JP5741175B2 - Confidential data generating device, concealed data generating method, concealing device, concealing method and program - Google Patents

Description

  The present invention relates to a concealed data generation apparatus that generates music data for concealing dialogue voice.

  Dialogue voices exchanged at medical institutions (reception counters at dispensing pharmacies, etc.), consultation counters at financial institutions and insurance companies, interview rooms at law firms, mobile phone counters, restaurants used for dinner, etc. In many cases, personal information that is not desirable to be heard by a person or confidential information of a company is included. However, in the past, there are many facilities where only simple partitioning is used. This is because it is not always easy to install a furniture with a sound insulation function like a karaoke box or to perform interior work due to space and cost constraints of offices and shops. Therefore, there is a demand for a technique for concealing dialogue voices with little modification to the current equipment.

  As one method for concealing sound, there is an active silencing method (ANC: Active Noise Control: see Patent Document 1) that electrically silences, but since the target is limited to stationary noise, However, this method cannot be applied to sounds that change significantly with time.

As another method for concealing the sound, there is a method using BGM (BackGround Music). For example, BGM is often played at shopping centers, cocktail parties, restaurants, and the like. This is intended to mitigate hustle noise using the human auditory masking effect. However, humans also have a completely opposite characteristic to the auditory masking effect, called the cocktail party effect. The cocktail party effect is a selective listening of voice that a conversation of a person who is interested can be naturally heard even when many people are chatting each other like a cocktail party.
Humans work to interpolate partially masked speech with a larger sound source (BGM, etc.) due to the cocktail party effect and listen to the speech of interest, so the speech is completely concealed by normal BGM. It cannot be expected until it becomes. In order to solve such a problem, two methods of (1) energy masking and (2) information masking have been proposed.

  (1) About energy masking, it describes in patent document 2, for example. Patent Document 2 describes that white noise (noise that has a tendency of power to be substantially uniform regardless of frequency in at least an audible range) or the like is flown as a masking sound and the sound is masked by an auditory masking effect. ing.

  (2) Information masking is described in Patent Documents 3 and 4, for example. In Patent Document 3, a sound signal is received from a microphone installed in a certain acoustic space, the received sound signal is scrambled to generate a masking sound, and another acoustic space (a space where the sound signal is not desired to leak). It is described that sound is emitted. Japanese Patent Application Laid-Open No. 2004-228561 describes analyzing dialogue voice recorded in real time, processing the dialogue voice to generate a masking sound, and outputting it.

However, in the method described in Patent Document 2, a masking sound having a high sound pressure flows all the time, and it has been pointed out that it is difficult to hear the chat of people in the waiting room and the conversation during the interview.
Further, in the methods described in Patent Documents 3 and 4, there is a problem that the masking sound gives an unpleasant feeling to humans. In addition, a problem has been pointed out that a microphone for recording, a high-speed signal processing device, and the like are required, which is expensive. In order to relieve the unpleasant masking sound, a method of further synthesizing BGM is conceivable, but another problem arises that the sound pressure increases and becomes troublesome.

In view of this, the present inventors have invented a concealed data generating device that can generate concealed data that is comfortable for humans and has a high concealing effect at low cost (see Patent Document 5). In addition, the inventor has invented a concealed data generation device and the like that can make the masking effect work evenly at any reproduction location of the concealed data without spending labor (see Patent Document 6).
In Patent Document 5 and Patent Document 6, when setting a filter function for emphasizing the masking effect on the sound for the BGM signal, the average value spectrum of the BGM music signal using the maximum value spectrum of a typical sound signal is used. A method for setting a filter function based on the divided value has been proposed.

Japanese Patent No. 2544899 JP 2010-031501 A Japanese Patent No. 42456060 Japanese Patent No. 4336552 Japanese Patent Application No. 2010-192133 Japanese Patent Application No. 2011-000929

By the way, in the methods of Patent Document 5 and Patent Document 6, it is easy to set a filter function so that frequency components of 5 kHz to 10 kHz that contain many human audio signal components are emphasized. The frequency band of 5 kHz to 10 kHz is a region where the sensitivity characteristic of the human auditory system is relatively low. However, if the filter function is used to play music that has been subjected to filter processing, the sensitivity characteristic of the human auditory system. Since the reproduction volume is set based on a frequency band of less than 4 kHz, which is high, the volume in the frequency band of 5 kHz to 10 kHz is remarkably increased, and the timbre may change unnaturally and become troublesome.
The equal loudness curve indicating the sensitivity characteristic of the human auditory system is standardized as ISO226 based on measurement data by Fletcher & Manson et al. The ISO 226 standard is further improved so that a low frequency band of 1 kHz or less matches the sensitivity characteristics of the human auditory system.

  The present invention has been made in view of the above-described problems, and its purpose is to maintain the tone of music to be reproduced equal to the original sound while reducing the volume while enhancing the masking effect on the audio signal. An object of the present invention is to provide a concealed data generation device or the like that can exert a predetermined masking effect even when reproduced.

In order to achieve the above-described object, the first invention is a concealment data generation device for generating concealment data that is music data for concealing a conversational voice, and stores voice data and music data stored in advance. Frequency analysis is performed on each of the voice data, and a voice maximum value spectrum Vv (j) (j is a frequency) which is the maximum spectrum in the time axis direction of the voice data is calculated and averaged in the time axis direction of the music data The frequency analysis means for calculating the music average value spectrum Vm (j), which is the spectrum obtained, and the value based on the voice maximum value spectrum Vv (j) correspond to each other by the value based on the music average value spectrum Vm (j). The division value spectrum Div (j), which is a value divided for each frequency j, is calculated, and each value of the division value spectrum Div (j) corresponds to each other. Filter function creating means for creating a filter function F (j) by multiplying a value based on an auditory sensitivity correction curve L (j) defining a weight of human auditory sensitivity for each frequency j, and the music data is predetermined. Filtering means for generating the concealed data by dividing the divided frames into frames f, and performing Fourier transform on each divided frame f, multiplying by the filter function F (j), and performing Fourier inverse transform, A concealed data generation device characterized by comprising:
According to the first invention, while enhancing the masking effect on the audio signal, the timbre of the music to be reproduced can be maintained at the same level as the original sound, and the predetermined masking effect can be exerted even if the volume is reduced.

The auditory sensitivity correction curve L (j) used by the filter function creating means in the first invention is defined based on, for example, an equal loudness curve of 40 phones.
Forty phones is an average loudness level when listening to normal speech and music, and an appropriate filter function can be created.

Further, the filtering means in the first aspect of the present invention provides the maximum scalar value of the complex spectrum within a predetermined frequency range for the complex spectrum multiplied by the filter function F (j) for each frame f. And further performing a correction for multiplying each element of the complex spectrum by a predetermined scale value within a range in which the scalar value of the element does not exceed the maximum scalar value, and then performing the inverse Fourier transform It is desirable to do.
As a result, it is possible to bring the low-frequency part closer to flatness like white noise while maintaining the state of the discrete spectral characteristics, and further enhance the masking effect while maintaining the timbre of the music. .

Further, the filter function creating means in the first invention replaces the voice maximum value spectrum Vv (jc) (jc is a specific frequency) with a maximum value within a range higher than the frequency jc. Then, a replacement speech maximum value spectrum is calculated, and a replacement music average value spectrum is calculated by replacing the music average value spectrum Vm (jc) with an average value within a range before and after the frequency jc, and the replacement speech maximum value is calculated. A value obtained by dividing the value spectrum by the replacement music average value spectrum is preferably the division value spectrum Div (j).
Since the masking has a property that it tends to work on the high sound side (frequency is on the high frequency side), if the sound maximum spectrum Vv (j) is replaced with the maximum value within the range on the high frequency side than the frequency j, the sound The correction is performed to nonlinearly shift the spectrum to the low frequency side in the frequency direction, and as a result, the masking effect can be enhanced.

In addition, the filter function creating means in the first invention multiplies each value of the filter function F (j) by a value based on the auditory sensitivity correction curve L (j), and then a range before and after the frequency j. It is desirable to smooth the filter function F (j) by replacing it with an average value.
As a result, the filter function becomes smooth, and as a result, the concealment data finally generated becomes music data comfortable for humans.

In the first aspect of the present invention, the frequency analysis means uses, as the music average spectrum Vm (f, j), a spectrum obtained by averaging a spectrum averaged in the time axis direction over M frames before and after each frame f of the music data. The filter function creating means calculates a value based on the voice maximum value spectrum Vv (j) as the divided value spectrum Div (f, j) as the division value spectrum Div (f, j), and the music average value spectrum Vm corresponding to the frame f. A value obtained by dividing each frequency j corresponding to each other by a value based on (f, j) is calculated. Further, for each value of the divided value spectrum Div (f, j), for each frequency j corresponding to each other, It is desirable to create the filter function F (f, j) by multiplying a value based on the auditory sensitivity correction curve L (j).
As a result, it is possible to generate concealment data in which the masking effect works evenly at any reproduction location without spending manpower.

