utterance section exists even in the self utterance section. The utterance section is finely divided, and the self utterance section cannot be accurately grasped.

  In the present embodiment, the accuracy of self-other identification is, for example, the accuracy rate is 80%. Note that the accuracy rate decreases in the method of performing self-other identification using only one of the first target direction enhancement spectrum and the second target direction enhancement spectrum described above. Further, in the method described with reference to FIGS. 3 to 5, the accuracy rate further decreases, and the accuracy rate becomes, for example, 30%.

  Note that the wearer's utterance region may move when the wearer changes the orientation of the face. In this case, the angle θ between the utterance part of the wearer and the first microphone 11 and the second microphone is changed within the movable range of the wearer's face, and the voice of the wearer is changed accordingly. The time difference Δ until reaching the second microphone changes. Therefore, in order to cope with this, several (for example, three) time differences Δ according to the angle θ are obtained, and the time difference Δ when the non-target direction signal spectrum β becomes the smallest is adopted.

<System application examples and host device functions>
In the system of the present embodiment, information related to utterances (hereinafter referred to as utterance information) obtained as described above by a plurality of terminal devices 10 is collected in the host device 20. For example, the host device 20 analyzes the conversation relationship between the wearers using the information obtained from the plurality of terminal devices 10.

FIG. 15 is a diagram illustrating a situation in which a plurality of wearers each wearing the terminal device 10 of the present embodiment are talking. FIG. 16 is a diagram illustrating an example of utterance information of the terminal devices 10A and 10B in the conversation state of FIG.
As shown in FIG. 15, consider a case where two wearers A and B who wear the terminal device 10 are talking. At this time, the voice recognized as the utterance of the wearer in the terminal device 10A of the wearer A is recognized as the utterance of the other person in the terminal device 10B of the wearer B. On the contrary, the voice recognized as the voice of the wearer in the terminal device 10B is recognized as the voice of the other person in the terminal device 10A.

  The utterance information is sent to the host device 20 independently from the terminal device 10A and the terminal device 10B. At this time, the utterance information acquired from the terminal device 10A and the utterance information acquired from the terminal device 10B are opposite in the identification result of the speaker (wearer and other person), as shown in FIG. Information indicating the utterance status such as the length of time and the timing when the speaker is switched is approximated. Therefore, the host device 20 according to the present embodiment compares the information acquired from the terminal device 10A with the information acquired from the terminal device 10B, thereby determining that these information indicate the same utterance situation, and the wearer Recognizing that A and the wearer B are talking. Here, the information indicating the utterance state includes at least the length of the utterance time in each utterance for each utterer described above, the start time and end time of each utterance, and the time (timing) at which the utterer is switched. As described above, the time information related to the utterance is used. Note that in order to determine the utterance situation related to a specific conversation, only part of the time information related to these utterances may be used, or other information may be additionally used.

  By analyzing the conversation relationship between the wearers of the terminal device 10 in this way, it is possible to analyze the communication tendency in the entire group of wearers. Specifically, for example, by examining the correlation between the number of conversation participants, the time during which the conversation was conducted, the degree of interaction, the degree of activity, and the frequency of occurrence of conversation, It is determined whether there is a tendency for the conversation of the aspect to be performed.

<Description of the program>
Note that the processing performed by the voice analysis unit 15 of the terminal device 10 in the present embodiment is realized by cooperation between software and hardware resources. That is, a CPU (not shown) inside the control computer provided in the terminal device 10 executes a program that realizes each function of the terminal device 10 and realizes each of these functions.

Therefore, the processing performed by the voice analysis unit 15 of the terminal device 10 is arranged at different distances from the wearer's utterance part on the computer, and acquires the voice uttered from the wearer's utterance part. A function of acquiring information related to the sound generated by the microphone 12 and a Fourier transform of the information related to the sound generated by the first microphone 11 and the second microphone 12 as a frequency spectrum representing the relationship between frequency and intensity. The signal emitted from the wearer's utterance part is different from the utterance part of the wearer from the function of obtaining the first frequency spectrum and the second frequency spectrum, respectively, and the first frequency spectrum and the second frequency spectrum. First extracted spectrum processed to be larger than the signal emitted from the direction A function of extracting, by a different process and function of extracting first extraction spectrum, and a first frequency spectrum and the second frequency spectrum, the signal emitted from speaking portion of the wearer is speaking portion of the wearer The function of extracting the second extracted spectrum that has been processed to be larger than the signal emitted from a direction different from the above, and the signal emitted from the wearer's utterance region from the first frequency spectrum and the second frequency spectrum were suppressed. The function of extracting the suppression spectrum, subtracting the suppression spectrum from the first extraction spectrum, performing the flooring process to obtain the first enhancement spectrum, and subtracting the suppression spectrum from the second extraction spectrum A function for obtaining two enhancement spectra, and inverse Fourier transform of the first enhancement spectrum and the second enhancement spectrum. A function of calculating the relationship of, can be regarded as a program for actualizing.

