JP5662712B2 - Voice changing device, voice changing method and voice information secret talk system - Google Patents

Description

  The present invention relates to a voice changing device that changes voice using a non-linear function, a voice changing method, and a voice information secret talk system including the voice changing device.

  With the enforcement of the Personal Information Protection Law, there is an increasing need to protect conversation information in banks and offices. Conventionally, sound insulation / soundproofing that physically separates the space, BGM / masking system that conceals conversational speech with other noise / music, etc. in an open plan office have been proposed.

For the purpose of concealing voice information,
(1) Masking system that masks the target speech with other stationary noise
(2) Shading system concealed by indoor background noise and air conditioning noise
(3) Sound insulation / sound insulation (the target room is spatially separated and acoustically separated)
Etc. The example (1) attempts to (forcefully) erase the presence of speech, and is positioned as energy masking. This is used, for example, in an open plan office booth or conference room.

  An example of the system (1) is reported in Non-Patent Document 1. There, a dedicated generator or speaker is installed inside the ceiling, etc., and masking sound is generated to conceal the sound. The principle is that it generates music and noise that is not in the way of conversation (contrast with conversation), conceals the contents of speech by reducing so-called S / N, and reduces clarity and intelligibility. Or to conceal the content of the conversation to the extent that it cannot be understood. The system includes a control device (signal processing device), a power amplifier, and the like that control the masking sound to the optimum level according to the conversation level, background noise, and the like.

  In addition, as an example using this technology, noise for masking was radiated from the partition into the booth, and the target space area was limited to the booth to suppress the increase in the noise level in the entire room. There is.

  An example of the system (2) is reported in Non-Patent Document 2. There are reports of “Sound Shading System” that uses indoor background noise itself and air-conditioning noise that is familiar everyday as radiating masking noise. In this system, a speaker is installed at the top of the partition for the purpose of protecting the privacy of conversation, in contrast to a visually interrupting partition for the purpose of ensuring privacy at a bank counter. A masking sound is reproduced from this speaker, thereby preventing the leakage / transmission of conversation contents to a person on the other side of the partition. The sound to be reproduced is the sound generated based on the crowds of the city or the air conditioning noise of the room.

  As an example of the system of (3), there is sound insulation partitioned as a separate room or soundproof partitioned by a partition.

JP 2008-233671 A

KOKUYO Press Release, Sound Masking, October 18, 2006 Akiko Sugimoto, Takahiro Nakamura, Shiro Ise, “Evaluation of Sound Shading System that generates sound field in consideration of ease of conversation and privacy”, Acoustical Society of Japan Spring Meeting 2005, Proceedings, p. 817 The Institute of Electronics, Information and Communication Engineers, Auditory and Speech, 1973, p. 370-371 Iwata, Kobayashi, Takeda, Itakura, “Analysis of phonetic features contained in human speech-like noise”, Journal of the Acoustical Society of Japan, May 1, 1997, 53 (5), p. 337-345

The inventor has recognized the following problems regarding the above-described masking / shading technique.
(I) Since the original sound emits a new sound having no context, the noise level in the indoor space can be raised with a sense of incongruity.
(II) Noise, that is, a masking sound can be heard even when the sound is not generated.
(III) By emitting another sound (noise / music) unrelated to conversation, it is possible to give a sense of incongruity to a speaker, a talker, and other people in the room.
(IV) Concealment of speech information includes the basic problem that it is difficult to succeed with noise and BGM due to the auditory nature of distinguishing and recognizing different things (envelope and spectrum are similar) Audio waveforms are more difficult to distinguish for auditory recognition).

  As for (I), the relative level of noise necessary for completely masking the original voice is empirically about 15 dB (see Non-Patent Document 3). From this point of view, the method of masking the sound by playing noise and music (masking approach) requires much louder noise and music than the original sound, whether masking or shading. The room noise level can be greatly increased.

Regarding (II), there is a sense of incongruity that a sound is produced even when there is no utterance. In the first place, playing noise and music when there is no utterance is useless from the viewpoint of concealing conversation content. Not only is it wasteful, but it also _{increases the} room equivalent noise level (L _Aeq : A-weighted equivalent sound level = the root mean square sound pressure level of the audio signal corrected with the A characteristic, that is, the average noise level). Can result. Even when “HSL noise (Human Speech-like noise)” (see Non-Patent Document 4) created from music or voice is used instead of noise, it is difficult to distinguish from general BGM.

  In addition, the approach (3) is considerably large in cost and hinders a feeling of opening, and is not suitable for use in an open plan office or the like.

  Further, in the sound masking system described in Patent Document 1, a method is described in which the speech speed of an input sound (voice) is analyzed, divided and processed by a frame length corresponding to this, and the processed speech is synthesized. However, since this system “stores the input sound (voice) temporarily in about 2 seconds and performs a series of processing”, the processed sound is generated from the past sound that is different from the sound that is the masking target. . Therefore, the relevance between the processed voice and the voice targeted for masking is low, and the masking effect is not sufficient.

  The present invention has been made in view of these problems, and its purpose is to conceal the contents of audio at real time or at a processing speed equivalent to real time, while suppressing the increase in noise level and listener discomfort. The provision of technology.

  One embodiment of the present invention relates to a sound changing device. The voice changing device includes a partial extraction unit that extracts a signal of a change target portion from a voice signal representing a voice being uttered, and a nonlinear that changes a signal of the change target portion extracted by the partial extraction unit using a nonlinear function A changing unit; and an output unit that outputs at least the signal of the change target portion changed by the non-linear changing unit to a sound output unit capable of outputting the sound to a region where the sound being spoken is received.

  According to this aspect, it is possible to output, in real time, a sound obtained by performing non-linear processing on the sound being uttered in a region where the sound being uttered is being listened to.

  Another aspect of the present invention is a speech information secret talk system. The voice information secret speech system is modified by a sound collecting unit that receives a voice being uttered and generates a voice signal representing the voice, a voice changing device that changes a voice signal generated by the sound collecting unit, and a voice changing device. Voice output means for converting the received voice signal into a voice and outputting the voice signal to a region where the voice being spoken is received. The voice changing device includes a partial extraction unit that extracts a signal of the change target portion from the voice signal generated by the sound collecting means, and a nonlinear that changes the signal of the change target portion extracted by the partial extraction unit using a nonlinear function A changing unit, and an output unit that outputs a signal of the change target portion changed by at least the nonlinear changing unit to the audio output unit.

