JP2017211637A - Acoustic signal processing device, acoustic signal processing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To allow transmission of noise to be suppressed.SOLUTION: An acoustic signal processing device according to one embodiment includes: a beam former unit for generating a target signal, i.e. an acoustic signal corresponding to sound arriving from an object direction for each of a plurality of object directions on the basis of acoustic signals on a plurality of channels output by a plurality of microphones; a feature quantity calculation unit for calculating a feature quantity in each object direction on the basis of the target signal of each object direction; and a direction selection unit for selecting an output direction from among the object directions on the basis of the feature quantity of each object direction.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an acoustic signal processing device, an acoustic signal processing method, and a program.

  In a voice conference system or a video conference system (hereinafter referred to as “voice conference system”), a voice call is performed by transmitting voice collected by a microphone to a communication partner. In order to realize a comfortable voice call, it is important to collect the spoken voice with high accuracy. Conventionally, as a method for collecting spoken voice, a method is known in which a microphone having directivity is installed near a conference participant and only a microphone installed near the speaker is turned on. As another method, there is also known a method of installing microphones at a plurality of locations and mixing the outputs of all the microphones.

  However, the above-described conventional method has a problem that the microphone collects noise (sound of turning paper or moving chair), and the collected noise is transmitted to the communication partner.

  The present invention has been made in view of the above problems, and an object thereof is to suppress transmission of noise.

  An acoustic signal processing device according to an embodiment generates a target signal, which is an acoustic signal corresponding to a sound arriving from a target direction, for a plurality of target directions, based on a plurality of channel acoustic signals output from a plurality of microphones. A beamformer unit that calculates a feature value in each target direction based on the target signal in each target direction, and the plurality of targets based on the feature value in each target direction. A direction selection unit that selects an output direction from the directions.

  According to each embodiment of the present invention, transmission of noise can be suppressed.

The figure which shows an example of a function structure of the acoustic signal processing apparatus which concerns on 1st Embodiment. The figure explaining the distance from a sound source to two microphones. The figure which shows the acoustic signal from two microphones of FIG. The figure which shows an example of the hardware constitutions of the acoustic signal processing apparatus which concerns on 1st Embodiment. The flowchart which shows an example of operation | movement of the acoustic signal processing apparatus which concerns on 1st Embodiment. The figure explaining the specific example of operation | movement of the acoustic signal processing apparatus which concerns on 1st Embodiment. The figure which shows an example of a function structure of the acoustic signal processing apparatus which concerns on 2nd Embodiment. The flowchart which shows an example of operation | movement of the acoustic signal processing apparatus which concerns on 2nd Embodiment. The figure explaining the specific example of the calculation method of the feature-value in 3rd Embodiment. The flowchart which shows an example of the 1st selection method in 4th Embodiment. The flowchart which shows an example of the 2nd selection method in 4th Embodiment. The figure which shows the specific example of the 2nd selection method in 4th Embodiment.

  Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In addition, regarding the description of the specification and the drawings according to each embodiment, constituent elements having substantially the same functional configuration are denoted by the same reference numerals and overlapping description is omitted.

<First Embodiment>
The acoustic signal processing apparatus 1 according to the first embodiment will be described with reference to FIGS. The acoustic signal processing apparatus 1 according to the present embodiment collects sound with a plurality of omnidirectional (omnidirectional) microphones, executes predetermined processing on the acoustic signals of a plurality of channels output by the omnidirectional microphone, The presence or absence is determined and the output direction is selected. The acoustic signal processing apparatus 1 can be applied to an audio conference system or the like.

(Configuration of acoustic signal processing apparatus 1)
FIG. 1 is a diagram illustrating an example of a functional configuration of an acoustic signal processing device 1 according to the present embodiment. The acoustic signal processing device 1 in FIG. 1 includes a sound collection unit 11, an acoustic signal storage unit 12, a beam former unit 13, a candidate signal storage unit 14, a feature amount calculation unit 15, a feature amount storage unit 16, A direction selection unit 17 and an output unit 18 are provided.

  The sound collection unit 11 collects external sound and outputs an acoustic signal (electric signal) corresponding to the collected sound. The sound collection unit 11 does not have directivity and collects sound in all directions. As will be described later, the sound collection unit 11 is realized by a plurality of omnidirectional microphones. Therefore, the sound collection unit 11 outputs a plurality of channels of acoustic signals.

  The acoustic signal storage unit 12 stores the acoustic signals of a plurality of channels output from the sound collection unit 11 for each channel. The acoustic signal storage unit 12 includes, for example, a FIFO (First in First Out) ring buffer.

  The beam former unit 13 performs beam forming processing on the multi-channel acoustic signals output from the sound collecting unit 11. The beam forming processing includes processing for generating a target signal S in the target direction D based on acoustic signals of a plurality of channels, processing for generating a noise signal N by forming a blind spot (null point) in the target direction D, including.

  Here, an outline of the beam forming process will be described with reference to FIGS. FIG. 2 is a diagram for explaining the distance from the sound source SS to the two microphones M1 and M2. FIG. 3 is a diagram illustrating acoustic signals output from the microphones M1 and M2 of FIG.

  As shown in FIG. 2, when the sound source SS exists in a certain target direction D, the sound from the sound source SS is collected by the microphones M1 and M2. At this time, since the distance from the sound source SS to the microphone M1 is different from the distance from the sound source SS to the microphone M2, the sound from the sound source SS arrives at the microphones M1 and M2 at different timings.

  For example, in the example of FIG. 2, the distance from the sound source SS to the microphone M2 is longer by dsin θ than the distance from the sound source SS to the microphone M1. d is an installation interval of the microphones M1 and M2. θ is an angle formed by a line orthogonal to a straight line passing through the microphones M1 and M2 and a straight line passing through the sound source SS and the microphones M1 and M2.

  Therefore, the sound from the sound source SS arrives at the microphone M2 with a delay of dsin θ / c from the microphone M1. c is the speed of sound. As a result, as shown in FIG. 3, the microphone M2 outputs an acoustic signal corresponding to the sound from the sound source SS with a delay of dsin θ / c from the microphone M1.

  For this reason, by adding the acoustic signal output from the microphone M1 with a delay of dsin θ / c to the acoustic signal output from the microphone M2, a signal (target signal S) that emphasizes the acoustic signal corresponding to the sound from the sound source SS is added. ) Can be generated. Further, a signal (noise signal N) obtained by reducing the acoustic signal corresponding to the sound from the sound source SS by subtracting the acoustic signal output from the microphone M1 by delaying by dsin θ / c from the acoustic signal output from the microphone M2. Can be generated.

  As described above, the beamformer unit 13 generates the target signal S and the noise signal N in the target direction D by adding and subtracting the acoustic signals from the plurality of microphones with a predetermined delay time, respectively.