According to a first aspect of the present invention, there is provided music data storage means for storing a plurality of music data, and music data selection for selecting a single music data from the music data stored by the music data storage means And means for generating the concealment data based on the single music data selected by the music data selection means.
Thereby, a plurality of concealment data can be generated based on a plurality of music data.

2nd invention is the concealment data storage means which memorize | stores the said some concealment data which the concealment data production | generation apparatus of 1st invention produces | generates, The said concealment data memorize | stored by the said concealment data storage means A concealment data selection unit that selects the single concealment data from among the data, and a concealment data reproduction unit that reproduces the single concealment data selected by the concealment data selection unit This is a concealment device.
According to the second invention, the concealed data generating device of the first invention can be physically separated, and the concealed data created in advance can be obtained without operating the concealed data generating device of the first invention. It can be played at any time.

The concealed data reproducing means in the second invention is preferably constituted by a flat speaker having a mechanism for uniformly radiating the concealed data from a predetermined plane as a sound wave having a wavefront close to a plane wave.
As a result, the energy amount of the sound wave attenuated in the process of being propagated to the concealment target position is smaller in the concealed data than in the conversation voice, and the energy amount of the concealed data is relatively larger than that in the conversation voice. Therefore, the masking effect can be enhanced.

A third invention is a concealment data generation method for generating concealment data, which is music data for concealing dialogue voice, and performs frequency analysis on each of prestored voice data and music data. And calculating a voice maximum value spectrum Vv (j) (j is a frequency) which is a maximum spectrum in the time axis direction of the voice data, and a music average spectrum which is a spectrum averaged in the time axis direction of the music data A frequency analysis step for calculating Vm (j) and a value obtained by dividing a value based on the maximum speech spectrum Vv (j) for each corresponding frequency j by a value based on the music average spectrum Vm (j). A certain division value spectrum Div (j) is calculated, and further, for each value of the division value spectrum Div (j), a human listening is performed for each frequency j corresponding to each other. A filter function creating step for creating a filter function F (j) by multiplying a value based on the auditory sensitivity correction curve L (j) defining the sensitivity weight, and a frame f that is a unit of a predetermined section of the music data. A filtering step for generating the concealed data by performing Fourier transform on each of the divided frames f, multiplying by the filter function F (j), and performing Fourier inverse transform. This is a method for generating confidential data.
According to the third aspect of the invention, the masking effect on the audio signal can be enhanced, the timbre of the music to be reproduced is maintained at the same level as the original sound, and the predetermined masking effect can be exerted even when the volume is reduced.

4th invention is the concealment data storage step which memorize | stores the said some concealment data produced | generated by the concealment data generation method of 3rd invention, The said concealment data memorize | stored by the said concealment data storage step Including a concealment data selection step for selecting a single concealment data from among the above, and a concealment data reproduction step for reproducing the single concealment data selected by the concealment data selection step. It is the concealment method characterized by this.
According to the fourth invention, the concealed data generation method of the second invention can be physically separated, and the concealed data created in advance can be obtained without using the concealed data generation method of the second invention. It can be played at any time.

According to a fifth aspect of the present invention, frequency analysis is performed on each of voice data and music data stored in advance in a computer, and a voice maximum value spectrum Vv (j) (which is a maximum spectrum in the time axis direction of the voice data). j is a frequency), a frequency analysis step of calculating a music average value spectrum Vm (j) that is a spectrum averaged in the time axis direction of the music data, and a value based on the voice maximum value spectrum Vv (j) Is divided by a frequency j corresponding to each other by a value based on the music mean value spectrum Vm (j), and a divided value spectrum Div (j) is calculated. By multiplying the value by a value based on the auditory sensitivity correction curve L (j) defining the weight of the human auditory sensitivity for each frequency j corresponding to each other, A filter function creating step for creating a filter function F (j), the music data is divided into frames f which are predetermined section units, each of the divided frames f is Fourier transformed, and the filter function F (j) is A computer-readable program for executing the filtering step of generating the concealment data by multiplication and inverse Fourier transform.
By installing the fifth invention on a general-purpose computer, the concealed data generating apparatus of the first invention or the concealed data generating method of the third invention can be realized on a general-purpose computer.

  According to the present invention, a component of a frequency band with low human auditory sensitivity in BGM music is emphasized due to the filter processing, so that setting the playback volume to a high level is suppressed, and the timbre changes unnaturally. It is possible to avoid the masking effect equivalent to or higher than the conventional one while suppressing the playback volume of the BGM music more than the conventional one. That is, it is possible to improve the concealment effect in a comfortable acoustic environment with a lower reproduction volume than before.

Overview of the concealment device Hardware configuration diagram of the concealment data generation device Diagram showing an example of equal loudness curves Diagram showing an example of auditory sensitivity correction curve Flow chart showing the flow of concealment processing Diagram showing the flow of concealment data generation processing Diagram explaining frequency analysis processing (1) Diagram explaining frequency analysis processing (2) FIG. (1) explaining filter function creation processing FIG. (2) explaining the filter function creation processing FIG. 3 illustrates the filter function creation process (3) Diagram explaining filtering process (1) Diagram for explaining the filtering process (2) Fig. 3 illustrates filtering processing. First installation example of concealment device Second installation example of concealment device The figure which shows the audio | voice maximum value spectrum of an Example and a comparative example. The figure which shows the music average value spectrum of an Example and a comparative example The figure which shows the filter function of the comparative example The figure which shows the music signal after the filtering process of a comparative example The figure which shows the auditory sensitivity correction curve of an Example The figure which shows the filter function of an Example The figure which shows the music signal after the filtering process of an Example (no compression) The figure which shows the music signal after the filtering process (with compression) of an Example

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a schematic diagram of the concealment device 1. As shown in FIG. 1, the concealment device 1 includes at least a concealment data generation device 2 and a music playback device 3.
The concealment data generation device 2 is, for example, a computer or the like, and generates concealment data 7 that is music data for concealing the dialogue voice. The storage unit of the concealed data generation device 2 stores at least audio data 4, music data 5, and auditory sensitivity correction curve 6. These data will be described later.
The music playback device 3 is composed of a music player and a speaker, and plays back the concealment data 7. The storage unit of the music playback device 3 stores at least concealment data 7 generated by the concealment data generation device 2.

The concealment device 1 can take various configurations depending on the application. As shown in FIG. 1, the concealment data generation device 2 and the music playback device 3 that constitute the concealment device 1 may be different housings or a single housing.
Further, the concealed data generation device 2 and the music playback device 3 may be connected by wire as shown in FIG. 1, may be connected wirelessly, may be connected via a network, It does not have to be connected.
When the concealment data generation device 2 and the music playback device 3 are not connected, the concealment data generation device 2 reads the concealment data 7 by a storage medium (a computer such as a CD, MD, USB memory, or SD card and a music player). The music player 3 inputs the concealment data 7 from the storage medium.

At least the music playback device 3 is installed in an acoustic space where it is desired to conceal the dialogue voice. As such an acoustic space, for example, a waiting room adjacent to a reception counter such as a dispensing pharmacy can be considered. Then, the music playback device 3 plays back the concealment data 7 in such a waiting room.
Here, the concealment data 7 generated by the concealment data generation device 2 according to the embodiment of the present invention is such that even if there is no partition between the reception counter and the waiting room, a person who is in the waiting room has a normal volume level. It is possible to conceal it to such an extent that the content of the dialogue voice of the reception counter cannot be heard.
Other acoustic spaces in which the music playback device 3 is installed include waiting spaces adjacent to counters such as financial institutions, insurance companies and mobile phone stores, corridors adjacent to interview rooms such as law offices, and reception areas such as companies Private rooms such as rooms and restaurants.

FIG. 2 is a hardware configuration diagram of the concealed data generation device 2. Note that the hardware configuration in FIG. 2 is an example, and various configurations can be adopted depending on the application and purpose.
The concealed data generation apparatus 2 includes a control unit 21, a storage unit 22, a media input / output unit 23, a communication control unit 24, an input unit 25, a display unit 26, a peripheral device I / F unit 27, etc. via a bus 28. Connected.