  It should be noted that any combination of the above-described constituent elements, or those obtained by replacing the constituent elements and expressions of the present invention with each other between apparatuses, methods, systems, computer programs, recording media storing computer programs, and the like are also included in the present invention. It is effective as an embodiment of

  ADVANTAGE OF THE INVENTION According to this invention, the content of an audio | voice can be concealed, suppressing the increase in a noise level and a listener's discomfort.

It is explanatory drawing which divides into a category the conventional approach regarding masking, and the approach which concerns on embodiment. It is a perspective view which shows typically the booth provided with the audio | voice information secret talk system which concerns on embodiment. It is a block diagram which shows typically the function and structure of the audio | voice information confidential system of FIG. It is a side view which shows the structure of the IT partition of FIG. It is a block diagram which shows the function and structure of SD controller part SD of FIG. It is explanatory drawing for demonstrating the determination criteria of the signal of the change object part in a part determination part. It is explanatory drawing which shows an example of the process in a 2nd change part. It is a wave form diagram which shows the waveform of the audio | voice signal showing the maskee and masker in a listener position. It is a flowchart which shows a series of processes in an SD controller part and a speaker. It is a graph which shows the relationship between the difference between a masker and a maskee, and a recognition rate. It is a block diagram which shows typically the function and structure of the audio | voice information confidential system which concerns on a 1st modification. It is a block diagram which shows typically the function and structure of the audio | voice information confidential system which concerns on a 2nd modification.

  The present invention will be described below based on preferred embodiments with reference to the drawings. The same or equivalent components, members, and processes shown in the drawings are denoted by the same reference numerals, and repeated descriptions are omitted as appropriate.

  Particularly in offices and the like, it is desirable to conceal only the voice information, that is, the voice content without impairing the openness and smoothness of communication of the open plan space. However, the conventional technology using BGM and masking basically adds another sound having a different property from that of speech, so that there is a dislike of increasing the sense of incongruity and background noise in the room. The embodiment of the present invention changes the structure of a sound signal itself collected by a microphone or the like using a nonlinear function, and outputs the changed sound substantially in real time with respect to the original sound. Concealing / blocking the content of the conversation without raising the background noise, ideally only the content of the conversation, a smooth and comfortable secret environment is realized.

  FIG. 1 is an explanatory diagram showing a conventional approach related to masking and an approach according to an embodiment divided into categories. (A) is SR (Sound Reinforcement) / PA (Public Address) using electroacoustics. These are conventional technologies that increase the volume and clarity and “make them sound better”. (F) is Sound Insulation, which is a conventional technique for acoustically separating a space and making it “not audible” as much as possible. On the other hand, the approach according to the embodiment is SD (Speech Deformation) of (e). By processing the original voice of the conversation person and outputting it in near real time, the conversation contents are not heard but not heard. It is a kind of voice information disruption (hearing) technique that makes it “unknown”. Further, (b) EM, (c) SS, and (d) IM according to the prior art can increase the noise level in the room or the target space region to increase the unpleasantness and discomfort, SD hardly causes an increase in noise level.

In the embodiment of the present invention, in order to reduce the overall volume of processing voice (hereinafter referred to as “masker”) added to the original speech (hereinafter referred to as “maskee”) that is the speech being uttered, Is possible.
In order to increase the masking effect of the masker, ANC (Active Noise Control) or PNC (Passive Noise Control) technology with fixed parameters is used in combination.

FIG. 2 is a perspective view schematically showing the booth 2 in which the audio information secret system 100 according to the embodiment is provided. FIG. 3 is a block diagram schematically showing the function and configuration of the speech information secret system 100 of FIG.
The voice information secret story system 100 is provided in a booth 2 partitioned by a simple partition, such as a bank consultation counter. The audio information secret system 100 includes a microphone Mic, an SD controller unit SD, two power amplifiers PA, and two speakers SP. The speaker SP and the SD controller unit SD may be incorporated in the IT partition 4 that visually separates the booths.

  A customer 6 who has a conversation with a counselor is a speaker. The speaker's maskee H ′ (t) is collected by a microphone Mic provided at or near the counter portion. The maskee H ′ (t) collected by the microphone Mic is converted into an audio signal and sent to the SD controller unit SD. This audio signal is non-linearly changed by the SD controller unit SD. The audio signal that has undergone nonlinear processing in the SD controller section SD is output as a masker H (t) from the speaker SP to the left and right adjacent booths 2 'via the power amplifier PA.

  Since Muskie H '(t) goes around the air in the adjacent booth 2', the voice being spoken by the customer 6 depends on the listener 8 (a different person from the customer 6) in the adjacent booth 2 '. Can be heard. However, in the present embodiment, the masky H ′ (t) that leaks through the air is combined with the masker H (t) and reaches the listener 8 in the adjacent booth 2 ′. Therefore, the listener 8 cannot understand the content of the conversation included in the maskee H ′ (t) due to the disturbance caused by the masker H (t).

  The speaker SP outputs a masker H (t) toward the adjacent booth 2 'next to the booth 2 where the SD controller unit SD and the microphone Mic to which the speaker SP is connected are installed. Here, the adjacent booth 2 ′ is an area where the muskey H ′ (t) leaking around the air is received. In other words, the masker H (t) is output from the speaker SP so that the maskee H ′ (t) and the masker H (t) reach the listener 8 substantially in real time. The main body that guarantees the real-time property may be the SD controller unit SD or the speaker SP, but in the following, the SD controller unit SD performs the real-time property of the maskee H ′ (t) and the masker H (t). The case where the changed audio signal is output to the speaker SP will be described.

  FIG. 4 is a side view showing the configuration of the IT partition 4 of FIG. The IT partition 4 has a stacked structure in which a first sound absorbing layer 142, a sound insulating layer 144, and a second sound absorbing layer 146 are stacked in this order. Each of the first sound absorbing layer 142 and the second sound absorbing layer 146 is a glass wool layer having a thickness of 20 mm. The sound insulation layer 144 is a gypsum board having a thickness of 12 mm.