  In the example of FIG. 2, since it is assumed that the sound source SS is sufficiently far from the microphones M1 and M2, the straight line passing through the sound source SS and the microphone M1 is parallel to the straight line passing through the sound source SS and the microphone M2. is there. In the example of FIG. 2, the case where the sound collection unit 11 is realized by two microphones M1 and M2 is shown, but the sound collection unit 11 may be realized by three or more microphones.

  A plurality of target directions D are set in advance. Hereinafter, n target directions D from the first to the nth are set (n = 8 in the case of FIG. 6). The i-th target direction D is set as a target direction Di (1 ≦ i ≦ n). The target directions D1 to Dn are different directions. Further, the target signal S and the noise signal N in the target direction Di are set as the target signal Si and the noise signal Ni, respectively.

  The target signal Si is an acoustic signal corresponding to the sound arriving from the target direction Di. The processing for generating the target signal Si corresponds to processing for extracting the acoustic signal corresponding to the sound arriving from the target direction Di from the acoustic signals corresponding to the sound arriving from all directions output from the sound collecting unit 11. .

  The beamformer unit 13 generates the target signal Si by adding the plurality of channels of acoustic signals after delaying them by a predetermined delay time. That is, when sound arrives from the target direction Di, the beamformer unit 13 delays and adds the acoustic signals in each target direction Di so that the acoustic signals corresponding to the sound are emphasized. A signal Si is generated. The target signal Si generated in this way corresponds to the adder output of a delay and sum beamformer in which the beam point (high sensitivity direction) is formed in the target direction Di.

  The delay time of the acoustic signal is set in advance for each target direction Di so that the phases of sounds coming from the target direction Di match. As described above, the delay time can be set according to the installation interval d of the plurality of omnidirectional microphones constituting the sound collection unit 11, the target direction Di, the installation position of the omnidirectional microphone that outputs an acoustic signal, and the like. It is.

  When the beamformer unit 13 delays and adds the acoustic signals of a plurality of channels by a predetermined delay time as described above, the acoustic signal in the target direction Di is emphasized. On the other hand, the acoustic signals in directions other than the target direction Di are not enhanced as much as the acoustic signals in the target direction Di even if they are delayed and added by the same delay time. That is, as a result, a signal in which only the acoustic signal in the target direction Di is enhanced is obtained. The beam former 13 outputs the signal thus obtained as the target signal Si.

  The method for generating the target signal Si is not limited to the above method. For example, the beam former 13 may amplify the acoustic signal of each channel and adjust the signal level before adding the acoustic signal of each channel, or filter the acoustic signal of each channel to remove unnecessary frequencies. Components may be removed.

  The noise signal Ni is an acoustic signal corresponding to a sound arriving from a direction different from the target direction Di, that is, a direction other than the target direction Di. The noise signal Ni corresponds to a noise component that can be included in the target signal Si.

  In the present embodiment, the generation process of the noise signal Ni removes the acoustic signal corresponding to the sound arriving from the target direction Di from the acoustic signals corresponding to the sound arriving from all directions output from the sound collection unit 11. It corresponds to the processing. In other words, the generation process of the noise signal Ni corresponds to a process of acquiring an acoustic signal other than the target direction Di by forming a null point (low sensitivity direction) in the target direction Di.

  Specifically, in the present embodiment, the beamformer unit 13 subtracts the acoustic signals of a plurality of channels by delaying each by a predetermined delay time, thereby obtaining an acoustic signal corresponding to the sound arriving from the target direction Di. The noise signal Ni is generated by removing the noise signal Ni. The noise signal Si generated in this way corresponds to a subtracter output of a delay and sum beamformer having a null point formed in the target direction Di. The method for setting the delay time of the acoustic signal is the same as that for the target signal Si.

  When the beamformer unit 13 subtracts the sound signals of a plurality of channels as described above, the sound signal in the target direction Di is reduced. On the other hand, acoustic signals in directions other than the target direction Di are not reduced as much as acoustic signals in the target direction Di. As a result, a signal in which the acoustic signal in the target direction Di is reduced is obtained. The beam former 13 outputs the signal thus obtained as a noise signal Ni.

  Note that the method of generating the noise signal Ni is not limited to the above method. For example, before subtracting the acoustic signal of each channel, the beamformer unit 13 may amplify the acoustic signal of each channel to adjust the signal level, or filter the acoustic signal of each channel to remove unnecessary frequencies. Components may be removed.

  The beam former 13 performs beam forming processing for each target direction Di, and generates a target signal Si and a noise signal Ni, respectively. The beam former 13 may execute the beam forming process for each target direction Di in order or in parallel.

  The candidate signal storage unit 14 stores candidate signals for each target direction Di. The candidate signal is a signal that is a candidate for the output signal output by the output unit 18. In the present embodiment, the candidate signal storage unit 14 stores the target signal Si for each target direction Di generated by the beamformer unit 13 as a candidate signal for each target direction Di.

  The feature quantity calculation unit 15 calculates the feature quantity C in each target direction Di based on the target signal Si and noise signal Ni in each target direction Di. The feature amount C in the target direction Di is set as the feature amount Ci. The feature amount Ci is an arbitrary acoustic feature amount that can be calculated based on the target signal Si and the noise signal Ni that indicates the state of the sound in the target direction Di. The feature quantity Ci may be a time domain acoustic feature quantity or a frequency domain acoustic feature quantity. When calculating the feature quantity Ci in the frequency domain, it is preferable that the feature quantity calculation unit 15 is capable of executing a fast Fourier transform (FFT). Thereby, the feature amount Ci can be calculated in a short time.

  The feature amount storage unit 16 stores the feature amount Ci of each target direction Di calculated by the feature amount calculation unit 15.

  The direction selection unit 17 determines the presence or absence of a sound source in each target direction Di based on the feature amount Ci of each target direction Di. In addition, the direction selection unit 17 selects the output direction Dout based on the determination result of the presence or absence of the sound source and the feature amount Ci. The output direction Dout is the target direction Di of the candidate signal that the output unit 18 outputs as an output signal. The output direction Dout is any one of the n target directions Di.

  When it is determined that a sound source is present, the direction selection unit 17 selects, as the output direction Dout, the target direction Di having the maximum sound coming from the sound source among the target directions Di determined to be present. On the other hand, when the direction selection unit 17 determines that there is no sound source, the target direction Di having the smallest incoming sound in the target direction Di (the target direction Di having the smallest noise signal Ni in the present embodiment) is selected. The output direction Dout is selected.

  The output unit 18 outputs the candidate signal in the output direction Dout as an output signal. In the present embodiment, since the candidate signal is the target signal Si, the output unit 18 outputs the target signal Sout in the output direction Dout as an output signal. The output of the output signal includes outputting the output signal to an external device of the acoustic signal processing device 1 and outputting an output signal from the acoustic output device (speaker) of the acoustic signal processing device 1.