The control unit 21 includes a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), and the like.
The CPU calls and executes a program stored in the storage unit 22, ROM, recording medium or the like to a work memory area on the RAM, drives and controls each device connected via the bus 28, and provides a concealed data generation device 2 to realize the processing described later.
The ROM is a non-volatile memory, and permanently stores a boot program of the concealed data generation device 2, a program such as BIOS, data, and the like.
The RAM is a volatile memory, and temporarily stores a program, data, and the like loaded from the storage unit 22, ROM, recording medium, and the like, and includes a work area used by the control unit 11 to perform various processes.

The storage unit 22 is an HDD (hard disk drive), and stores a program executed by the control unit 21, data necessary for program execution, an OS (operating system), and the like. With respect to the program, a control program corresponding to an OS (operating system) and an application program for causing a computer to execute processing described later are stored.
Each of these program codes is read by the control unit 21 as necessary, transferred to the RAM, read by the CPU, and executed as various means.

The media input / output unit 23 (drive device) inputs / outputs data, for example, a CD drive (-ROM, -R, -RW, etc.), DVD drive (-ROM, -R, -RW, etc.), MD drive, etc. And other media input / output devices.
The communication control unit 24 has a communication control device, a communication port, and the like, and is a communication interface that mediates communication between the concealed data generation device 2 and the network, and performs communication control between other devices via the network. Do. The network may be wired or wireless.

The input unit 25 inputs data and includes, for example, a keyboard, a pointing device such as a mouse, and an input device such as a numeric keypad.
An operation instruction, an operation instruction, data input, and the like can be performed on the concealed data generation apparatus 2 via the input unit 25.
The display unit 26 includes a display device such as a CRT monitor and a liquid crystal panel, and a logic circuit (such as a video adapter) for realizing the video function of the concealed data generation device 2 in cooperation with the display device.

The peripheral device I / F (interface) unit 27 is a port for connecting the peripheral device to the concealed data generation device 2, and the concealed data generation device 2 is connected to the peripheral device via the peripheral device I / F unit 27. Send and receive data. The peripheral device I / F unit 27 is configured by a USB, an SD card reader, or the like.
The bus 28 is a path that mediates transmission / reception of control signals, data signals, and the like between the devices.

FIG. 3 is a diagram illustrating an example of an equal loudness curve. The equal loudness curve is standardized by ISO226. An equal loudness curve is a change in the physically measured sound pressure level at which the human is audibly audibly heard at the same loudnell level based on the change in frequency for each level of loudness. It is a curve which shows. The unit of the loudness level is phon (phone, phone, horn). The unit of the sound pressure level is dB (decibel).
In FIG. 3, the horizontal axis represents frequency [Hz] and the vertical axis represents sound pressure level [dB], and an equal loudness curve is defined for each loudness level. In FIG. 3, an equal loudness curve of 0 (minimum audible sound field), 10, 20, 30,..., 130 [phon] is illustrated.

As can be seen from FIG. 3, as the loudness level increases, the difference between the maximum sound pressure level and the minimum sound pressure level for each equal loudness curve decreases. That is, the difference between the maximum sound pressure level and the minimum sound pressure level in the equal loudness curve of 0 [phon] is the largest, and the difference between the maximum sound pressure level and the minimum sound pressure level in the equal loudness curve of 130 [phon] is Smallest.
In the embodiment of the present invention, an “audience sensitivity correction curve” to be described later is defined using an equal loudness curve of 40 [phon], which is an average loudness level when listening to normal speech and music. If the environment of the acoustic space where the concealment data 7 is reproduced can be predicted to some extent, an equal loudness curve may be selected according to the environment.

FIG. 4 is a diagram illustrating an example of an auditory sensitivity correction curve. The auditory sensitivity correction curve is used by the concealment data generation device 2. The auditory sensitivity correction curve is used when correcting a “filter function” described later.
In FIG. 4, the upper part for each frequency indicates the sound pressure level of the equal loudness curve, and the lower part for each frequency indicates the sound pressure level of the auditory sensitivity correction curve when 500 Hz is used as a reference (0 dB). . For example, for a frequency of 20 [Hz], the sound pressure level of the equal loudness curve is 90.0 [dB], and the sound pressure level of the auditory sensitivity correction curve is −53.0 [dB]. For example, for a frequency of 30 [Hz], the sound pressure level of the equal loudness curve is 77.0 [dB], and the sound pressure level of the auditory sensitivity correction curve is −40.0 [dB].
In the example shown in FIG. 4, “Sound pressure level of auditory sensitivity correction curve (lower value) = minimum value of sound pressure level of equal loudness curve at 500 Hz (= 37.0) −sound pressure level of equal loudness curve (upper row). The sound pressure level of the auditory sensitivity correction curve is obtained by “
The auditory sensitivity correction curve calculation process is not limited to this example. For example, the auditory sensitivity correction curve may be obtained by folding the equal loudness curve of FIG. 3 according to a predetermined straight line parallel to the horizontal axis. Further, the frequency used as a reference for the auditory sensitivity correction curve does not need to be set to 500 Hz which is a minimum value on the equal loudness curve.

Here, the significance of the auditory sensitivity correction curve will be described.
The sensitivity of the human auditory system changes depending on the frequency, and has a characteristic that the sensitivity decreases as a low frequency of 300 Hz or lower or a high frequency of 5 kHz or higher reaches a peak around 4 kHz. However, since the audio signal includes many relatively low frequency band components of 5 kHz to 10 kHz in the music signal, the filter function works to emphasize these frequency band components. Since the emphasized frequency band is a band where the human auditory sensitivity is relatively low, the volume is set based on a frequency band of 4 kHz or less where the auditory sensitivity is high during reproduction. Then, in conjunction with this, the volume of the frequency band of 5 kHz to 10 kHz is unnaturally increased, and noise is generally generated. Therefore, as described later, by generating a filter function by superimposing an auditory characteristic curve, it is possible to prevent the degree of emphasis in the frequency band of 5 kHz to 10 kHz from being suppressed and resulting in an unnatural timbre. In addition, the frequency band of 300 Hz to 3.4 kHz, which has a high human auditory sensitivity and is rich in formant components for identifying speech, is emphasized, and masking can easily work effectively without increasing the playback volume.

FIG. 5 is a flowchart showing the flow of concealment processing.
As illustrated in FIG. 5, the control unit 21 of the concealed data generation device 2 stores the audio data 4 and the music data 5 in the storage unit 22 (S101). A plurality of music data 5 may be stored.
The voice data 4 is not a conversation voice in the acoustic space to be concealed, but fixed sample data. That is, the concealment data generation device 2 according to the embodiment of the present invention does not use the conversation target conversation voice sampled in real time. The voice data 4 is a dialogue voice in which various male voices and female voices recorded in advance are mixed.
The music data 5 is arbitrary. For example, it is not always necessary to include melodies, rhythms, and progression of harmony that are meaningful to the listener, and natural sounds such as river murmurs may be used. Music data that contains many frequency components similar to the conversational speech to be concealed can easily work with a masking effect. Therefore, it is desirable to include vocal music data in the sense of enhancing the masking effect. However, since voice noise data is noisy, it is realistic to use chamber music with only instrumental data and less instrumentation. The concealment data generation device 2 generates concealment data 7 for each music data 5.

Next, the control part 21 of the concealment data generation apparatus 2 selects the single music data 5 (S102). Selection of the music data 5 may be instructed by the user via the input unit 25.
Next, the control unit 21 of the concealed data generation device 2 performs concealment data 7 generation processing based on the single music data 5 selected in S102 (S103). Details of the process for generating the concealment data 7 will be described later.
A plurality of the concealment data 7 may be generated by repeating the processes of S102 and S103.

Next, the music playback device 3 stores the concealment data 7 generated in S103 (S104). A plurality of concealment data 7 may be stored.
Next, the music playback device 3 selects a single anonymized data 7 (S105). The selection of the concealment data 7 is generally performed automatically based on a playlist (reproduction program) defined in advance, but may be instructed by the user.
Next, the music playback device 3 plays back the single anonymized data 7 selected in S105 (S106). The reproduction volume is appropriately changed according to a user instruction in accordance with a change in environment.

As described above, the concealment device 1 can conceal the conversation voice in the acoustic space A so that it is not heard by a person in the acoustic space B separated by a predetermined distance.
Below, the detail of the production | generation process of the concealment data 7 is demonstrated.

FIG. 6 is a diagram showing the flow of the concealment data generation process. As shown in FIG. 6, the concealment data generation process includes a frame extraction process 31, a frequency analysis process 32, a filter function creation process 33, and a filtering process 34.
Here, an outline of each process will be described, and details will be described later.

  The frame extraction process 31 receives the audio data 4 and the music data 5 and divides them into frames f of a predetermined section for each to generate the audio frame group 10 and the music frame group 11.