  FIG. 5 is a block diagram showing the function and configuration of the SD controller unit SD of FIG. Each block shown here can be realized in hardware by an element such as a CPU (central processing unit) or a mechanical device, and in software by a computer program or the like. Describes functional blocks realized by collaboration. Accordingly, it is understood by those skilled in the art who have touched this specification that these functional blocks can be realized in various forms by a combination of hardware and software.

  The SD controller unit SD includes an A / D unit 20, a partial extraction unit 30, a nonlinear change unit 40, and an output unit 50.

  The maskee H ′ (t) collected by the microphone Mic is converted into an audio signal, and the audio signal is input to the A / D unit 20 through a microphone amplifier (not shown). The A / D unit 20 converts an audio signal that is an analog signal into digital data. The audio signal digitized by the A / D unit 20 is digital data in which a voltage value corresponding to the magnitude of sound pressure is associated with time, for example.

The partial extraction unit 30 extracts the signal of the change target portion from the audio signal digitized by the A / D unit 20. The partial extraction unit 30 includes a signal division unit 32, a partial determination unit 34, a first envelope acquisition unit 36, and a first switch 39.
The partial extraction unit 30 has at least two modes, ie, approximately one mountain extraction mode and random division mode, regarding the extraction of the signal of the change target portion. In the partial extraction unit 30, both modes are configured to be selectable. In the present embodiment, the user switches the mode by switching the first switch 39. The first switch 39 may be implemented as a hardware switch or may be implemented as a software switch.

(Approximately one mountain extraction mode)
When the first switch 39 is set so as to connect the A / D unit 20 and the first envelope acquisition unit 36, the partial extraction unit 30 operates in approximately one mountain extraction mode. In the approximately one mountain extraction mode, the first envelope acquisition unit 36 acquires data indicating the envelope of the audio signal. This data is digital data in which a voltage value corresponding to the size of the envelope is associated with time, for example. Hereinafter, data indicating an envelope is simply referred to as an envelope. The first envelope acquisition unit 36 includes a square sound pressure acquisition unit 37 and a low-pass filter 38.

  The squared sound pressure acquisition unit 37 acquires the squared sound pressure waveform of the voice signal digitized by the A / D unit 20. The squared sound pressure acquisition unit 37 squares the audio signal and obtains a squared sound pressure waveform by multiplying by a predetermined coefficient as necessary.

The low-pass filter 38 averages the squared sound pressure waveform acquired by the squared sound pressure acquisition unit 37 with a time constant of several msec to several hundred msec. That is, the low-pass filter 38 performs a low-pass filter process on the squared sound pressure waveform. Thereby, a change faster than the time constant is removed from the squared sound pressure waveform, and a smooth waveform is obtained. In this embodiment, this smooth waveform is the envelope of the audio signal. It should be understood by those skilled in the art who have touched this specification that the envelope of the audio signal may be obtained by other methods. In the present embodiment, the envelope is data indicating a change in the amplitude of the audio signal in a broad sense.
The low-pass filter 38 takes the square root of the low-pass filtered data if necessary.

  The part determination unit 34 detects a rising part that continuously rises from several dB to several tens dB, for example, 5 dB or more, from the envelope of the audio signal obtained by the low-pass filter 38. Next, the part determining unit 34 detects a descending part that descends continuously several dB to several tens dB, for example, 5 dB or more after the ascending part. The part determination part 34 determines the audio | voice signal between a raise part and the fall part corresponding to it as a signal of a change object part. In many cases, the envelope of the signal of the change target portion determined in this way is approximately one mountain.

  FIG. 6 is an explanatory diagram for explaining the determination criteria of the signal of the change target part in the part determination unit 34. FIG. 6A is an explanatory diagram for explaining a case where the signal of the change target part is determined based on the detection of the rising part and the falling part in the part determination unit 34. FIG. 6A shows an exemplary audio signal waveform 211 and its envelope 208. The part determination unit 34 detects the rising part 202 based on the rate of change of the envelope 208. Next, the part determining unit 34 detects the descending part 204 after the ascending part 202. The part determining unit 34 converts the audio signal of the section 206 (the section sandwiched between the time t1 before the peak 203 and the time t2 after the peak 203) sandwiched between the rising section 202 and the descending section 204 to the signal to be changed. Determine as.

  Note that the part determination unit 34 may determine the signal of the change target part by another method. For example, the part determination unit 34 may detect a part where the envelope is inflated and determine a sound signal corresponding to the part as a signal of the part to be changed. Or the part determination part 34 may detect the peak of an envelope, and may determine the audio | voice signal of the area which has a predetermined length before and behind as a signal of a change object part. Alternatively, the part determining unit 34 may determine the audio signal of the continuous section where the envelope exceeds a predetermined level as the signal of the change target part.

  FIG. 6B is an explanatory diagram for explaining a case where the signal of the change target portion is determined based on the peak detection in the portion determination unit 34. FIG. 6B shows an exemplary audio signal waveform 212 and its envelope 214. The partial determination unit 34 detects the peak 216 of the envelope 214. The part determining unit 34 determines the audio signal of the section 218 having a predetermined length before and after the peak 216 as the signal to be changed.

  FIG. 6C is an explanatory diagram for explaining a case where the signal of the change target portion is determined based on the envelope level in the portion determination unit 34. FIG. 6C shows an exemplary audio signal waveform 220 and its envelope 222. The part determining unit 34 detects a continuous section 226 in which the envelope 222 exceeds a predetermined level 224, and determines the audio signal in the section 226 as a signal to be changed. In this case, depending on how to obtain a predetermined level, the signal of the change target portion may include two or more peaks.

  As described above, various methods for determining the signal of the change target portion are conceivable. Such a large number of options is preferable in the sense that it provides a great degree of freedom for making the concealment of conversation contents by SD more effective.

  In addition, what can be said through these various determination methods is that a group of signals is determined based on the waveform of an audio signal, particularly based on its statistical properties, and the group of signals thus determined is That is, it is determined as a signal of the part to be changed. That is, the change target portion is adaptively determined according to the incoming audio signal. In this case, according to the experience of the present inventor as a person skilled in the art and preliminary experiments, it has been found that the conversation content disturbance effect is higher than, for example, the case where audio signals are cut out at predetermined intervals. . In particular, according to an experiment conducted by the present inventor, when approximately one peak of an envelope is extracted as a change unit, the disturbance effect is more effective than, for example, cutting out at a constant period or using a consonant or vowel as a change unit. It was found to be expensive.