  FIG. 4 is a diagram illustrating an example of a hardware configuration of the acoustic signal processing device 1 according to the present embodiment. The acoustic signal processing device 1 in FIG. 4 includes a microphone array 100 and a computer 200.

  The microphone array 100 is an array composed of a plurality of omnidirectional microphones installed at a predetermined installation interval d, and is connected to the computer 200. Each omnidirectional microphone of the microphone array 100 outputs an acoustic signal corresponding to the collected sound. When the microphone array 100 includes M omnidirectional microphones, the microphone array 100 outputs an M-channel acoustic signal. The sound collection unit 11 is realized by the microphone array 100.

  The computer 200 includes a processor 201, a memory 202, a microphone interface 203, an input device 204, a display device 205, a communication device 206, a sound output device 207, and a bus 208.

  The processor 201 executes a program stored in the memory 202, and realizes the beam former unit 13, the feature amount calculation unit 15, the direction selection unit 17, and the output unit 18 of the acoustic signal processing device 1. The processor 201 is, for example, a central processing unit (CPU), a digital signal processor (DSP), an application specified integrated circuit (ASIC), a field programmable gate array (FPGA), or a programmable logic device (PLD).

  The memory 202 stores programs executed by the processor 201 and various data. The memory 202 implements the acoustic signal storage unit 12, the candidate signal storage unit 14, and the feature amount storage unit 16. The memory 202 is, for example, a RAM (Random Access Memory), a DRAM (Dynamic RAM), an SRAM (Static RAM), a hard disk, an optical disk, a flash memory, or the like.

  The microphone interface 203 is an interface that mediates communication between the microphone array 100 and the computer 200. The microphone interface 203 is an analog front end (AFE) including an AD (Analog to Digital) converter. The microphone interface 203 converts an analog signal (acoustic signal) output from the microphone array 100 into a digital signal and stores it in the memory 202. The microphone interface 203 inputs a control signal from the processor 201 to the microphone array 100. The microphone interface 203 may be realized by a program (software).

  The input device 204 is, for example, a keyboard, a mouse, a push button, a touch panel, or the like. A user can operate the acoustic signal processing device 1 via the input device 204.

  The display device 205 is, for example, a liquid crystal display, a plasma display, a cathode ray tube display, a lamp, or the like. The display device 205 may display the output direction Dout and the like.

  The communication device 206 is, for example, a modem, a hub, and a router. The output signal of the acoustic signal processing device 1 is transmitted to an external device via the communication device 206.

  The sound output device 207 is a speaker, a buzzer, or the like. The sound output device 207 may output the output signal of the sound signal processing device 1. When the acoustic signal processing device 1 is applied to an audio conference system or the like, the acoustic output device 207 outputs an acoustic signal received from a communication partner.

  The bus 208 connects the processor 201, the memory 202, the microphone interface 203, the input device 204, the display device 205, the communication device 206, and the sound output device 207 to each other.

(Operation of the acoustic signal processing apparatus 1)
FIG. 5 is a flowchart showing an example of the operation of the acoustic signal processing apparatus 1 according to the present embodiment. The acoustic signal processing apparatus 1 performs the operation of FIG. 5 at predetermined time intervals. In the following, it is assumed that the acoustic signal processing device 1 operates in units of frames having a predetermined time width. Let p be the frame number of the current frame. Further, it is assumed that the sound collection unit 11 constantly outputs an acoustic signal during the operation of the acoustic signal processing device 1.

  When the start time of the current frame (p) comes, first, the acoustic signal storage unit 12 stores the acoustic signal output by the sound collection unit 11 for each channel (step ST101). In the example of FIG. 5, the acoustic signal storage unit 12 only needs to be able to store an acoustic signal for one frame or more.

  Next, the beam former unit 13 reads out the acoustic signal of each channel from the acoustic signal storage unit 12, and executes beam forming processing for each target direction Di based on the read out acoustic signal. Thereby, the target signal Si and the noise signal Ni in each target direction Di in the current frame (p) are generated (step ST102). The beam former 13 passes the generated target signal Si and noise signal Ni in each target direction Di to the feature amount calculator 15. The beam forming process is as described above.

  Further, the beam former unit 13 stores the generated target signal Si as a candidate signal in the candidate signal storage unit 14. The candidate signal storage unit 14 stores the stored candidate signal (target signal Si) for each target direction Di (step ST103).

  Subsequently, the feature quantity calculation unit 15 calculates the feature quantity Ci of each target direction Di based on the target signal Si and noise signal Ni of each target direction Di received from the beam former unit 13 (step ST104). The feature amount Ci of the current frame (p) is defined as a feature amount Ci (p).

  Hereinafter, a case where the characteristic amount C is a signal to noise ratio (SNR) will be described as an example. SNR is the ratio of the signal component to the noise component. The SNR corresponds to the volume of sound coming from the sound source. Hereinafter, the feature amount Ci (p) of the current frame (p) is assumed to be SNRi (p). SNRi (p) is calculated by the following equation.

  In Equation (1), f is the frequency, Si (f) is the signal level of the component having the frequency f included in the target signal Si, Ni (f) is the signal level of the component having the frequency f included in the noise signal Ni, fmin is a lower limit frequency, and fmax is an upper limit frequency. SNRi (p) calculated by Expression (1) is an SNR that is band-limited to a band from the lower limit frequency fmin to the upper limit frequency fmax.

  The lower limit frequency fmin can be set to an arbitrary frequency. For example, the lower limit frequency fmin may be set to 20 Hz (the lower limit frequency of the audible region). Since the calculation amount of SNRi (p) decreases as the lower limit frequency fmin increases, the calculation process of SNRi (p) can be speeded up.

  The upper limit frequency fmax can be set to an arbitrary frequency. For example, the upper limit frequency fmax may be set to 20 kHz (upper limit frequency of the audible region). As the upper limit frequency fmax is lower, the calculation amount of SNRi (p) is reduced, so that the calculation process of SNRi (p) can be speeded up.

  The upper limit frequency fmax is preferably set so that no spatial aliasing occurs in the target signal Si and the noise signal Ni. Spatial aliasing here refers to signal aliasing caused by the installation interval d of omnidirectional microphones.

  Specifically, the upper limit frequency fmax is preferably set to be lower than the lower limit frequency fnyq (first frequency) at which spatial aliasing occurs (fmax <fnyq). The lower limit frequency fnyq at which spatial aliasing occurs is calculated by the following equation.

  In Expression (2), d is the installation interval of the omnidirectional microphone, and c is the speed of sound. Spatial aliasing occurs in the frequency component of the target signal Si and the noise signal Ni having a frequency f equal to or higher than the lower limit frequency fnyq. On the other hand, spatial aliasing does not occur in the frequency component having the frequency f less than the lower limit frequency fnyq of the target signal Si and the noise signal Ni.