The frequency analysis process 32 receives the audio frame group 10 and the music frame group 11 and outputs the audio maximum value spectrum data 12 and the music average value spectrum data 13. In the frequency analysis process 32, the control unit 21 of the concealed data generation device 2 performs frequency analysis on each claim of the audio frame group 10 and the music frame group 11, and is the maximum spectrum in the time axis direction of the audio frame. This is a process of calculating a single voice maximum value spectrum Vv (j) (j is a frequency) and calculating a music average value spectrum Vm (j) which is a spectrum averaged in the time axis direction of the music frame.
Note that the subscript “v” of Vv (j) is an initial of voice. Further, the subscript “m” of Vm (f, j) is an initial of music.

  Further, the frequency analysis processing 32 may output the music average value spectrum data 13 for each frame f. That is, the control unit 21 of the concealed data generation device 2 calculates a spectrum averaged in the time axis direction over the M frames before and after the music frame (M) as the music average value spectrum Vm (f, j). You may do it.

  Here, as for M, for example, “M (number) × frame length (second)” is preferably about several seconds. This is because if “M (pieces) × frame length (seconds)” is too short, the music will sound unnatural, and if “M (pieces) × frame length (seconds)” is too long, This is because the masking effect, that is, the part where the voice concealment does not work properly becomes conspicuous.

Since the audio data 4 has a large spectrum time series variation and includes a silent part, the average value cannot be appropriately evaluated. Therefore, in the embodiment of the present invention, only one voice maximum value spectrum Vv (j) is calculated.
The music data 5 is replaced with a music average value spectrum Vm (j) obtained by averaging the instantaneous spectrum in the time axis direction for each instantaneous spectrum in units of frames (the energy amount ignoring the phase component). Alternatively, the music data 5 is replaced with a music average value spectrum Vm (f, j) obtained by averaging instantaneous spectra corresponding to a predetermined number of frames before and after every frame f.

The filter function creation process 33 receives the maximum speech spectrum data 12 and the average music spectrum data 13 and outputs the filter function data 14 for each frame f. In the filter function creation process 33, the control unit 21 of the concealed data generation device 2 sets a value based on the voice maximum value spectrum Vv (j) for each frequency j corresponding to each other by a value based on the music average value spectrum Vm (j). Divide value spectrum Div (j), which is a value divided by 2, and further, auditory sensitivity defining weights of human auditory sensitivity for each frequency j corresponding to each value of divided value spectrum Div (j) In this process, a filter function F (j) is created by multiplying a value based on the correction curve L (j). Here, the unit of the auditory sensitivity correction curve L (j) is not a dB shown in FIG. 4 but a dimensionless value. Specifically, if the dB value shown in FIG. 4 is d, 10 ^{d / 20.} Given in.

  Further, the filter function creation processing 33 may output the filter function data 14 for each frame f. That is, the control unit 21 of the concealed data generation device 2 uses the value based on the maximum speech value spectrum Vv (j) as the divided value spectrum Div (f, j) as the music average value spectrum Vm (f , j) is calculated by dividing each frequency j corresponding to each other by a value based on, j), and for each value of the divided value spectrum Div (f, j), an auditory sensitivity correction curve for each frequency j corresponding to each other. The filter function F (f, j) may be created by multiplying a value based on L (j).

  The filtering process 34 receives the music data 5 and the filter function data 15 and outputs the concealment data 7 for each frame f. In the filtering process 34, the control unit 21 of the concealed data generation device 2 divides the music data 5 into frames f that are predetermined intervals, and performs Fourier transform on each of the divided frames f to obtain a filter function F (j). Is the process of generating the concealment data 7 by performing Fourier inverse transform.

  In the following, a case where the frequency analysis process 32 and the filter function creation process 33 create the music average spectrum Vm (f, j) and the filter function F (f, j) for each frame f will be described as an example. To do. By explaining this example, the case where the frequency analysis process 32 and the filter function creation process 33 create the music average value spectrum Vm (j) and the filter function F (j) instead of every frame f will be explained. Needless to say.

  7 and 8 are diagrams for explaining the frequency analysis processing. As shown in FIGS. 7 and 8, the frequency analysis process 32 includes a (narrow sense) frequency analysis 32 a, an instantaneous spectrum calculation process 41, and an average spectrum calculation process 42.

First, frequency analysis processing for the audio data 4 will be described.
For example, the sampling frequency Fs is “44100 Hz”, and the number of samples N is “4096”. The number of frames Fv included in the audio data 4 is determined by the sampling frequency Fs and the number of samples N.
In the frame extraction process 31, the control unit 21 of the concealed data generation device 2 performs N / 2 samples on each of the monaural audio signal having the sampling frequency Fs (the sum of LR (left and right) in the case of stereo). For each interval (ie, overlap by N / 2 samples), each N Fv frames are extracted.

Next, in the frequency analysis process 32a, the control unit 21 extracts the f-th frame data Xv (f, i) (f = 0,..., Fv-1; i = 0,. For 1), Fourier transformation is performed using the Hanning window function H (i) = 0.5−0.5 cos (2πi / N).
Next, the control unit 21 calculates the real part Av (f, j) (f = 0,..., Fv−1; j = 0,..., N−1) of the conversion data, and the imaginary part Bv (f , J) (f = 0,..., Fv-1; j = 0,..., N-1) and the time-series maximum value spectrum Vv (j) of the intensity values, respectively, as calculate.

  In FIG. 7, the frequency analysis 32a is performed on the frame 1 (corresponding to f = 0) to the frame F (corresponding to f = Fv−1) of the audio frame data Xv (f, i). ~ The voice spectrum F is calculated and the voice maximum value spectrum Vv (j) is calculated.

Next, frequency analysis processing for music data 5 will be described.
As with the audio data 4, the sampling frequency Fs is “44100 Hz” and the number of samples N is “4096”. The number of frames Fm included in the music data 5 is determined by the sampling frequency Fs and the number of samples N.
In the frame extraction process 31, the control unit 21 of the concealment data generation device 2 performs N / 2 samples for each monaural music signal having a sampling frequency Fs (the sum of LR (left and right) in the case of stereo). For each interval (ie, overlap by N / 2 samples), N Fm frames are extracted.

Next, in the frequency analysis process 32a, the control unit 21 extracts the f-th frame data Xm (f, i) (f = 0,..., Fm−1; i = 0,. For 1), Fourier transformation is performed using the Hanning window function H (i) = 0.5−0.5 cos (2πi / N).
Next, as the instantaneous spectrum calculation process 41, the control unit 21 calculates an instantaneous spectrum that is an energy amount ignoring the phase component for each frame. Moreover, the control part 21 calculates the average spectrum which is an average value of the instantaneous spectrum of M frames (M pieces) before and after as an average spectrum calculation process 42.

Specifically, the control unit 21 calculates the real part Am (f, j) (f = 0,..., Fm−1; j = 0,..., N−1) of the converted data, and the imaginary part Bm. (F, j) (f = 0,..., Fm-1; j = 0,..., N-1) and M / 2 frames before and after the target frame as a midpoint (M / 2) Average spectrum Vm (f, j) (f = 0,..., Fm-1; j = 0,..., N / 2) of M frames (M) (M <Fm) in total. Each is calculated as follows:
However, in the case of the top portion of the music data 5, that is, when f <M / 2, it is impossible to take the average of M / 2 frames before and after (M / 2 pieces), so Vm (f, j) = Vm ( M / 2, j). Similarly, in the case of the rear part of the music data 5, that is, when f> Fm−M / 2, it is impossible to take the average of M / 2 frames before and after (M / 2 pieces), so Vm (f, j) = Vm (Fm-M / 2-1, j).

FIG. 7 shows frequency analysis processing for the frame f and the frame f + 1 of the music data 5 as an example.
In FIG. 7, the frequency analysis 32a is performed on the frames 1 to M + 1 of the music frame data Xm (f, i), the time series average from the frames 1 to M is calculated, and the music average for the frame f is calculated. It is illustrated that a value spectrum Vm (f, j) is calculated. Similarly, FIG. 7 illustrates that the time series average from frame 2 to frame M + 1 is calculated, and the music average value spectrum Vm (f + 1, j) for frame f + 1 is calculated.

  In addition, FIG. 8 illustrates, as a supplementary explanation of FIG. 7, that the instantaneous spectrum is calculated for each frame by the instantaneous spectrum calculation process 41 using the music data 5 as an input. Also, the average spectrum calculation process 42 calculates the average value of the instantaneous spectrum of the preceding and following M frames (M) for the frame to be processed, replaces it with the average value spectrum, and outputs the music average value spectrum data 13. It is shown in the figure.

  9 to 11 are diagrams illustrating the filter function creation process. The filter function creation process 33 includes a critical bandwidth correction process 43 shown in FIG. 9, a division process 44 shown in FIG. 10, and an auditory sensitivity correction process 45 and a smoothing process 46 shown in FIG. 11.