Returning to FIG.
The part determining unit 34 outputs a part of the audio signal that has not been determined as the signal to be changed to the delay adjusting unit 52.

(Random split mode)
When the first switch 39 is set to connect the A / D unit 20 and the signal division unit 32, the partial extraction unit 30 operates in the random division mode. In the random division mode, the signal division unit 32 divides the audio signal digitized by the A / D unit 20 in a period having a random length. The length of the period varies between several tens of milliseconds to several hundreds of milliseconds. Alternatively, the length of the period varies within a range of ± several tens of% to several hundreds of% with respect to a certain period. For example, the length of the period changes as follows: 11 msec, 10 msec, 12 msec,.

The part determining unit 34 determines a signal corresponding to one of the periods divided by the signal dividing unit 32 in the audio signal as a signal of the change target part. The part determination unit 34 may select all the divided periods as the change target part, or may select every other period as the change target part. In the latter case, the part determining unit 34 outputs the audio signal of the part corresponding to the period not selected as the change target part to the delay adjusting unit 52.
In the random division mode, since the randomness is added to the length of the period, the naturalness of the masker H (t) is improved.

The non-linear changing unit 40 changes the change target portion extracted by the partial extracting unit 30 in real time or near real time using a non-linear function. The nonlinear changing unit 40 includes a first changing unit 42, a second changing unit 44, a third changing unit 46, and a second switch 48.
The nonlinear changing unit 40 has at least three modes of a first change mode, a second change mode, and a third change mode. These modes are configured to be selectable in the nonlinear changing unit 40. In the present embodiment, the user switches the mode by switching the second switch 48. Note that the second switch 48 may be implemented as a hardware switch or a software switch.

(First change mode)
When the second switch 48 is set to connect the partial determination unit 34 and the first change unit 42, the nonlinear change unit 40 operates in the first change mode. In the first change mode, the first change unit 42 acquires the signal of the change target portion determined by the portion determination unit 34 and performs nonlinear processing on the signal. The first change unit 42 includes a second envelope acquisition unit 62, a first nonlinear processing unit 64, and an integration unit 66.

  The second envelope acquisition unit 62 has the same configuration as the first envelope acquisition unit 36. That is, the second envelope acquisition unit 62 acquires an envelope from the signal of the change target portion extracted by the partial extraction unit 30. Alternatively, when the substantially single mountain mode is used in the partial extraction unit 30, the second envelope acquisition unit 62 uses the envelope corresponding to the signal of the change target portion from the envelope acquired by the first envelope acquisition unit 36. May be obtained.

The first nonlinear processing unit 64 processes the signal of the change target portion extracted by the partial extraction unit 30 using a nonlinear function. As the nonlinear function, for example, a function based on absolute value and logarithmic transformation is used. In particular, the first nonlinear processing unit 64 calculates the logarithm (log ₂ | y (t) |) of the absolute value (| y (t) |) of the signal (y (t)) of the change target portion with respect to the base 2.

  The integrating unit 66 changes the signal of the change target portion processed by the first nonlinear processing unit 64 based on the envelope acquired by the second envelope acquiring unit 62. In particular, the integrating unit 66 integrates the envelope acquired by the second envelope acquiring unit 62 and the calculation result in the first nonlinear processing unit 64. Thereby, even when the shape of the envelope is broken by the processing in the first nonlinear processing unit 64, the shape of the envelope can be recovered by the processing in the integrating unit 66.

  The first changing unit 42 repeats the above process for each signal of the change target portion determined by the portion determining unit 34, and outputs the signal thus processed to the delay adjusting unit 52.

(Second change mode)
When the second switch 48 is set to connect the partial determination unit 34 and the second change unit 44, the nonlinear change unit 40 operates in the second change mode. In the second change mode, the second change unit 44 acquires the signal of the change target portion determined by the portion determination unit 34 and performs nonlinear processing on the signal. The second changing unit 44 includes a replacement unit 68 and a second nonlinear processing unit 70.

The replacement unit 68 replaces the signal value at a certain time with the signal value at another time in the signal of the change target portion.
The second nonlinear processing unit 70 processes the signal of the change target portion replaced by the replacement unit 68 using a nonlinear function.

FIG. 7 is an explanatory diagram illustrating an example of processing in the second changing unit 44. In FIG. 7, the horizontal axis represents time, and the vertical axis represents voltage. A solid line 228 in FIG. 7 shows a waveform of an audio signal as an analog signal input to the A / D unit 20. It is assumed that the voice signal of the section 230 is extracted by the partial extraction unit 30 as the signal of the change target part. The signal of the change target portion is digital data, and voltage values y ₀ = f (t ₀ ) and y ₁ = f (t ₁ ) corresponding to the times t ₀ , t ₁ ,..., T _N (N is a natural number). ),..., Y _N = f (t _N ). Here, it is assumed that t ₀ <t _N and the times are arranged at equal intervals.
In FIG. 7, the first data point 232 is (t ₀ , y ₀ ), the second data point 234 is (t _N−i , y _N−i ) (i is a natural number, 0 ≦ i ≦ N), the third data Point 236 indicates (t _N , y _N ).

The replacement unit 68 changes the signal of the part to be changed using a quasi-function y ′ = f (t _N −t _i ). For example, for the time t _i , the replacement unit 68 replaces y _i with y ′ _i = f (t _N −t _i ) = f (t _N−i ) = y _N−i . The fourth data point 238 after such replacement is denoted by (t _i , y ′ _i = y _N−i ). A one-dot chain line 240 in FIG. 7 shows the waveform of the signal replaced by the replacement unit 68.

The second nonlinear processing unit 70 changes the signal replaced by the replacement unit 68 using a nonlinear function Y = g (y ′) such as a logarithm. For example, for the fourth data point 238, the second non-linear processing unit 70 sets y ′ _i to Y _i = g (y ′ _i ) = g (y _N−i ). The fifth data point 242 after such a change is indicated by (t _i , Y _i = g (y _N−i )). A two-dot chain line 244 in FIG. 7 indicates the waveform of the signal changed by the second nonlinear processing unit 70.