  Therefore, by setting the upper limit frequency fmax to be less than the lower limit frequency fnyq, the SNRi (p) can be calculated using the frequency components of the target signal Si and the noise signal Ni in which no spatial aliasing has occurred. As a result, SNRi (p) can be calculated with high accuracy.

  The feature quantity calculation unit 15 stores the calculated SNRi (p) of each target direction Di in the feature quantity storage unit 16. The feature amount storage unit 16 stores the stored SNRi (p) for each target direction Di. In the example of FIG. 5, the feature quantity storage unit 16 stores SNRi (feature quantity Ci) for j frames (j> 1).

  Next, the direction selection unit 17 reads the SNRi (p) of the current frame (p) and the SNRi (p−1) of the previous frame (p−1) for each target direction Di from the feature amount storage unit 16, and the SNRi It is determined whether the amount of change in SNRi (p) with respect to (p-1) is greater than or equal to a preset threshold value TH1 (step ST105). That is, the direction selection unit 17 determines whether the following expression is satisfied.

  This determination corresponds to processing for detecting a change in the state of the sound in the target direction Di. More specifically, this corresponds to processing for detecting the generation and disappearance of a sound source in the target direction Di. The reason is as follows.

  If no sound source exists in the target direction Di in the previous frame (p−1), SNRi (p−1) is a small value. When a sound source is generated in the target direction Di in the current frame (p), SNRi (p) has a larger value than SNRi (p−1). As a result, SNRi (p) changes (increases) abruptly with respect to SNRi (p−1).

  When a sound source exists in the target direction Di in the previous frame (p−1), SNRi (p−1) has a large value. When the sound source disappears in the current direction (p) in the target direction Di, SNRi (p) becomes a smaller value than SNRi (p−1). As a result, SNRi (p) changes (decreases) sharply with respect to SNRi (p−1).

  If no sound source exists in the target direction Di in the previous frame (p−1), SNRi (p−1) is a small value. If no sound source exists in the target direction Di even in the current frame (p), SNRi (p) is a value that is substantially the same as SNRi (p−1). As a result, the amount of change in SNRi (p) with respect to SNRi (p−1) becomes small.

  When a sound source exists in the target direction Di in the previous frame (p−1), SNRi (p−1) has a large value. When a sound source is present in the target direction Di even in the current frame (p), SNRi (p) is a value that is substantially the same as SNRi (p−1). As a result, the amount of change in SNRi (p) with respect to SNRi (p−1) becomes small.

  Therefore, by appropriately setting the threshold TH1, when SNRi (p) changes sharply with respect to SNRi (p-1), that is, when a sound source is generated in the target direction Di and when the sound source is generated from the target direction Di. When it disappears, it can be detected. An appropriate value of the threshold value TH1 may be determined by experiment.

  Next, the direction selection part 17 correct | amends SNRi (p) based on the determination result of step ST105, ie, the detection result of the state change of the sound of the object direction Di. The corrected SNRi (p) is defined as C_SNRi (p).

  When the change amount of SNRi (p) with respect to SNRi (p−1) is less than the threshold TH1 (NO in step ST105), the direction selection unit 17 sets C_SNRi (p) to the average value of SNRi for j frames. (Step ST106). Let the mean value of SNRi for j frames be mean_SNRi (p). The direction selection unit 17 reads SNRi for j frames from the feature amount storage unit 16 and calculates mean_SNRi (p) by the following equation.

  The direction selection unit 17 stores the calculated mean_SNRi (p) in the feature amount storage unit 16 as C_SNRi (p). The feature amount storage unit 16 stores mean_SNRi (p) stored in the direction selection unit 17 as C_SNRi (p) (step ST107). Thereafter, the direction selection unit 17 selects the output direction Dout based on C_SNRi (p).

  In this way, by selecting the output direction Dout based on the averaged SNRi, erroneous selection of the output direction Dout due to instantaneous SNRi fluctuation can be suppressed. For example, it is possible to suppress erroneous selection caused by an instantaneous decrease in SNRi (p) due to the utterance of the speaker or the utterance of unvoiced sound. The erroneous selection here means that the direction in which the speaker is present is not selected as the output direction Dout.

  On the other hand, when the amount of change in SNRi (p) with respect to SNRi (p−1) is greater than or equal to threshold value TH1 (YES in step ST105), direction selection unit 17 sets C_SNRI (p) to SNRi (p) ( Step ST108). Specifically, the direction selection unit 17 stores SNRi (p) in the feature amount storage unit 16 as C_SNRi (p). The feature amount storage unit 16 stores the SNRi (p) stored in the direction selection unit 17 as C_SNRi (p) (step ST107).

  The direction selection unit 17 sets C_SNRi (p) for all target directions Di. Subsequently, the direction selection unit 17 reads C_SNRi (p) of each target direction Di from the feature amount storage unit 16, and the difference between the maximum value and the minimum value of C_SNRi (p) is greater than or equal to a preset threshold value TH2. It is determined whether there is any (step ST109). That is, the direction selection unit 17 determines whether the following expression is satisfied.

  In formula (5), max (C_SNRi (p)) is the maximum value of C_SNRi (p). min (C_SNRi (p)) is the minimum value of C_SNRi (p).

  This determination corresponds to a process for determining that a sound source exists. This is because when there is a sound source, the difference between SNRi (p) in the target direction Di where the sound source exists and SNRi (p) in the target direction Di where there is no sound source is large. By appropriately setting the threshold TH2, it can be determined that a sound source exists. An appropriate value of the threshold value TH2 may be determined by experiment.

  When the difference between the maximum value and the minimum value of C_SNRi (p) is greater than or equal to threshold value TH2 (YES in step ST109), direction selection unit 17 determines that a sound source exists. Then, the direction selection unit 17 selects the target direction Di having the maximum C_SNRi (p) as the output direction Dout (step ST110). The output direction Dout corresponds to the direction of the target direction Di where the sound coming from the sound source is the largest among the sound sources determined to be present.

  When the difference between the maximum value and the minimum value of C_SNRi (p) is less than the threshold value TH2 (NO in step ST109), the direction selection unit 17 sets the difference between the maximum value and the minimum value of C_SNRI (p) in advance. It is determined whether or not the threshold value TH3 is exceeded (step ST111). The threshold value TH3 is set smaller than the threshold value TH2 (TH2> TH3). That is, the direction selection unit 17 determines whether the following expression is satisfied.

  This determination corresponds to a process for determining that there is no sound source. This is because when there is no sound source, the difference in SNRi (p) in each target direction Di is small. By appropriately setting the threshold TH3, it can be determined that there is no sound source. An appropriate value for the threshold TH3 may be determined by experiment.