First, the critical bandwidth correction processing 43 will be described with reference to FIG.
In the critical bandwidth correction processing 43, the control unit 21 of the concealed data generation device 2 replaces the voice maximum value spectrum Vv (j) with a maximum value within a predetermined range for each frequency j. This is a process for creating a replacement speech maximum value spectrum Vv ′ (j). Also, the critical bandwidth correction processing 43 replaces the music average value spectrum Vm (f, j) for each frame f with an average value within a predetermined range for each frequency j, thereby replacing the replacement music average value spectrum Vm. This is a process for creating '(f, j).
FIG. 9 shows critical bandwidth correction processing for frame f and frame f + 1 as an example.

  The critical bandwidth is a frequency range (referred to as a critical bandwidth, Bark) over which masking extends around a frequency component Vv (j) or Vm (f, j) of a certain frequency j. As an approximate expression of the critical bandwidth, E.I. The Zwicker equation is known. In general, it has been found that the critical bandwidth increases as the frequency increases.

  The unit of fr in Equation (7) is also “Hz”. When fr and Bz (fr) are converted into the dimension of the number of points of Fourier transform in the present embodiment, the following expression is obtained.

  In the critical bandwidth correction processing 43, the control unit 21 of the concealed data generation device 2 converts the frequency component Vv (j) to jc = j− (1−α) × Bz () for each frequency j with respect to the audio signal spectrum. j) to jc = j + α × Bz (j). That is, the control unit 21 calculates the replaced spectrum (replacement speech maximum value spectrum) Vv ′ (j) for j = 0,..., N / 2 as the following equation.

α is a real number from 0 to 1, and usually α = 1.0. According to Expression (9), correction is performed to nonlinearly shift the voice spectrum in the frequency direction to the low frequency side.
Since the masking has a property that it tends to work on the high sound side (frequency is on the high frequency side), if the sound maximum spectrum Vv (j) is replaced with the maximum value within the range on the high frequency side than the frequency j, the sound The correction is performed to nonlinearly shift the spectrum to the low frequency side in the frequency direction, and as a result, the masking effect can be enhanced.

  On the other hand, for the music signal spectrum, the control unit 21 performs processing for each frame f, and changes the frequency component Vm (f, j) for each frequency j from jc = j−0.5 × Bz (j) to jc. = J + 0.5 × Bz (j) Replace with an average value. That is, the control unit 21 calculates the replaced spectrum (replacement music average value spectrum) Vm ′ (f, j) for j = 0,..., N / 2 as follows:

  According to the equation (10), the music spectrum is smoothed in the frequency direction.

  In FIG. 9, W (j) represents the calculation range at the time of replacement. It is illustrated that a single replacement speech maximum value spectrum Vv ′ (j) is calculated for the speech maximum value spectrum Vv (j). A replacement music average value spectrum Vm ′ (f, j) is calculated for the music average value spectrum Vm (f, j), and a replacement music is calculated for the music average value spectrum Vm (f + 1, j). It is shown that the average value spectrum Vm ′ (f + 1, j) is calculated.

Next, the division process 44 will be described with reference to FIG.
In the division processing 44, the control unit 21 of the concealed data generation device 2 uses a frequency based on a value based on the audio maximum value spectrum Vv (j) for each frame f by a value based on the music average value spectrum Vm (j). This is a process of calculating a value divided for each j as a divided value spectrum Div (f, j). In particular, for each frame f, the control unit 21 obtains a value obtained by dividing the replacement speech maximum value spectrum Vv ′ (j) by the replacement music average value spectrum Vm ′ (f, j) as a divided value spectrum Div (f, j). It is desirable to do.
FIG. 10 shows division processing for frame f and frame f + 1 as an example.

Next, the auditory sensitivity correction process 45 and the smoothing process 46 will be described with reference to FIG.
The auditory sensitivity correction processing 45 multiplies each value of the divided value spectrum Div (f, j) by a value based on the auditory sensitivity correction curve L (j) for each corresponding frequency j, thereby correcting the divided value. This is a process for creating a spectrum Div ′ (f, j).
FIG. 11 shows an auditory sensitivity correction process for frame f and frame f + 1 as an example.

Specifically, the control unit 21 is based on, for example, a 40 phon equal loudness curve for each value of the division value spectrum Div (f, j) for each frequency (j = 0,..., N / 2). Is multiplied by a value based on the auditory sensitivity correction curve L (j) defined as follows.
For example, the control unit 21 substitutes the value shown in the lower part of FIG. 4 for the variable dB, calculates 10 ^{dB / 20} as the magnification value, and multiplies this magnification value.

As shown in the example of FIG. 4, the auditory sensitivity correction curve L (j) is expressed by “Sound pressure level of auditory sensitivity correction curve = 500 Hz minimum value of sound pressure level of equal loudness curve−Sound pressure level of equal loudness curve”. You may ask.
Further, the sign of the auditory sensitivity correction curve L (j) may be inverted by appropriately offsetting the equal loudness curve to the minus side. The reason for adding the offset is that simply inverting the sign will amplify the waveform amplitude.
The control unit 21 may multiply the auditory sensitivity correction curve L (j) a plurality of times.

  In the human auditory organ, it is considered that processing similar to the auditory sensitivity correction processing 45 is performed on speech and music. Therefore, the concealment data 7 generated by the control unit 21 multiplying the auditory sensitivity correction curve L (j) once, the auditory sensitivity correction curve L (j) is once in the human auditory organ that listens to the data. It is considered to be multiplied. That is, in total, the concealment data 7 is considered to be heard by a human being multiplied by the auditory sensitivity correction curve L (j) twice. On the other hand, since the conversational speech to be concealed is also considered to be multiplied by the auditory sensitivity correction curve L (j) once in the auditory organ, the concealment data 7 is superior to the conversational speech by one extra multiplication. Will work.

Further, the smoothing process 46 is performed by the control unit 21 of the concealed data generation device 2 replacing the corrected division value spectrum Div ′ (f, j) with an average value within a range before and after the frequency j. This is a process of smoothing the division value spectrum Div ′ (f, j).
FIG. 11 shows a smoothing process for the frame f and the frame f + 1 as an example.

  Specifically, the control unit 21 performs, for each frequency (j = 0,..., N / 2), a predetermined tap number T (<N) with respect to the corrected division value spectrum Div ′ (f, j). / 2), let F (f, j) be the result of applying a smoothing filter as in the following equation.

β is a proportionality constant (real value) for adjusting the sound pressure. If the sound pressure of the audio signal is the same as the sound pressure of the music signal, β = 1.0.
An upper limit value and a lower limit value of F (f, j) are set in advance. For example, if the median value is 1, the upper limit value is 10 times “10” and the lower limit value is 1/10 “0.1”. When the division result exceeds the upper limit value or falls below the lower limit value, the control unit 21 sets an upper limit value or a lower limit value for F (f, j), respectively.

As shown in FIG. 11, the corrected division value spectrum Div ′ (f, j) is a function having many extreme values (maximum values and minimum values). In particular, there are places where division by 0 occurs in some places, and at those places, extreme values having an upper limit value become discontinuous points. If the corrected division value spectrum Div ′ (f, j) is directly used as a filter function, concealment data 7 that is difficult to hear for humans is generated. Therefore, in the embodiment of the present invention, the smoothing process 46 is performed.
As shown in FIG. 11, by performing the smoothing process 46, the filter function F (f, j) is a smooth function with few extreme values.

  Note that the concealment data 7 that is easier to hear for humans can be generated by performing the smoothing process 46 after the auditory sensitivity correction process 45 than performing the auditory sensitivity correction process 45 after the smoothing process 46.

12-14 is a figure explaining a filtering process. The filtering process 34 includes a Fourier transform process 47 and a filter function multiplication process 48 shown in FIG. 12, and a frequency dimension compression process 49 and a Fourier inverse transform process 50 shown in FIG.
In the frequency analysis process 32 and the filter function creation process 33 described above, calculation is performed on a real value, but in the filtering process 34, calculation is performed on an instantaneous spectrum having a complex value.

In the Fourier transform processing 47, the control unit 21 of the concealment data generation device 2 performs music frame data Xml (f, i) and Xmr (f, i) (f = 0,..., Fm−1; i = 0 ,..., N-1) is Fourier transformed to calculate a source complex spectrum.
FIG. 12 shows, as an example, Fourier transform processing for frame f and frame f + 1.