Returning to FIG.
The second changing unit 44 repeats the above processing for each signal of the change target portion determined by the portion determining unit 34, and outputs the signal thus processed to the delay adjusting unit 52.
In addition, the process in the 2nd change part 44 is not restricted to the above-mentioned process. For example, it is possible to employ other things in the form of t ₀ and t _N and magnitude relations and Quasi function f.

(Third change mode)
When the second switch 48 is set to connect the partial determination unit 34 and the third change unit 46, the nonlinear change unit 40 operates in the third change mode. In the third change mode, the third change unit 46 acquires the signal of the change target portion determined by the portion determination unit 34 and performs nonlinear processing on the signal. The third changing unit 46 includes a preprocessing unit 72, an LPC analysis unit 74, a residual processing unit 76, a frequency characteristic conversion unit 78, and a synthesis unit 80.

  The third changing unit 46 performs formant conversion on the signal of the change target part. The formant conversion technology is used when returning the modified sound to a sound close to the original sound, such as in deep sea work using helium gas.

  The formant conversion process is performed as follows. The pre-processing unit 72 performs pre-emphasis on the signal of the change target part. The LPC analysis unit 74 performs linear prediction (LPC) analysis on the signal that has been subjected to pre-emphasis in the preprocessing unit 72, and divides the signal into a vocal tract frequency characteristic and a sound source (residual signal). The frequency characteristic converter 78 transforms the frequency characteristic of the vocal tract. The residual processing unit 76 down-samples the residual signal so as to have a desired frequency. Alternatively, the residual processing unit 76 uses the residual signal as it is. The synthesizer 80 synthesizes the output of the frequency characteristic converter 78 and the output of the residual processor 76. The signal synthesized in the synthesizing unit 80 is a signal indicating a modified processed voice in which the pitch frequency (sound fundamental frequency) is the same but the formant is changed when compared with the original signal to be changed. Therefore, the contents of the modified processed speech are generally unintelligible.

  The third changing unit 46 repeats the above process for each signal of the change target part determined by the part determining unit 34, and outputs the signal thus processed to the delay adjusting unit 52.

  The output unit 50 acquires the signal of the change target portion subjected to the nonlinear processing from the nonlinear change unit 40 and the signal that is not the change target portion from the partial extraction unit 30. The output unit 50 converts them into analog signals and outputs them to the speaker SP via the power amplifier PA. The output unit 50 includes a delay adjustment unit 52 and a D / A unit 54.

  The delay adjustment unit 52 generates an output audio signal to be output by connecting the signal of the change target portion subjected to nonlinear processing and the signal that is not the change target portion. The delay adjustment unit 52 adjusts the timing at which the output audio signal is output from the output unit 50 according to the time required for propagation of the maskee H ′ (t). In particular, the delay adjustment unit 52 gives a predetermined delay to the output audio signal. This delay is within a range where the masker H (t) delay with respect to the masker H '(t) at the listener 8 position is such that the masker H' (t) and the masker H (t) can be said to be substantially real time. Set to fit.

  The fact that the maskee H '(t) and the masker H (t) are substantially in real time (semi-realtime) means that the maskee H' (t) and the masker H (t) are in the adjacent booth 2 '. At least partially overlapping. Alternatively, the signal of the change target portion output from the output unit 50 is converted into sound by the speaker SP, and the converted sound is received while the maskee H ′ (t) is received in the adjacent booth 2 ′. It is to be output to the adjacent booth 2 ′. Alternatively, the signal of the change target portion output from the output unit 50 is converted into sound by the speaker SP, and the converted sound is adjacent to the portion of the maskee H ′ (t) corresponding to the signal of the change target portion. This is to be output to the adjacent booth 2 ′ while listening in the booth 2 ′. In other words, the portion of the maskee H ′ (t) corresponding to the signal of the change target portion and the portion of the masker H (t) corresponding to the signal of the change target portion are at least partially in the adjacent booth 2 ′. It is to superimpose on.

  When the voice information secret system 100 is introduced, the positions of the microphone Mic and the speaker SP are determined, and the position of the assumed customer 6 and the assumed position of the listener 8 are also determined to some extent. In addition, the processing time in the SD controller unit SD can be estimated to some extent. Therefore, at the time of introducing the voice information secret talk system 100, the propagation time of the maskee H ′ (t) and the propagation time of the masker H (t) from the customer 6 to the listener 8 can be estimated to some extent. The delay in the delay adjusting unit 52 is set by calculating back from a desired value of the delay of the masker H (t) with respect to the maskee H ′ (t) at the listener 8 position.

If the masker H (t) has a large delay with respect to the maskee H ′ (t), echoes or reverberations may occur at the listener 8 position. Therefore, the delay adjusting unit 52 gives a delay to the output audio signal such that the delay of the masker H (t) with respect to the maskee H ′ (t) at the listener 8 position is a value that does not cause such a sense of incongruity. give. This delay is determined by experiment, but is typically about 100 msec or less.
Further, as described above, the present inventor has come up with the idea that it is advantageous to handle audio signals in units of approximately one mountain in order to control the understanding of audio information. From this point of view, it is preferable that the delay in the delay adjustment unit 52 be set to a value corresponding to the time width of the approximately one peak portion of the audio signal, particularly smaller than that. This is because an interaction between approximately one mountain portion of the masker and approximately one mountain portion of the musky is expected.

  The D / A unit 54 converts the output audio signal provided with the delay by the delay adjustment unit 52 into an analog audio signal for driving the speaker SP, and outputs the analog audio signal to the power amplifier PA.

  FIG. 8 is a waveform diagram showing waveforms of audio signals representing the maskee H ′ (t) and the masker H (t) at the listener 8 position. FIG. 8A is a waveform diagram showing the waveform of an audio signal representing the maskee H ′ (t). The waveform in FIG. 8 (a) is obtained by converting the original voice “ANO KARETOWA MOSOTONAGAINDAYO ZITSUWA” into a voice signal with the microphone Mic. In FIG. 8A, the vertical axis represents signal intensity in arbitrary units, and the horizontal axis represents time. FIG. 8B is a waveform diagram showing a waveform of an audio signal generated by processing the audio signal of FIG. 8A in the SD controller unit SD using the approximately one mountain extraction mode and the first change mode. It is. The portion indicated by N in the waveform shown in FIG. 8B corresponds to the portion indicated by M in the waveform shown in FIG. The same applies to FIG. The difference between the audio signal in FIG. 8B and the audio signal in FIG. 8C is the delay value provided by the delay adjustment unit 52.