  When the difference between the maximum value and the minimum value of C_SNRi (p) is less than the threshold value TH3 (NO in step ST111), the direction selection unit 17 determines that there is no sound source. Then, the direction selection unit 17 selects the target direction Di having the smallest C_SNRi (p) as the output direction Dout (step ST112). This output direction Dout corresponds to the target direction Di with the smallest incoming noise.

  This is because in the absence of a sound source, all sounds coming from each target direction Di are noise, and the magnitude of C_SNRi (p) is considered to correspond to the magnitude of noise. .

  On the other hand, when the difference between the maximum value and the minimum value of C_SNRi (p) is greater than or equal to threshold value TH3 (YES in step ST111), direction selection unit 17 determines that the presence or absence of the sound source is unknown. Then, the direction selection unit 17 selects the target direction Di selected as the output direction Di in the previous frame (p-1) as the output direction Dout of the current frame (p) (step ST113).

  When selecting the output direction Dout, the direction selection unit 17 notifies the output unit 18 of the selected output direction Dout. The output unit 18 notified of the output direction Dout reads the candidate signal (target signal Sout) in the output direction Dout from the candidate signal storage unit 14, and outputs the read candidate signal as an output signal (step ST114).

  FIG. 6 is a diagram for explaining a specific example of the operation of the acoustic signal processing device 1. In the example of FIG. 6, eight target directions Di (i = 1 to 8) are set at equal intervals. Hereinafter, a case where the sound source SS is generated in the target direction D2 in the current frame (p) will be described as an example. It is assumed that no sound source SS exists in the previous frame (p-1).

  When the start time of the current frame (p) comes, the acoustic signal storage unit 12 stores the acoustic signal output from the sound collection unit 11 for each channel (step ST101). The acoustic signal of each channel includes an acoustic signal corresponding to the sound arriving from the sound source SS in the target direction D2.

  Next, the beam former unit 13 performs beam forming processing based on the acoustic signal stored in the acoustic signal storage unit 12, and generates the target signal Si and the noise signal Ni in each target direction Di (step ST102). . Thereby, target signals S1 to S8 and noise signals N1 to N8 are generated. Since the sound source SS exists in the target direction D2, the target signal S2 has a relatively large value. The target signals S1 to S8 are stored in the candidate signal storage unit 14 (step ST103).

  Subsequently, the feature amount calculator 15 calculates SNR1 (p) in the target direction D1 based on the target signal S1 and the noise signal N1 (step ST104). In the same way, the feature amount calculation unit 15 calculates SNR2 (p) to SNR8 (p). Since the target signal S2 has a relatively large value, SNR2 (p) has a relatively large value. The SNRi (p) calculated in this way is stored in the feature amount storage unit 16.

  The direction selection unit 17 reads SNRi (p) from the feature amount storage unit 16 and determines whether the change amount of SNRi (p) with respect to SNRi (p−1) is greater than or equal to the threshold TH1 for each target direction Di (step ST105).

  In the target directions D1, D3 to D8, since there is no sound source SS in the previous frame (p-1) or the current frame (p), the amount of change in SNRi (p) with respect to SNRi (p-1) is small, and the threshold value It becomes less than TH1 (NO in step ST105). Therefore, the direction selection unit 17 sets C_SNRi (p) of the target directions D1, D3 to D8 to mean_SNRi (p) (step ST106).

  On the other hand, in the target direction D2, since the SNRi (p) sharply increases with respect to the SNRi (p-1) due to the occurrence of the sound source SS in the current frame (p), the SNRi (p-1) The amount of change in SNRi (p) is greater than or equal to threshold value TH1 (YES in step ST105). Therefore, the direction selection unit 17 sets C_SNR2 (p) in the target direction D2 to SNR2 (p) (step ST108).

  The C_SNRi (p) of each target direction Di set in this way is stored in the feature amount storage unit 16 (step ST109).

  Next, the direction selection unit 17 determines whether the difference between the maximum value and the minimum value of C_SNRi (p) is greater than or equal to the threshold value TH2 (step ST109). The maximum value of C_SNRi (p) is C_SNR2 (p). The minimum value of C_SNRi (p) is one of C_SNR1 (p) and C_SNR3 (p) to C_SNR8 (p). Here, it is assumed that the minimum value is C_SNR1 (p).

  Since the difference between C_SNR2 (p) (= SNR2 (p)) and C_SNR1 (p) is greater than or equal to the threshold value TH2 (YES in step ST109), the direction selection unit 17 selects the target direction D2 as the output direction Dout. (Step ST110).

  The output unit 18 that is notified that the output direction Dout is the target direction D2 reads the target signal S2 in the target direction D2 from the candidate signal storage unit 14, and outputs the read target signal S2 as an output signal (step ST114). .

  As described above, when there is no sound source, the acoustic signal processing device 1 according to the present embodiment selects the target direction Di with the smallest noise as the output direction Dout, and outputs an output signal corresponding to the sound from the output direction Dout. Output. Thereby, the sound that has arrived from the target direction Di with the smallest noise is transmitted to the communication partner. As a result, noise transmission can be suppressed.

  Moreover, when the sound source exists, the acoustic signal processing device 1 according to the present embodiment selects the target direction Di where the sound source exists as the output direction Dout, and outputs an output signal corresponding to the sound from the output direction Dout. Thereby, the sound that has arrived from the target direction Di in which the sound source exists, that is, the sound that has arrived from the sound source is transmitted to the communication partner. As a result, sound arriving from the sound source can be collected with high accuracy and transmitted to the communication partner.

  When this acoustic signal processing device 1 is applied to an audio conference system or the like, the acoustic signal processing device 1 transmits a sound arriving from the target direction Di with the smallest noise to a communication partner when there is no speaker (sound source). . In addition, when there is a speaker (sound source), the acoustic signal processing device 1 transmits the voice spoken by the speaker to the communication partner. That is, the acoustic signal processing device 1 can suppress noise transmitted to the communication partner while accurately collecting the speech uttered by the speaker. As a result, the acoustic signal processing device 1 according to the present embodiment can realize a comfortable conference environment.

  Moreover, the acoustic signal processing device 1 according to the present embodiment collects sound with an omnidirectional microphone. Therefore, by applying the acoustic signal processing device 1, it is possible to configure an audio conference system or the like at a lower cost than a conventional audio conference system or the like that collects sound with a directional microphone.

  Further, the acoustic signal processing device 1 according to the present embodiment determines the presence / absence of a sound source in the target direction Di based on the change amount of the current (current frame) feature value Ci with respect to the past (previous frame) feature value Ci. . Therefore, the acoustic signal processing apparatus 1 can collect the sound when the sound source is generated without missing it. This is because when the sound source is generated, the acoustic signal rises steeply and the feature value Ci greatly changes.