  In the Fourier transform processing 47, the control unit 21 performs a sampling frequency Fs stereo audio signal (one is set to 0 in the case of a monaural signal) at each N / 2 sample interval (that is, N / 2 samples). And Hanning window function H (i) = 0.5 with respect to f-th music frame data Xml (f, i) and Xmr (f, i) extracted by N Fm frames each. Fourier transform is performed using −0.5 cos (2πi / N), and real parts Aml (f, j) and Amr (f, j) and imaginary part Bml (f, j) of the transformed data are obtained as follows. ) And Bmr (f, j) (f = 0,..., Fm) −1; j = 0,.

The filter function multiplication process 48 is a process in which the control unit 21 calculates a modified complex spectrum by multiplying the source complex spectrum by the filter function F (f, j).
FIG. 12 shows filter function multiplication processing for frame f and frame f + 1 as an example.

  In the filter function multiplication process 48, the control unit 21 multiplies all frequency components in a predetermined frequency section [j1, j2] for each frame f by using Fm filter functions F (f, j). That is, the control unit 21 performs conversion according to the following expression in each frame f = 0,..., Fm−1 and each frequency j = j1,.

Next, the frequency dimension compression process 49 and the inverse Fourier transform process 50 will be described with reference to FIG.
For each frame f, the frequency dimension compression processing 49 obtains a maximum scalar value of the complex spectrum within a predetermined frequency range for the complex spectrum multiplied by the filter function F (j). This is a process of calculating a re-modified complex spectrum by applying correction for multiplying each element by a predetermined scale value within a range in which the scalar value of the element does not exceed the maximum scalar value.
FIG. 13 shows frequency dimension compression processing for frame f and frame f + 1 as an example.

In the frequency dimension compression processing 49, the control unit 21 applies the complex spectrum component multiplied by the filter function F (f, j) for each frame f within the range of j = j1,..., J2. scalar values ^{{Aml '(f, j)} 2 + Bml' (f, j) 2} 1/2 the value of the maximum, and the scalar value ^{{Amr '(f, j)} 2 + Bmr' (f, j) ² } Values that maximize ^1/2 are calculated as Mml (f) and Mmr (f) for each LR channel. And the control part 21 multiplies 1 or more real value Scl (for example, Scl = 2.0) like following Formula.

The resulting scalar value multiplied by equations (20)-(23) is {Aml ″ (f, j) ² + Bml ″ (f, j) ² } ^1/2 > Mml (f), or { When Amr ″ (f, j) ² + Bmr ″ (f, j) ² } ^1/2 > Mmr (f), Scl should not exceed Mml (f) and Mmr (f) as follows: Correct and multiply.

When {Aml ″ (f, j) ² + Bml ″ (f, j) ² } ^1/2 > Mml (f), the control unit 21 determines that Scl ′ = Mml (f) / {Aml ′ (f, j) Calculate ² + Bml ′ (f, j) ² } ^1/2 . And the control part 21 multiplies Scl 'as following Formula.

Similarly, when {Amr ″ (f, j) ² + Bmr ″ (f, j) ² } ^1/2 > Mmr (f), the control unit 21 determines that Scl ′ = Mmr (f) / {Amr ′. (F, j) ² + Bmr ′ (f, j) ² } ^1/2 is calculated. And the control part 21 multiplies Scl 'as following Formula.

Here, the significance of the frequency dimension compression processing 49 will be described.
When the filter function is generated by performing the auditory sensitivity correction process 45, the music signal subjected to the filtering process using the filter function has a spectral characteristic that approaches 1 / f to 1 / f ² characteristic, and a low-frequency gradient. Becomes steep. The music signal originally has a discrete frequency characteristic, but the music signal that has been subjected to the filtering process as described above is far from the characteristic of white noise on which masking works most effectively.
On the other hand, for example, the air conditioner noise proposed in Japanese Patent Laid-Open No. 2010-031501 has a continuous spectrum characteristic of 1 / f curve, and the masking effect is slightly smaller than white noise having a flat characteristic. However, converting a music signal into a continuous spectrum adds noisy discomfort and is no longer music.
Therefore, in the embodiment of the present invention, the low-frequency part is made slightly flat like white noise while maintaining the state of discrete spectral characteristics by the frequency dimension compression processing 49. As a result, the masking effect can be enhanced while maintaining the timbre of the music.

FIG. 14 schematically shows the difference in operation between the time-dimensional compression process, which is a general compression process, and the frequency-dimensional compression process 49 in the embodiment of the present invention.
When the time-dimensional compression process is performed, the sound pressure increases as a whole, and the time undulation is reduced. That is, the concealment data 7 generated by performing the time-dimensional compression process feels troublesome for humans. In addition, since there is no significant change in the frequency characteristics, an overall increase in masking effect cannot be expected.
On the other hand, when the frequency dimension compression processing 49 is performed, flat white noise characteristics increase. Further, the temporal amplitude change is maintained. That is, the concealment data 7 generated by performing the frequency dimension compression processing 49 has a high masking effect and does not feel troublesome for humans.

  Returning to the description of FIG. In the inverse Fourier transform process 50, the control unit 21 performs an inverse Fourier transform of the re-modified complex spectrum calculated by the frequency dimension compression process 49, and the concealed frame data Xml ′ (f, i) and Xmr ′ (f, i ) (F = 0,..., Fm-1; i = 0,..., N-1).

The result of the frequency dimension compression processing 49 for each element of Am1 ′ (f, j), Bml ′ (f, j), Amr ′ (f, j), Bmr ′ (f, j) of each frame f Aml ″ (f, j), Bml ″ (f, j), Amr ″ (f, j), Bmr ″ (f, j).
In the inverse Fourier transform process 50, the control unit 21 converts the frame f−1 converted immediately before the concealment frame data Xml ′ (f, i) and Xmr ′ (f, i) of the frame f to be converted. When there is the concealment frame data Xml ′ (f−1, i) and Xmr ′ (f−1, i), the following equation is taken into consideration that both overlap by N / 2 samples on the time axis: To calculate.

The concealment data generation process according to the embodiment of the present invention has been described above. In the filter function creation process 33 according to the embodiment of the present invention, a value based on the voice maximum value spectrum Vv (j) is represented as a music average value spectrum. Divide value spectrum Div (j), which is a value divided for each corresponding frequency j by a value based on Vm (j), is calculated, and further, each value of divided value spectrum Div (j) corresponds to each other. The filter function F (j) is created by multiplying the value based on the auditory sensitivity correction curve L (j) defining the weight of the human auditory sensitivity for each frequency j. As a result, while enhancing the masking effect on the audio signal, the timbre of the music to be reproduced is maintained at the same level as the original sound, and the predetermined masking effect can be exerted even if the sound is reproduced at a reduced volume.
Further, in the filtering process 34 according to the embodiment of the present invention, the maximum scalar value of the complex spectrum within a predetermined frequency range is obtained for the complex spectrum multiplied by the filter function F (j) for each frame f. Further, after applying correction for multiplying each element of the complex spectrum by a predetermined scale value within a range where the scalar value of the element does not exceed the maximum scalar value, inverse Fourier transform is performed. . As a result, the masking effect can be further enhanced while maintaining the timbre of the music.
In the embodiment of the present invention, it is possible to avoid unnatural changes in the timbre of the BGM music by applying the filter processing, and while suppressing the playback volume of the BGM music more than before, it is equal to or higher than the conventional one. The masking effect can be exerted, and the concealment effect can be improved in a quieter and more comfortable acoustic environment than before.

  Next, an installation example of the concealment device will be described with reference to FIGS. 15 and 16. In the example illustrated in FIGS. 15 and 16, it is assumed that the concealment data 7 is generated by the concealment data generation device 2 and stored in the music player 52 that is the music playback device 3.

FIG. 15 shows a first installation example of the concealment device 1.
In the example shown in FIG. 15, the left side is the interview space 60 and the right side is the waiting space 65 with the flat speakers 51 a and 51 b interposed therebetween.
In the interview space 60, an interview counter table 61, a clerk chair 62, a visitor chair 63, and the like are installed. The interview counter table 61 is divided by a partition 64. A waiting sofa 65 is installed in the waiting space 65. When the customer visits the store, the customer waits in the waiting space 65, and in turn is called to the interview space 60 to interview the store clerk.

The flat speakers 51a and 51b are composed of a panel having a honeycomb structure and a speaker (exciter), such as a Posula sound panel (registered trademark of the present applicant).
The panels of the flat speakers 51a and 51b are preferably about the size of a partition that the clerk or visitor at the interview counter table 61 cannot look into from the waiting space 65, but even a signboard with an area of only A3 size is sufficient. Demonstrate the effect. That is, even if the conversation voice 71 reaches the waiting sofa 65 without being physically blocked by the flat speakers 51a and 51b, a sufficient masking effect can be obtained by the concealment data 7 of the present invention.
The Posula (registered trademark of the present applicant) sound panel can be manufactured to a width of about 1 meter.