  Comparing the envelope of FIG. 8A with the envelopes of FIG. 8B and FIG. 8C, it can be seen that there is not much change. In other words, there is not much change in voice intonation and intonation. However, when the audio signals of FIG. 8B and FIG. 8C are converted into audio by the speaker SP and output as the masker H (t), at the listener 8 site, the masky H ′ (t) and the masker H ( t) is synthesized and heard, and its meaning is difficult to understand. In other words, it is often “I don't know” (may be heard by other sounds).

  Depending on the positional relationship between the microphone Mic, the speaker SP, the customer 6, and the listener 8, the masker H (t) is positioned earlier than the maskee H '(t) when the delay adjustment unit 52 does not give a delay. May also be reached. That is, there is a case where the delay of the masker H (t) with respect to the maskee H ′ (t) at the listener 8 position is negative. If the delay given by the delay adjusting unit 52 is reduced, for example, as shown in FIG. 8C, the delay of the masker H (t) with respect to the maskee H ′ (t) at the listener 8 position is −D1 (D1 is Positive). In this case, the listener 8 is listening to the masker H (t) generated based on the future maskey H ′ (t) that has not yet been listened to.

  When the delay provided by the delay adjusting unit 52 is increased, the delay of the masker H (t) with respect to the maskee H ′ (t) at the listener 8 position becomes zero at a certain value, and then increases. For example, as shown in FIG. 8B, the delay of the masker H (t) with respect to the maskee H ′ (t) at the listener 8 position can be D2 (D2 is positive). From the viewpoint of temporal masking (temporal masking), there are cases where the masking effect is higher when the masker is slightly delayed than when the masker and the maskee are set at the same timing. This is because the auditory sense also has one aspect of understanding the content by changing the voice envelope over time. Therefore, in such a case, it is preferable to increase the delay given by the delay adjusting unit 52 so that the delay of the masker H (t) with respect to the maskee H ′ (t) at the listener 8 position is positive.

  Further, depending on the positional relationship between the microphone Mic, the speaker SP, the customer 6, and the listener 8, the masker H (t) is received much later than the maskee H ′ (t) when the delay adjusting unit 52 does not give a delay. The position of the listener 8 may be reached. In this case, in order to conceal the information by synthesizing the maskee H '(t) and the masker H (t) in the real time in the listener 8 position, the SD processing time in the SD controller unit SD is shortened. Must. In some cases, the accuracy of nonlinear processing must be sacrificed due to the existence of this time constraint, that is, the SD processing time must be shortened. However, the purpose of this embodiment is to reduce the intelligibility and intelligibility of speech, and the accuracy of the assumed / scheduled processing itself is not the purpose. Therefore, in this embodiment, if the condition that it becomes difficult to understand the meaning content of the maskee H ′ (t) due to the superposition of the maskers H (t), the processing accuracy does not become a big problem. This is because there are an infinite number of “conditions that make it difficult to understand the semantic content”.

  FIG. 9 is a flowchart showing a series of processes in the speech information secret system 100. The microphone Mic collects the maskee H ′ (t) and generates an audio signal (step 302). The A / D unit 20 acquires an audio signal representing the maskee H ′ (t) from the microphone Mic (step 304). The partial extraction unit 30 extracts the signal of the change target portion from the audio signal acquired by the A / D unit 20 and A / D converted (step 306). The nonlinear changing unit 40 changes the signal of the change target portion extracted by the partial extracting unit 30 using a nonlinear function (step 308). The output unit 50 outputs the signal of the change target portion changed by the nonlinear changing unit 40 to the speaker SP (step 310). The loudspeaker SP converts the received signal into a voice to obtain a masker H (t), and outputs the masker H (t) to the adjacent booth 2 'where the maskee H' (t) is being listened to (step 312).

  The operation of the speech information secret system 100 having the above configuration will be described. Consider a case in which a customer 6 sits in a bank booth 2 and consults with a bank counselor about, for example, a loan. At this time, it is assumed that there is a listener 8 in the adjacent booth 2 ′ next to the booth 2 and an application for opening an account is being made. Customer 6 explains the circumstances of applying for a loan, such as the worsening of the cash flow of his business. Of course, it is better for the listener 8 not to leak such a story. In particular, in the speech information secret speech system 100 according to the present embodiment, the non-linear processing of the speech being spoken by the customer 6 is mainly the listener in near real time. 8, the listener 8 cannot understand the utterance content of the customer 6. In addition, when there is no utterance by the customer 6, there is substantially no output from the speaker SP to the adjacent booth 2 ', so that the noise level in the adjacent booth 2' is not increased unnecessarily.

  In the above-described embodiment, the SD controller unit SD may include a storage device, and examples of such a storage device are a hard disk and a memory. In addition, based on the description of the present specification, each block is stored in a CPU (not shown), an installed application program module, a system program module, a memory that temporarily stores data read from the hard disk, or the like. It will be understood by those skilled in the art who have touched this specification that it can be realized.

  According to the speech information secret system 100 according to the present embodiment, the following operational effects can be obtained.

(1) According to the speech information secret system 100 according to the present embodiment, the content, that is, the information included in the conversation speech is concealed instead of concealing or deleting the presence of the conversation itself. In this regard, the inventor has recognized the following.
In open-plan offices, bank counters, and securities company lobby counters, especially at customer service counters that are partitioned by simple partitions, concealing the content of a conversation is possible if the contents of the conversation cannot be understood by anyone other than the person who is speaking. The point serves its purpose well. In other words, the voice itself may be heard as long as the content of the conversation is not leaked. Rather, when the presence of a speaker can be visually recognized, it is more natural to preserve the speech spectrum and envelope (sound quality, intonation, and intonation). The voice information secret system 100 according to the present embodiment corresponds to the above viewpoints and needs, and conceals conversation contents in a more natural manner.

  Although the envelope is preserved, in this embodiment, the degree of preservation is, for example, that the envelope of the masker is slightly shifted in time from the envelope of the maskee, or the shape of both envelopes is Allow a little different. That is, the masker envelope and the masky envelope are preserved to a similar degree. According to the inventor's experience as a person skilled in the art and preliminary experiments, if the masker envelope and the masky envelope are not equal, but are similar, the speech information disturbance effect Was found to be higher.