  The program for realizing the present embodiment can be provided not only by communication means but also by storing it in a recording medium such as a CD-ROM.

  Although the present embodiment has been described above, the technical scope of the present invention is not limited to the scope described in the above embodiment. It is clear from the description of the scope of the claims that various modifications or improvements added to the above embodiment are also included in the technical scope of the present invention.

  For example, in the above-described example, the self-other identification is determined by the terminal device 10, but is not limited thereto, and may be performed by the host device 20. In the speech analysis system 1 in this embodiment, the processing performed by the speech analysis unit 15 is performed by, for example, the data analysis unit 23 of the host device 20. In the speech analysis system 1, the data analysis unit 23 of the host device 20 functions as a signal analysis device (identification means).

  In the above-described example, the signal arriving from the non-target direction is used as the speech of another person. However, the present invention is not limited to this. For example, noise may be used.

  Further, in the above-described example, the case where the signal is audio is illustrated, but the present invention is not limited to this. For example, the signal may be an electromagnetic wave such as light, infrared rays or radio waves, or a vibration wave such as a seismic wave. In this case, the signal acquisition means is not a microphone but a sensor for capturing these electromagnetic waves and vibration waves.

DESCRIPTION OF SYMBOLS 1 ... Voice analysis system, 10 ... Terminal device, 11 ... 1st microphone, 12 ... 2nd microphone, 15 ... Voice analysis part, 20 ... Host apparatus, 151 ... Voice information acquisition part, 152 ... Frequency spectrum calculation part, 153 ... First target direction signal spectrum extraction unit, 154 ... second target direction signal spectrum extraction unit, 155 ... non-target direction signal spectrum extraction unit, 156 ... target direction enhancement spectrum calculation unit, 157 ... correlation calculation unit, 158 ... Self-other identification part

Claims (12)