  In addition, the acoustic signal processing device 1 according to the present embodiment corrects the current feature amount Ci in consideration of the feature amount Ci in a past fixed period (a period of j frames), and based on the corrected feature amount Ci. To determine the presence or absence of a sound source. Therefore, it is possible to suppress erroneous selection of the output direction Dout due to instantaneous fluctuation of the acoustic signal (feature amount Ci).

  In this embodiment, steps ST111 and ST113 may be omitted. Thereby, the calculation amount of the selection process of the output direction Dout can be reduced, and the selection process can be speeded up. When steps ST111 and ST113 are omitted, the presence or absence of a change in the direction of the sound source may be determined based on the determination in step ST109. That is, in the case of NO in step ST109, the process in step ST112 may be executed.

Second Embodiment
The acoustic signal processing apparatus 1 according to the second embodiment will be described with reference to FIGS. In the present embodiment, an acoustic signal processing device 1 having high directivity will be described.

(Configuration of acoustic signal processing apparatus 1)
FIG. 7 is a diagram illustrating an example of a functional configuration of the acoustic signal processing device 1 according to the present embodiment. The acoustic signal processing device 1 of FIG. 7 further includes a candidate signal generation unit 19. Other functional configurations and hardware configurations are the same as those in the first embodiment.

  The candidate signal generation unit 19 generates a candidate signal Oi in each target direction Di based on the target signal Si and noise signal Ni in each target direction Di generated by the beamformer unit 13, and uses the generated candidate signal Oi as a candidate signal. Store in the storage unit 14. In the present embodiment, the candidate signal storage unit 14 stores the candidate signal Oi for each target direction Di generated by the candidate signal generation unit 19.

  The candidate signal generation unit 19 generates the candidate signal Oi by removing the noise signal Ni (noise component) from the target signal Si. As a result, a candidate signal Oi having SNRi improved from the target signal Si is generated.

  The method of removing the noise signal Ni from the target signal Si is, for example, the MMSE-STSA method (Minimum Mean-Square-Error Short-Time Spectral Amplitude estimator), but is not limited thereto. The candidate signal generation unit 19 is realized by the processor 201 executing a program stored in the memory 202.

(Operation of the acoustic signal processing apparatus 1)
FIG. 8 is a flowchart showing an example of the operation of the acoustic signal processing apparatus 1 according to the present embodiment. The flowchart in FIG. 8 includes step ST115. Other steps are the same as in the first embodiment.

  In this embodiment, in step ST102, the beamformer unit 13 that has generated the target signal Si and noise signal Ni in each target direction Di generates the generated target signal Si and noise signal Ni in each target direction Di. 15 and the candidate signal generator 19.

  When receiving the target signal Si and the noise signal Ni in each target direction Di, the candidate signal generator 19 removes the noise signal Ni from the target signal Si in each target direction Di, thereby obtaining the candidate signal Oi in each target direction Di. Generate (step ST115). The method for generating the candidate signal Oi is as described above. Then, the candidate signal generation unit 19 stores the generated candidate signal Oi in the candidate signal storage unit 14. The candidate signal storage unit 14 stores the stored candidate signal Oi for each target direction Di (step ST103).

  The subsequent processing is the same as in the first embodiment. However, in this embodiment, in step ST114, the output unit 18 outputs the candidate signal Oout in the output direction Dout as an output signal.

  As described above, the acoustic signal processing device 1 according to the present embodiment can output the candidate signal Oi whose SNRi is improved from the target signal Si as an output signal. Therefore, according to the present embodiment, the directivity of the acoustic signal processing device 1 can be made sharper.

<Third Embodiment>
The acoustic signal processing apparatus 1 according to the third embodiment will be described with reference to FIG. In the present embodiment, an acoustic signal processing device 1 that can calculate a feature value Ci without using a noise signal Ni will be described.

  The functional configuration of the acoustic signal processing device 1 according to the present embodiment is the same as that of the first embodiment except for the beam former unit 13 and the feature amount calculation unit 15. The hardware configuration of the acoustic signal processing device 1 is the same as that of the first embodiment. Hereinafter, the beam former unit 13 and the feature amount calculation unit 15 in the present embodiment will be described.

  The beam former 13 generates a target signal Si in each target direction Di based on the acoustic signal. However, the beam former 13 does not generate the noise signal Ni.

  The feature amount calculation unit 15 performs the feature of the target direction Di based on the target signal Si in the target direction Di (first target direction) and the target signal Sj in the other target direction Dj (second target direction). The quantity Ci is calculated. In the present embodiment, the target signal Sj in the other target direction Dj is used as the noise signal Ni in the first embodiment, that is, the noise component of the target signal Si. The target signal Sj is an acoustic signal corresponding to a sound coming from a target direction Dj different from the target direction Di.

  The target direction Dj can be arbitrarily selected from n−1 target directions D excluding the target direction Di. However, since the target signal Sj is used as a noise component of the target signal Si, it is preferable that the target signal Sj does not include an acoustic signal corresponding to the sound arriving from the target direction Di. Therefore, the target direction Dj is preferably close to the direction opposite to the target direction Di (a direction different from the target direction Di by 180 degrees), and more preferably the direction opposite to the target direction Di. By using such a target signal Sj as a noise component of the target signal Si, the feature amount calculator 15 can calculate the feature amount Ci with high accuracy.

  The method for calculating the feature amount Ci is the same as that in the first embodiment. In the present embodiment, the noise signal Ni in the first embodiment may be read as the target signal Sj. For example, SNRi (p) in the target direction Di is calculated by the following equation.

  FIG. 9 is a diagram illustrating a specific example of the method for calculating the feature amount Ci in the present embodiment. In the example of FIG. 9, eight target directions Di (i = 1 to 8) are set at equal intervals. For example, the feature amount C2 in the target direction D2 is calculated based on the target signal S2 and the target signal S6. The target signal S6 is used as a noise component of the target signal S2. The feature amount C6 in the target direction D6 is calculated based on the target signal S6 and the target signal S2. The target signal S2 is used as a noise component of the target signal S6.

  As described above, according to the acoustic signal processing device 1 according to the present embodiment, the feature amount Ci can be calculated without using the noise signal Ni. Therefore, even when the noise signal Ni cannot be calculated by the beamformer unit 13 (the subtracter output of the delayed sum beamformer cannot be obtained), the feature amount Ci can be calculated.

  In the present embodiment, a part of the target signal Si may be used only as another target signal Sj (noise component). For example, in FIG. 9, it can be considered that the target signals S5 to S8 are used only as noise components of the target signals S1 to S4, respectively. In this case, the feature amounts C1 to C4 are calculated, and the feature amounts C5 to C8 are not calculated. Therefore, the output direction Dout is selected from the target directions D1 to D4.