The music player 52 is connected to the flat speakers 51a and 51b, and reproduces the concealment data 7 according to the embodiment of the present invention.
In the example shown in FIG. 15, the planar speakers 51a and 51b output BGM sound L72a and BGM sound R72b, which are masker sounds, respectively (stereo reproduction). The BGM sound may be reproduced in monaural, and the number and arrangement position of the flat speakers may be changed as appropriate according to the environment.

  The flat speakers 51a and 51b preferably have a mechanism that causes the music player 52 to uniformly radiate from the plane as sound waves in which the wavefront of the concealment data 7 is close to a plane wave. Thereby, the energy amount of the sound wave attenuated in the process of propagating to the waiting space 65 is smaller in the BGM sounds 72a and 72b output from the flat speakers 51a and 51b than in the conversational sound 71 uttered from the interview space 60. Thus, the amount of energy of the BGM sounds 72a and 72b is relatively larger than that of the conversation voice 71 uttered from the interview space 60, so that the masking effect can be enhanced. An example of such flat speakers 51a and 51b is disclosed in Japanese Patent Laid-Open No. 2007-301888. The speaker disclosed in Japanese Patent Application Laid-Open No. 2007-301888 is constituted by a panel having a fine tube structure array, and uniformly emits sound waves close to plane waves.

Here, the reason why the planar speakers 51a and 51b can enhance the masking effect by radiating sound waves close to a plane wave will be described.
As shown in FIG. 15, the conversation voice 71 reaches the waiting space 65 as the observation position as a spherical sound wave. Similarly, BGM reproduced from a normal dynamic speaker is also a spherical sound wave.
Here, in the case of a spherical wave, the surface area propagated in proportion to the square of the distance increases, and the energy concentrated on the sound source is dispersed. Therefore, the energy (sound pressure) is attenuated in inverse proportion to the square of the distance. It is known to do. On the other hand, in the case of a plane wave, energy is not attenuated so much even if the distance is long.

That is, BGM reproduced from a normal dynamic speaker is a sound wave of a spherical wave, and energy is attenuated when leaving, so if the volume is adjusted according to a customer who is waiting closer to the interview space 60, the interview space The masking effect may not work sufficiently for customers who are standing farther from 60.
On the other hand, if the planar speakers 51a and 51b that radiate sound waves close to plane waves are used, the reproduced BGM sound L72a and BGM sound R72b are plane wave sound waves, and the energy does not attenuate much even if they are separated. Even if the volume is adjusted according to the customer who is waiting at a close position, a sufficient masking effect is exerted on the customer who is waiting at a far position by the interview space 60.

FIG. 16 shows a second installation example of the concealment device 1.
In the example shown in FIG. 16, the first reception space 81a is on the left side of the flat speakers 51c and 51d, and the second reception space 81b is on the right side.
One reception table 82 and four chairs 83 are respectively installed in the first reception space 81a and the second reception space 81b.
In the first reception space 81a and the second reception space 81b, different customers are received independently.

The flat speakers 51c and 51d are composed of a honeycomb-structured panel and a speaker (exciter), such as a Posula sound panel (registered trademark of the present applicant). The flat speakers 51c and 51d shown in FIG. 16 have a larger width than the first installation example and also function as a partition.
The planar speakers 51c and 51d include a plurality of speakers (exciters), and BGM sound L72a and BGM sound R72b, which are masker sounds, are output from each speaker.
As in the first installation example, it is desirable that the flat speakers 51c and 51d have a mechanism that causes the music player 52 to uniformly radiate the wavefront of the concealment data 7 from the plane as a sound wave close to a plane wave.

As shown in FIG. 16, the first conversation voice 71a, which is a Musky sound, reaches the second reception space 81b, which is the observation position, as a spherical sound wave. Similarly, the second conversation voice 71b, which is a Musky sound, reaches the first reception space 81a, which is the observation position, as a spherical sound wave.
For the first conversation voice 71a, in the second reception space 81b, the BGM sound L72a and BGM sound R72b output from the flat speaker 51d become masker sounds and exhibit a masking effect. Similarly, for the second conversation voice 71b, in the first reception space 81a, the BGM sound L72a and BGM sound R72b output from the flat speaker 51c become masker sounds and exhibit a masking effect.

As described above, the installation example of the concealment device 1 has been described. As described above, a BGM is played at a relatively low volume by using a flat speaker that emits a sound wave close to a plane wave as a speaker that reproduces a music signal. However, the voice concealment effect can be exhibited.
Further, the flat speaker can take various forms from a standing signboard of about A3 size to a partition of about 1 meter wide.
In addition, as the design of the panel surface of the flat speaker, interior materials such as wallpaper and poster advertisements can be used, and the unnaturalness on the interior that the speaker is exposed visually can be avoided.

  In the above description, the planar speaker is a standing signboard or partition, but the embodiment of the present invention is not limited to this. For example, a speaker can be built in the wall of the room, and BGM sound, which is a masker sound, can be output from all sides of the room.

Next, examples and comparative examples will be described with reference to FIGS. 17 and 18 show data used in the examples and comparative examples. 19 and 20 show the results of the comparative example. 21 to 24 show the results of the examples.
In the embodiment, the concealment data 7 is generated by performing the auditory sensitivity correction process 45 and the compression process 49. On the other hand, in the comparative example, the concealment data was generated without performing the auditory sensitivity correction process 45 and the compression process 49.

  FIG. 17 is a diagram showing the maximum audio spectrum of the example and the comparative example. FIG. 17 shows the maximum audio spectrum of the audio data 4 output by the frequency analysis process 32. This voice maximum spectrum has a peak at 12 to 13 kHz.

FIG. 18 is a diagram illustrating the music average spectrum of the example and the comparative example. FIG. 18 shows a music average value spectrum of the music data 5 output by the frequency analysis processing 32.
Referring to FIG. 18, it can be seen that this music average spectrum is close to a 1 / f curve.

FIG. 19 is a diagram illustrating a filter function of a comparative example. The filter function of the comparative example is created without performing the auditory sensitivity correction process 45 and the compression process 49.
Compared with FIG. 22 described later, referring to FIG. 19, it can be seen that the value of the frequency component of 5 kHz to 10 kHz is high. For this reason, when the filtering process is performed using the filter function of the comparative example, the frequency components of 5 kHz to 10 kHz that contain many human audio signal components are easily emphasized. The frequency band of 5 kHz to 10 kHz is a region where the sensitivity characteristic of the human auditory system is relatively low. However, if the music signal is filtered using this filter function, the timbre changes unnaturally and becomes cumbersome.

FIG. 20 is a diagram illustrating the music signal after the filtering process of the comparative example. FIG. 20 shows a music signal that has been filtered using the filter function of the comparative example.
As described above, the music signal shown in FIG. 20 feels annoying because the timbre changes unnaturally.
In addition, compared with FIGS. 23 and 24 described later, referring to FIG. 20, the high frequency region of 10 kHz or higher has a higher value. Although a human audio signal component exists as it is in a frequency component of 10 kHz or higher, it does not contribute much to masking because human auditory sensitivity is low. Therefore, the reproduction volume is set based on a band of 4 kHz or less where the auditory sensitivity is high, and the entire sound pressure level is unnecessarily increased. Therefore, it is difficult to obtain a masking effect unless the volume is increased considerably.

  FIG. 21 is a diagram illustrating an auditory sensitivity correction curve of the example. FIG. 21 plots the sound pressure level of the auditory sensitivity correction curve 6 for each frequency shown in the lower part of FIG.

FIG. 22 is a diagram illustrating a filter function of the embodiment. FIG. 22 shows a filter function created by the filter function creation process 33 using the auditory sensitivity correction curve 6 shown in FIG.
Compared to FIG. 19, referring to FIG. 22, it can be seen that the value of the frequency component of 5 kHz to 10 kHz is low. For this reason, when the filtering process 34 is performed using the filter function of the embodiment, it is possible to prevent the degree of emphasis in the frequency band of 5 kHz to 10 kHz from being suppressed and an unnatural tone. In addition, the frequency band of 300 Hz to 3.4 kHz, which has a high human auditory sensitivity and is rich in formant components for identifying speech, is emphasized, and masking can easily work effectively without increasing the playback volume.

FIG. 23 is a diagram illustrating a music signal after filtering processing (no compression) according to the embodiment. FIG. 23 shows a music signal that has been subjected to the filtering process 34 using the filter function of the embodiment. However, the music signal shown in FIG. 23 is not subjected to the frequency dimension compression processing 49.
As described above, the music signal shown in FIG. 23 has a natural tone color with the degree of enhancement in the frequency band of 5 kHz to 10 kHz being suppressed. Further, the frequency band of 300 Hz to 3.4 kHz, which has high human auditory sensitivity and is rich in formant components for identifying speech, is emphasized, and masking works effectively even if the reproduction volume is not increased so much.