Signal information acquisition for acquiring information related to the signals generated by the first signal acquisition means and the second signal acquisition means, which are arranged at different distances from the signal generation source and acquire signals emitted from the signal generation source. And
From the information about each signal generated by the first signal acquisition unit and the second signal acquisition unit, the first frequency spectrum and the second frequency spectrum are represented as frequency spectra representing the relationship between frequency and intensity. A frequency spectrum calculation unit to be obtained;
Extracting from the first frequency spectrum and the second frequency spectrum a first extracted spectrum that is processed such that a signal emitted from the signal generation source is larger than a signal emitted from a direction different from the signal generation source. A first extraction unit that
By a process different from that of the first extraction unit, a signal emitted from the signal generation source from a signal emitted from a direction different from the signal generation source from the first frequency spectrum and the second frequency spectrum. A second extraction unit that extracts a second extracted spectrum that has been processed to increase;
A third extraction unit for extracting, from the first frequency spectrum and the second frequency spectrum, a suppression spectrum in which a signal emitted from the signal generation source is suppressed;
The suppression spectrum is subtracted from the first extracted spectrum, and a first enhancement spectrum is obtained by performing a flooring process, and the second enhancement spectrum is subtracted from the second extraction spectrum. An enhanced spectrum calculation unit for obtaining
A correlation calculating unit that calculates correlation between the first enhanced spectrum and the second enhanced spectrum;
A signal analyzing apparatus comprising: The first extraction unit takes a difference between the first frequency spectrum and the second frequency spectrum, so that a signal emitted from the signal generation source is emitted from a direction different from the signal generation source. The signal analysis apparatus according to claim 1, wherein a larger process is performed. The second extraction unit adds the first frequency spectrum and the second frequency spectrum so that a signal emitted from the signal generation source is larger than a signal emitted from a direction different from the signal generation source. The signal analyzing apparatus according to claim 1, wherein the signal analyzing apparatus performs the following process.   The first signal acquisition means and the second signal acquisition means are arranged so that the signal generation source, the first signal acquisition means, and the second signal acquisition means are substantially in a straight line. The signal analysis device according to claim 1, wherein the signal analysis device is a signal analysis device.   The third extraction unit is caused by a difference in distance between the signal generation source, the first signal acquisition unit, and the second signal acquisition unit with respect to the first frequency spectrum and the second frequency spectrum. 5. The signal analysis apparatus according to claim 1, wherein the suppression spectrum is extracted by adjusting each time difference until the signal arrives and then obtaining each difference. 6.   The second extraction unit is caused by a difference in distance between the signal generation source, the first signal acquisition unit, and the second signal acquisition unit with respect to the first frequency spectrum and the second frequency spectrum. The signal analysis apparatus according to claim 3, wherein the second extracted spectrum is extracted by adding each after adjusting a time difference until the signal arrives.   The time difference is determined in consideration of a change in angle between the signal generation source and the first signal acquisition unit and the second signal acquisition unit, which is generated by the movement of the signal generation source. The signal analysis apparatus according to claim 5 or 6.   The signal analysis apparatus according to claim 1, wherein the signal is a voice. The frequency spectrum calculation unit performs Fourier transform on information about each signal generated by the first signal acquisition unit and the second signal acquisition unit, so that the first frequency spectrum and the second frequency are obtained. Obtain each spectrum,
9. The correlation calculation unit according to claim 1, wherein the correlation calculation unit calculates respective correlations after performing inverse Fourier transform on the first enhancement spectrum and the second enhancement spectrum. The signal analysis apparatus described. From the information about the signals generated by the first signal acquisition means and the second signal acquisition means, which are arranged at different distances from the signal generation source and acquire the signal emitted from the signal generation source, the frequency and the intensity A frequency spectrum calculating unit for obtaining a first frequency spectrum and a second frequency spectrum as frequency spectra representing the relationship;
The first enhanced spectrum and the second enhanced spectrum are separated from the first frequency spectrum and the second frequency spectrum as enhanced spectra in which a target direction signal that is a signal emitted from the signal generation source is enhanced. An enhanced spectrum calculation unit obtained in
A correlation calculating unit that calculates correlation between the first enhanced spectrum and the second enhanced spectrum;
A signal analyzing apparatus comprising: A signal acquisition unit comprising a first signal acquisition unit and a second signal acquisition unit, which are arranged at different distances from the signal generation source and acquire a signal emitted from the signal generation source;
Identifying means for identifying a target direction signal that is a signal emitted from the signal generation source and a non-target direction signal that is a signal other than a signal emitted from the signal generation source;
With
The identification means includes
A signal information acquisition unit for acquiring information related to the signals generated by the first signal acquisition unit and the second signal acquisition unit;
From the information about each signal generated by the first signal acquisition unit and the second signal acquisition unit, the first frequency spectrum and the second frequency spectrum are represented as frequency spectra representing the relationship between frequency and intensity. A frequency spectrum calculation unit to be obtained;
Extracting from the first frequency spectrum and the second frequency spectrum a first extracted spectrum that is processed such that a signal emitted from the signal generation source is larger than a signal emitted from a direction different from the signal generation source. A first extraction unit that
By a process different from that of the first extraction unit, a signal emitted from the signal generation source from a signal emitted from a direction different from the signal generation source from the first frequency spectrum and the second frequency spectrum. A second extraction unit that extracts a second extracted spectrum that has been processed to increase;
A third extraction unit for extracting, from the first frequency spectrum and the second frequency spectrum, a suppression spectrum in which a signal emitted from the signal generation source is suppressed;
The suppression spectrum is subtracted from the first extracted spectrum, and a first enhancement spectrum is obtained by performing a flooring process, and the second enhancement spectrum is subtracted from the second extraction spectrum. An enhanced spectrum calculation unit for obtaining
A correlation calculating unit that calculates correlation between the first enhanced spectrum and the second enhanced spectrum;
A signal analysis system comprising: On the computer,
A function of acquiring information about signals generated by the first signal acquisition unit and the second signal acquisition unit, which are arranged at different distances from the signal generation source and acquire a signal emitted from the signal generation source;
From the information about each signal generated by the first signal acquisition unit and the second signal acquisition unit, the first frequency spectrum and the second frequency spectrum are represented as frequency spectra representing the relationship between frequency and intensity. Each of the functions
Extracting from the first frequency spectrum and the second frequency spectrum a first extracted spectrum that is processed such that a signal emitted from the signal generation source is larger than a signal emitted from a direction different from the signal generation source. Function to
Due to processing different from the function of extracting the first extraction spectrum , the signal generated from the signal generation source is different from the direction of the signal generation source from the first frequency spectrum and the second frequency spectrum. A function of extracting a second extracted spectrum that has been processed to be larger than the emitted signal;
A function of extracting, from the first frequency spectrum and the second frequency spectrum, a suppression spectrum in which a signal emitted from the signal generation source is suppressed;
The suppression spectrum is subtracted from the first extracted spectrum, and a first enhancement spectrum is obtained by performing a flooring process, and the second enhancement spectrum is subtracted from the second extraction spectrum. The function to ask for
A function of calculating a correlation between the first enhanced spectrum and the second enhanced spectrum;
A program that realizes

Images