  One target signal Sj may be used as a noise component of a plurality of target signals Si. For example, in FIG. 9, it can be considered that the target signal S8 is used as a noise component of the target signals S1 to S7. In this case, feature amounts C1 to C7 are calculated based on the target signal S8.

  The feature amount Ci may be calculated using noise components calculated from the plurality of target signals Sj. For example, in FIG. 9, it is conceivable to calculate the feature amount C1 by using a signal obtained by averaging the target signals S4 to S6 as a noise component.

<Fourth embodiment>
The acoustic signal processing apparatus 1 according to the fourth embodiment will be described with reference to FIGS. In the present embodiment, an acoustic signal processing device 1 that can detect a desired type of sound by using a plurality of feature amounts C will be described.

  The functional configuration of the acoustic signal processing apparatus 1 according to the present embodiment is the same as that of the first embodiment except for the feature amount calculation unit 15 and the direction selection unit 17. The hardware configuration of the acoustic signal processing device 1 is the same as that of the first embodiment. Hereinafter, the feature amount calculation unit 15 and the direction selection unit 17 in the present embodiment will be described.

  The feature amount calculation unit 15 calculates a plurality of feature amounts Ci for each target direction Di and stores them in the feature amount storage unit 16. In the following, it is assumed that Q feature quantities from the first to the Qth are calculated as the feature quantity Ci. The qth feature amount Ci in the target direction Di is set as a feature amount Ciq. The plurality of feature amounts Ciq are collectively referred to as a feature amount Ci. At this time, the feature value Ci is represented by a vector including Q feature values Ciq as follows.

  In the present embodiment, the feature quantity Ciq is not limited to an acoustic feature quantity such as SNR. The feature quantity Ciq may be a higher-order statistic or a discriminator score. Higher order statistics are, for example, cartesis, cumulant, and the like. The classifier is, for example, a hidden Markov model, a Gaussian Mixture Model (GMM), a deep neural network (DNN), or the like. It is preferable that the feature amount Ciq has a high correlation with the feature of the type of sound to be detected (for example, voice).

  The direction selection unit 17 selects the output direction Dout based on the feature amount Ci of each target direction Di. Hereinafter, two methods for selecting the output direction Dout will be described.

(First selection method)
First, a first method for selecting the output direction Dout using the feature amount Ci will be described. The first selection method is a method of calculating the evaluation value Vi of the target direction Di based on the feature amount Ci and selecting the output method Dout based on the evaluation value Vi.

  The evaluation value Vi is, for example, a linear sum of Q feature values Ciq, but is not limited thereto. When the evaluation value Vi is a linear sum, the evaluation value Vi is represented by the following expression, where wq is a weighting coefficient of the feature quantity Ciq.

  The evaluation value Vi is preferably calculated so as to be larger (or smaller) as the probability that the sound source of the sound to be detected exists is higher. For example, when it is desired to detect speech, the evaluation value Vi is calculated so as to increase (or decrease) as the probability that a speaker is present is higher. Such an evaluation value Vi is made possible by using, for example, the score of a discriminator that identifies the presence or absence of a speaker (acoustic signal corresponding to speech) as the feature amount Ciq. What is necessary is just to set the weighting coefficient w of Formula (9) appropriately by experiment.

  FIG. 10 is a flowchart illustrating an example of the first selection method. The process of FIG. 10 corresponds to the process after step ST104 in the first embodiment. Therefore, at the start of the process of FIG. 10, the feature amount storage unit 16 stores the feature amount Ci of each target direction Di. Hereinafter, a case where the acoustic signal processing device 1 detects sound will be described as an example.

  First, the direction selection unit 17 reads the feature amount Ci of the target direction Di from the feature amount storage unit 16, and calculates the evaluation value Vi of each target direction Di based on the read feature amount Ci (step ST201). Here, it is assumed that the evaluation value Vi increases as the probability that a speaker is present is higher.

  Next, the direction selection unit 17 determines whether the difference between the maximum value (max (Vi)) and the minimum value (min (Vi)) of the evaluation value Vi is equal to or greater than a preset threshold value TH4 (step ST202). ).

  This determination corresponds to a process for determining that a speaker is present. This is because when there is a speaker, the difference between the evaluation value Vi of the target direction Di where the speaker is present and the evaluation value Vi of the target direction Di where there is no speaker is large. By appropriately setting the threshold value TH4, it can be determined that a speaker is present. An appropriate value of the threshold value TH4 may be determined by experiment.

  When the difference between the maximum value and the minimum value of evaluation value Vi is equal to or greater than threshold value TH4 (YES in step ST202), direction selection unit 17 determines that there is a speaker. Then, the direction selection unit 17 selects the target direction Di having the maximum evaluation value Vi as the output direction Dout (step ST203). This output direction Dout corresponds to the direction of the speaker determined to exist.

  When the difference between the maximum value and the minimum value of the evaluation value Vi is less than the threshold value TH4 (NO in step ST202), the direction selection unit 17 sets the difference between the maximum value and the minimum value of the evaluation value Vi in advance. It is determined whether or not the threshold value is TH5 or more (step ST204). The threshold value TH5 is set smaller than the threshold value TH4 (TH4> TH5).

  This determination corresponds to a process for determining that there is no speaker. This is because when there is no speaker, the difference between the evaluation values Vi in each target direction Di is small. By appropriately setting the threshold TH5, it can be determined that there is no speaker. An appropriate value for the threshold TH5 may be determined by experiment.

  When the difference between the maximum value and the minimum value of evaluation value Vi is less than threshold value TH5 (NO in step ST204), direction selection unit 17 determines that there is no speaker. Then, the direction selection unit 17 selects the target direction Di having the smallest evaluation value Vi as the output direction Dout (step ST205). This output direction Dout corresponds to the target direction Di with the smallest incoming noise.

  This is because, when there is no speaker, the sounds coming from each target direction Di are all noise, and the magnitude of the evaluation value Vi corresponds to the high reception sensitivity of noise. Because.

  On the other hand, when the difference between the maximum value and the minimum value of the evaluation value Vi is equal to or greater than the threshold value TH5 (YES in step ST204), the direction selection unit 17 determines that the presence or absence of the speaker is unknown. Then, the direction selection unit 17 selects the target direction Di selected as the output direction Di in the previous frame (p-1) as the output direction Dout of the current frame (p) (step ST206).

  When selecting the output direction Dout, the direction selection unit 17 notifies the output unit 18 of the selected output direction Dout. The output unit 18 notified of the output direction Dout reads the candidate signal in the output direction Dout from the candidate signal storage unit 14, and outputs the read candidate signal as an output signal (step ST207).