FIG. 10 is a graph showing the relationship between the difference between the masker and the maskee and the recognition rate. The vertical axis in FIG. 10 indicates the recognition rate in arbitrary units (in the example of FIG. 10, percentage (%)), and the horizontal axis indicates the degree of difference between the masker and the maskee in arbitrary units. The recognition rate is a recognition rate of independent words in a state where both the masker and the maskee are listening. Here, the difference between the masker and the maskee shows the difference in the envelope between the two.
If the difference between the masker and the maskee is close to zero, the recognition rate is high (almost 100%). Also, when the difference between the masker and the maskee is large, the recognition rate is high because the auditory sense makes it easy to distinguish between the two. The present inventor has conceived that, among them, the masker and the maskee are different from each other but are not easily distinguished, and therefore the recognition rate is the lowest. That is why hearing is tossed. In the present embodiment, for example, it is possible to adjust the delay in the delay adjustment unit 52 so that the difference between the masker and the maskee gives such a minimum value of the recognition rate.

  (2) For example, processing voice generated from a voice having a low relevance to Musky H ′ (t) received at the adjacent booth 2 ′, for example, a past voice, is sent to the adjacent booth 2 ′ for masking. When trying to do so, the information hiding effect cannot be obtained as much as expected due to the difference in the position of the silent part or the difference in articulation, and the unnaturalness increases. On the other hand, in the speech information secret system 100 according to the present embodiment, the non-linearly processed masker H (t) reaches the ear of the listener 8 substantially in real time with the masque H '(t). Therefore, compared with the above case, the degree of information concealment is high and the unnaturalness is low.

  (3) In the substantially single mountain mode of the embodiment, a substantially one mountain signal is extracted as the signal of the change target portion. In this case, the cut noise and the pasting are performed at a portion where the signal level of the maskee H ′ (t) is small, so that click noise due to nonlinear processing is reduced. That is, if the masker H ′ (t) is continuous in time, the masker H (t) is also substantially continuous. It is difficult for the envelope shape to collapse (intonation collapse) due to the windowing process.

  (4) The masker H (t) is created based on the speaker's own maskee H ′ (t), and is output from the speaker in parallel with the original voice. Therefore, especially in the first change mode and the second change mode, the spectrum and envelope of the maskee H ′ (t) can be preserved to some extent even if it becomes the masker H (t). As a result, since the spectrum and intonation of the masker H (t) are almost the same as those of the maskee H '(t), the sense of incongruity is naturally received by the listener.

  (5) When there is no maskee H '(t) on the time axis, that is, when there is no conversation, the masker H (t) is not output. That is, both overlap substantially in time. Therefore, an increase in the room noise level due to the masker H (t) during “no sound” when no sound is generated can be suppressed.

  (6) Dissipating a feeling of discomfort due to intermittent maskers and level fluctuations (disrupted when the conversation is stopped to reduced level) that may occur when using conventional technology, and other sounds (noise / music) that are unrelated to conversation This reduces the sense of discomfort for speakers, conversers, and other people in the room.

  (7) With respect to the physical sound insulation and private room formation in the prior art, no spatial blockage or movement is required, so that a feeling of openness and communication are hardly hindered.

  (8) Since the SD controller unit SD and the speaker SP are incorporated in the IT partition 4, the installation and installation of the system can be greatly simplified. In some cases, the microphone Mic may be incorporated in the IT partition 4. In this case, it is further simplified.

  (9) The IT partition 4 itself is subjected to sound absorption processing. Therefore, sound leakage to the adjacent booth can be reduced while increasing the clarity of the conversation voice in the booth.

  (10) The masker H (t) becomes a signal whose electrical signal correlation is not so high with the maskee H ′ (t) (original voice) by non-linear processing. Therefore, abnormalities due to feedback such as howling are less likely to occur during the operation of the speech information confidential system 100.

  (11) In the first change mode and the second change mode of the embodiment, it can be said that the carrier is subjected to nonlinear processing while the envelope of the audio signal representing the maskee H ′ (t) is substantially preserved. Therefore, such a change process can be performed in a short time.

  Heretofore, the configuration and operation of the audio information secret system 100 according to the embodiment and the SD controller unit SD included therein have been described. This embodiment is an exemplification, and it is understood by those skilled in the art that various modifications can be made to each component and combination of processes, and such modifications are within the scope of the present invention.

  In the embodiment, the case where the masker H (t) is output from one side of the adjacent booth has been described, but the present invention is not limited to this. For example, the masker H (t) may be output from the left and right sides of the adjacent booth by signal addition. FIG. 11 is a block diagram schematically showing the function and configuration of the audio information secret system according to the first modification. The audio information secret system according to the first modification includes a microphone Mic, an SD controller unit SD, four speakers SPa to SPd (SPd is not shown), and four power amplifiers PAa to PAd (PAd is not shown). Four adders 210a to 210d (210d not shown).

The audio signal that has undergone the processing in the SD controller unit SD is the adder 210a corresponding to the left speaker SPa of the booth 2, the adder 210b corresponding to the right speaker SPb of the booth 2, and the adjacent booth adjacent to the left of the booth 2. The signal is input to the adder 210c corresponding to the left speaker SPc of 2 ′ and the adder 210d (not shown) corresponding to the right speaker SPd (not shown) of the adjacent booth adjacent to the right of the booth 2. The audio signals input to the adders 210a to 210d are output from the speakers SPa to SPd via the corresponding power amplifiers PAa to PAd. The adder acquires and adds the audio signal that has undergone processing in the SD controller unit SD from the booths adjacent to the booth to which the speaker to which the speaker is connected outputs audio.
According to this modification, since the masker H (t) is output from both the left and right sides of the adjacent booth 2 ′, the conversation contents in the booth 2 are difficult to be transmitted to the listener 8.