FIG. 24 is a diagram illustrating a music signal after filtering processing (with compression) according to the embodiment. FIG. 24 shows a music signal that has been subjected to the filtering process 34 using the filter function of the embodiment. Also, the music signal shown in FIG. 24 is subjected to frequency dimension compression processing 49.
As described above, the music signal shown in FIG. 24 has a natural tone color with the degree of emphasis in the frequency band of 5 kHz to 10 kHz being suppressed. Further, the frequency band of 300 Hz to 3.4 kHz, which has high human auditory sensitivity and is rich in formant components for identifying speech, is emphasized, and masking works effectively even if the reproduction volume is not increased so much.
Furthermore, referring to FIG. 24 as compared with FIG. 23, it can be seen that the low frequency region is flat as a whole while taking discrete values. That is, the low-frequency part can be made slightly flat like white noise while maintaining the state of discrete spectral characteristics. As a result, the masking effect can be enhanced while maintaining the timbre of the music.

  The preferred embodiments of the concealed data generation device and the like according to the present invention have been described above with reference to the accompanying drawings, but the present invention is not limited to such examples. It will be apparent to those skilled in the art that various changes or modifications can be conceived within the scope of the technical idea disclosed in the present application, and these naturally belong to the technical scope of the present invention. Understood.

DESCRIPTION OF SYMBOLS 1 ......... Concealment apparatus 2 ......... Concealment data generation apparatus 3 ......... Music reproduction apparatus 4 ......... Audio data 5 ......... Music data 6 ......... Hearing sensitivity correction curve 7 ......... Concealment data 10 ......... Audio frame group 11 ......... Music frame group 12 ......... Maximum audio spectrum data 13 ......... Audio average spectrum data 14 ......... Filter function data 31 ......... Frame extraction process 32 ......... Frequency analysis processing 32a ......... frequency analysis 33 ......... filter function creation processing 34 ......... filtering processing 41 ......... instantaneous spectrum calculation processing 42 ......... average spectrum calculation processing 43 ......... critical bandwidth correction processing 44 ... …… Division processing 45 …… Hearing sensitivity correction processing 46 ………… Smoothing processing 47 ……… Fourier transform processing 48 ……… Filter function multiplication processing 49 ……… Frequency processing Multidimensional compression process 50 ... Fourier inverse transform process

Claims (12)

A concealment data generation device that generates concealment data that is music data for concealing dialogue voice,
Frequency analysis is performed on each of voice data and music data stored in advance, and a voice maximum value spectrum Vv (j) (j is a frequency) which is a maximum spectrum in the time axis direction of the voice data is calculated, Frequency analysis means for calculating a music mean value spectrum Vm (j), which is a spectrum averaged in the time axis direction of music data;
A division value spectrum Div (j), which is a value obtained by dividing a value based on the maximum speech value spectrum Vv (j) by a value based on the music average value spectrum Vm (j) for each corresponding frequency j, is calculated; Further, each value of the division value spectrum Div (j) is multiplied by a value based on an auditory sensitivity correction curve L (j) defining a weight of human auditory sensitivity for each frequency j corresponding to each other, thereby obtaining a filter. A filter function creating means for creating a function F (j);
The concealed data is generated by dividing the music data into frames f that are units of a predetermined section, Fourier transforming each divided frame f, multiplying by the filter function F (j), and inverse Fourier transform Filtering means to
A concealed data generating apparatus comprising: The concealment data generation device according to claim 1, wherein the auditory sensitivity correction curve L (j) used by the filter function creation means is defined based on an equal loudness curve of 40 phones. The filtering means obtains a maximum scalar value of the complex spectrum within a predetermined frequency range for the complex spectrum multiplied by the filter function F (j) for each frame f, and further, the complex spectrum And performing the inverse Fourier transform after performing a correction for multiplying a predetermined one or more scale values within a range in which the scalar value of the element does not exceed the maximum scalar value. The concealed data generation device according to claim 1. The filter function creating means includes
Replacing the voice maximum value spectrum Vv (jc) (jc is a specific frequency) with a maximum value in a range higher than the frequency jc to calculate a replacement voice maximum value spectrum;
Replacing the music average value spectrum Vm (jc) with an average value within a range before and after the frequency jc to calculate a replacement music average value spectrum;
The concealment data according to any one of claims 1 to 3, wherein a value obtained by dividing the replacement speech maximum value spectrum by the replacement music average value spectrum is the division value spectrum Div (j). Generator. The filter function creating means multiplies each value of the filter function F (j) by a value based on the auditory sensitivity correction curve L (j), and then replaces it with an average value within a range before and after the frequency j. 5. The concealed data generation device according to claim 1, wherein the filter function F (j) is smoothed. The frequency analysis means calculates, for each frame f, a spectrum averaged in the time axis direction over M frames before and after each frame f of the music data as the music average value spectrum Vm (f, j),
The filter function creation means sets a value based on the voice maximum value spectrum Vv (j) as the divided value spectrum Div (f, j) to the music average value spectrum Vm (f, j) corresponding to the frame f. A value obtained by dividing each frequency j corresponding to each other by a value based on the calculated value, and further, for each value of the divided value spectrum Div (f, j), the auditory sensitivity correction curve L (for each frequency j corresponding to each other is calculated. 6. The concealed data generation device according to claim 1, wherein the filter function F (f, j) is created by multiplying a value based on j). Music data storage means for storing a plurality of the music data;
Music data selection means for selecting a single piece of music data from the music data stored by the music data storage means;
Further comprising
The concealment data generation device according to claim 1, wherein the concealment data is generated based on the single music data selected by the music data selection unit. A concealed data storage unit that stores a plurality of the concealment data generated by the concealment data generation device according to any one of claims 1 to 7,
Concealment data selection means for selecting a single concealment data from the concealment data stored by the concealment data storage means;
Concealed data reproducing means for reproducing the single concealed data selected by the concealed data selecting means;
A concealment device comprising: 9. The concealment method according to claim 8, wherein the concealment data reproducing means is constituted by a flat speaker having a mechanism for uniformly radiating the concealment data from a predetermined plane as a sound wave having a wavefront close to a plane wave. apparatus. A concealment data generation method for generating concealment data, which is music data for concealing dialogue voice,
Frequency analysis is performed on each of voice data and music data stored in advance, and a voice maximum value spectrum Vv (j) (j is a frequency) which is a maximum spectrum in the time axis direction of the voice data is calculated, A frequency analysis step of calculating a music average value spectrum Vm (j) which is a spectrum averaged in the time axis direction of the music data;
A division value spectrum Div (j), which is a value obtained by dividing a value based on the maximum speech value spectrum Vv (j) by a value based on the music average value spectrum Vm (j) for each corresponding frequency j, is calculated; Further, each value of the division value spectrum Div (j) is multiplied by a value based on an auditory sensitivity correction curve L (j) defining a weight of human auditory sensitivity for each frequency j corresponding to each other, thereby obtaining a filter. A filter function creating step for creating a function F (j);
The concealed data is generated by dividing the music data into frames f that are units of a predetermined section, Fourier transforming each divided frame f, multiplying by the filter function F (j), and inverse Fourier transform A filtering step to
A method for generating concealed data, comprising: A concealed data storage step of storing a plurality of the concealed data generated by the concealed data generation method according to claim 10;
A concealment data selection step of selecting a single concealment data from the concealment data stored by the concealment data storage step;
A concealed data reproduction step of reproducing the single concealment data selected by the concealment data selection step;
The concealment method characterized by including this. On the computer,
Frequency analysis is performed on each of voice data and music data stored in advance, and a voice maximum value spectrum Vv (j) (j is a frequency) which is a maximum spectrum in the time axis direction of the voice data is calculated, A frequency analysis step of calculating a music average value spectrum Vm (j) which is a spectrum averaged in the time axis direction of the music data;
A division value spectrum Div (j), which is a value obtained by dividing a value based on the maximum speech value spectrum Vv (j) by a value based on the music average value spectrum Vm (j) for each corresponding frequency j, is calculated; Further, each value of the division value spectrum Div (j) is multiplied by a value based on an auditory sensitivity correction curve L (j) defining a weight of human auditory sensitivity for each frequency j corresponding to each other, thereby obtaining a filter. A filter function creating step for creating a function F (j);
The concealed data is generated by dividing the music data into frames f that are units of a predetermined section, Fourier transforming each divided frame f, multiplying by the filter function F (j), and inverse Fourier transform A filtering step to
A computer-readable program for executing the program.