Further, a PNC (Passive Noise Controller) may be used in combination to reduce the level of the maskee H ′ (t). The PNC intends to use a known ANC (Active Noise Control) adaptively at the time of adjustment, and to fix and use the set parameters at the time of operation.
FIG. 12 is a block diagram schematically showing the function and configuration of the speech information secret system according to the second modification. In this modification, the SD controller unit SD in FIG. 11 is replaced with a part surrounded by a broken line in FIG. In this part, the SD controller unit SD and the PNC unit PNC are provided in parallel, and an audio signal from the microphone Mic is input to the SD controller unit SD and the PNC unit PNC. A switch SW1 is provided on the output side of the SD controller unit SD, and the operation of the SD controller unit SD is controlled by the switch SW1. The output of the switch SW1 and the output of the PNC unit PNC are added by an adder 406 and output as sound from the speaker SP via the power amplifier PA.

  In this modification, a head torso simulator HATS (HATS: Head and Torso Simulator) connected to the sound source 402 via the amplifier 404 is placed at the speaker position P, and the PNC unit PNC is identified. The switch SW1 is opened to turn off the operation of the SD controller unit SD, and an appropriate audio signal is emitted from the HATS so that the output of the microphone Mic ′ placed at the listener position Q in the adjacent booth 2 ′ is minimized. System identification is performed by adaptively operating.

At this time, the impulse response including the microphone Mic and the speaker SP is −h (x), and the absolute value is substantially equal to that h (x) between the PNC speaker and the listener. Thereafter, the switch SW1 is closed, and the PNC unit is operated with the identified parameters fixed. Then, since the positions P and Q of the speaker and the listener and the positions of the microphone Mic and the speaker SP are substantially fixed, the level of the maskee H ′ (t) is effectively reduced, and the masker H (t) is dominant. Become. As a result, the effect of information masking is enhanced. If the level of the masker H (t) is lowered as necessary, the level of the entire system including the maskee H ′ (t), that is, the noise level in the room can be further reduced.
Note that the PNC function described above may be incorporated in a computer in which the SD controller unit SD is incorporated.

  Although ANC / PNC is an existing technology, it is not suitable for controlling a wide sound field all over three dimensions. On the other hand, in the case where the listener's head is present at a substantially fixed position in a narrow space surrounded by the partition of the counter, the sound reduction means is effective even in three dimensions.

  In processing the signal of the change target portion in the embodiment, a time window such as a Hanning window or zero cross detection may be used in combination to remove a click sound that may occur at the time of clipping. In this case, the uncomfortable feeling that can be given to the listener 8 or the people in the room is further reduced.

  In the embodiment, the partial extraction unit 30 has described the case where the signal of the change target portion is extracted from the audio signal in the approximately one-peak extraction mode or the random division mode, but the present invention is not limited thereto. For example, the partial extraction unit may exclude the silent part of the maskee H ′ (t) or a part below a certain level as the “silent part” from the change target part. The output unit 50 may output a part removed from the change target part as a silent part as it is as a silent part. In this case, an increase in the volume (sound pressure level) of the masker H (t) and thus the room noise level can be suppressed as much as possible. Conversely, when it is necessary to emphasize the disturbance effect, the extracted envelope may be subjected to processing such as logarithmic compression / decompression. The partial extraction unit may extract the entire audio signal as a signal of the change target portion.

  Although the present invention has been described based on the embodiments, the embodiments merely show the principle and application of the present invention, and the embodiments are defined in the claims. Needless to say, many modifications and arrangements can be made without departing from the spirit of the present invention.

  2 booths, 4 IT partitions, 6 customers, 8 listeners, 20 A / D sections, 30 partial extraction sections, 40 non-linear change sections, 50 output sections, 100 voice information secret talk system.

Claims (6)

A partial extractor for extracting a signal of a change target part from a voice signal representing a voice being uttered;
A non-linear changing unit that changes the signal of the change target portion extracted by the partial extracting unit using a non-linear function;
An output unit that outputs at least the signal of the change target portion changed by the non-linear change unit to a sound output unit capable of outputting sound to a region where the sound being uttered is received ;
The partial extraction unit is a signal in a section sandwiched between a first time before the peak of the envelope of the waveform of the audio signal and a second time after the peak, and a substantially one mountain-shaped signal, A voice changing device, characterized in that it is determined as a signal of a change target portion . The nonlinear changing unit is
From the signal of the change target portion extracted by the partial extraction unit, an envelope acquisition unit that acquires information indicating an envelope of the waveform;
A non-linear processing unit that processes the signal of the change target portion extracted by the partial extraction unit using a non-linear function,
The non-linear change unit is a change target extracted by the partial extraction unit based on information indicating the envelope acquired by the envelope acquisition unit and a signal of the change target part processed by the non-linear processing unit The voice changing device according to claim 1, wherein the signal of the portion is changed. The voice change device according to claim 1, wherein the non-linear changing unit performs formant transformation on the signal of the change target portion extracted by the partial extraction unit. The apparatus further comprises a timing adjusting unit that adjusts a timing at which the signal of the change target portion changed by the nonlinear changing unit is output from the output unit according to a time taken for propagation of the voice during the utterance. The voice changing device according to any one of claims 1 to 3 . Sound collecting means for receiving the voice being uttered and generating a voice signal representing the voice;
A sound changing device for changing a sound signal generated by the sound collecting means;
Voice output means for converting the voice signal changed by the voice changing device into voice and outputting the voice to the area where the voice being spoken is received,
The voice changing device is
A partial extractor for extracting a signal of a change target portion from the audio signal generated by the sound collecting means;
A non-linear changing unit that changes the signal of the change target portion extracted by the partial extracting unit using a non-linear function;
See containing at least an output unit for outputting a signal of the modified change target area by the non-linear changing unit to said audio output means, and
The partial extraction unit is a signal in a section sandwiched between a first time before the peak of the envelope of the waveform of the audio signal and a second time after the peak, and a substantially one mountain-shaped signal, A speech information secret talk system characterized by being determined as a signal of a part to be changed . Extracting a signal of a change target portion from a voice signal representing a voice being uttered;
Changing the extracted signal to be modified using a nonlinear function;
Converts a signal of the changed change target area in the voice, seen including the steps of: outputting the converted voice to the area where the sound in the speech is listening,
The extracting step is a signal in a section sandwiched between a first time before the peak of the envelope of the waveform of the audio signal and a second time after the peak, and a substantially one mountain-shaped signal, A voice changing method comprising a step of determining as a signal of a change target portion .

Images