  In the first selection method, steps ST204 and ST106 may be omitted. Thereby, the calculation amount of the selection process of the output direction Dout can be reduced, and the selection process can be speeded up. When steps ST204 and ST206 are omitted, the presence or absence of a change in the direction of the speaker may be determined based on the determination in step ST202. That is, in the case of NO in step ST202, the process in step ST205 may be executed.

  In the example of FIG. 10, the direction selection unit 17 determines the presence / absence of a speaker by relative evaluation of the evaluation value Vi, but may determine the presence / absence of a speaker by absolute evaluation of the evaluation value Vi. In this case, the direction selection unit 17 determines that there is a speaker when at least one evaluation value Vi is equal to or greater than the threshold value TH6, and selects the target direction Di having the maximum evaluation value Vi as the output direction Dout. When all the evaluation values Vi are less than the threshold value TH7, the direction selection unit 17 determines that there is no speaker, and selects the target direction Di having the smallest evaluation value Vi as the output direction Dout. The threshold values TH6 and TH7 may be determined as appropriate values through experiments. The threshold values TH6 and TH7 may be the same or different.

(Second selection method)
Next, a second method for selecting the output direction Dout using the feature amount Ci will be described. The second selection method is a method in which the presence / absence of a sound source to be detected is determined and the selection of the output direction Dout is performed using different feature amounts Ciq.

  FIG. 11 is a flowchart illustrating an example of the second selection method. The process in FIG. 11 corresponds to the process after step ST104 in the first embodiment. Therefore, at the start of the process of FIG. 11, the feature amount storage unit 16 stores the feature amount Ci of each target direction Di. Hereinafter, a case where the acoustic signal processing device 1 detects sound will be described as an example.

  Further, it is assumed that the feature amount Ci in each target direction Di includes a feature amount Ci1 and a feature amount Ci2. The feature amount Ci1 is SNRi, and the feature amount Ci2 is a score Gi. The score G is a determination result indicating the presence or absence of a speaker in the target direction Di (whether or not sound is included in the sound that has arrived from the target direction Di). The score Gi indicates that there is a speaker in the target direction Di when the value is 1, and indicates that there is no speaker in the target direction Di when the value is 0. Such a score G is obtained by inputting the target signal Si or the like into the GMM.

  First, the direction selection unit 17 reads the feature amount Ci of the target direction Di from the feature amount storage unit 16, and determines whether or not there is a speaker based on each read score Gi (feature amount Ci2) (step ST301). . The direction selection unit 17 determines that the speaker is present when one or more scores G having a value of 1 exist. On the other hand, if there is no score G having a value of 1, the direction selection unit 17 determines that there is no speaker.

  When a speaker is present (YES in step ST301), the direction selection unit 17 selects the target direction Di having the maximum SNRi as the output direction Dout from the target directions Di having a score Gi value of 1 (step ST30). ST302).

  On the other hand, when there is no speaker (NO in step ST301), the direction selection unit 17 selects the target direction Di having the smallest SNRi as the output direction Dout (step ST303).

  When selecting the output direction Dout, the direction selection unit 17 notifies the output unit 18 of the selected output direction Dout. The output unit 18 notified of the output direction Dout reads the candidate signal in the output direction Dout from the candidate signal storage unit 14, and outputs the read candidate signal as an output signal (step ST304).

  FIG. 12 is a diagram illustrating a specific example of the second selection method. In the example of FIG. 12, eight target directions Di (i = 1 to 8) are set at equal intervals. A sound source SS2 (for example, a speaker) that is not a speaker is present in the target direction D2. Due to the sound coming from the sound source SS2, the target direction D2 has a score G2 (feature amount C22) of 0 and an SNR2 (feature amount C21) of 20.

  A sound source SS5 is present in the target direction D5. The sound source SS5 is a speaker. Due to the sound (voice) coming from the sound source SS5, the target direction D5 has a score G5 (feature value C52) of 1 and an SNR5 (feature value C51) of 10. For simplicity of explanation, it is assumed that the scores Gi and SNRi of the other target directions Di are both 0.

  In this case, since the score G5 is 1, the direction selection unit 17 determines that there is a speaker (YES in step ST301). Then, the direction selection unit 17 selects, as the output direction Dout, the target direction D5 that is the target direction Di having the maximum SNRi among the target directions Di having the score G of 1 (step ST302). Thereafter, the output unit 18 outputs the candidate signal in the target direction D5 as an output signal (step ST304). That is, the acoustic signal processing device 1 outputs sound (voice) from the sound source SS5 (speaker).

  As described above, the acoustic signal processing apparatus 1 according to the present embodiment detects a desired type of sound (sound source) by using a plurality of feature amounts Ciq, and the target direction Di in which the sound (sound source) is detected. The output direction Dout can be selected from among the above. As a result, even when noise (sound of a different type from the desired type of sound) has arrived, the target direction Di in which the sound source of the desired type of sound exists is selected as the output direction Dout. Can do.

  For example, in the example of FIG. 12, the SNR2 of the noise from the sound source SS2 is larger than the SNR5 of the sound from the sound source SS5. However, as described above, the acoustic signal processing device 1 selects the target direction D5 as the output direction Dout and outputs an output signal corresponding to the sound from the sound source SS5.

  When the acoustic signal processing apparatus 1 according to the present embodiment is applied to an audio conference system or the like, the acoustic signal processing apparatus 1 suppresses the influence of noise, accurately collects voices coming from a speaker, Can be sent.

  It should be noted that the present invention is not limited to the configuration shown here, such as a combination with other elements in the configuration described in the above embodiment. These points can be changed without departing from the spirit of the present invention, and can be appropriately determined according to the application form.

1: Acoustic signal processing device 11: sound collecting unit 12: acoustic signal storage unit 13: beam former unit 14: candidate signal storage unit 15: feature amount calculation unit 16: feature amount storage unit 17: direction selection unit 18: output unit 19 : Candidate signal generation unit 100: Microphone array 200: Computer

Japanese Patent No. 5377518

Claims (11)

A beamformer unit that generates a target signal, which is an acoustic signal corresponding to a sound arriving from a target direction, based on the acoustic signals of a plurality of channels output from a plurality of microphones, for the plurality of target directions;
A feature amount calculation unit that calculates a feature amount in each target direction based on the target signal in each target direction;
A direction selection unit that selects an output direction from the plurality of target directions based on the feature amount of each target direction;
An acoustic signal processing device comprising: The direction selection unit determines the presence or absence of a sound source in each target direction based on the feature amount, and when the sound source is determined to be present, selects the target direction with the maximum sound coming from the sound source, The acoustic signal processing device according to claim 1, wherein when it is determined that the sound source does not exist, the target direction with the smallest incoming sound is selected. The feature amount calculation unit generates a noise signal that is an acoustic signal corresponding to a sound arriving from a direction different from the target direction for each target direction,
The acoustic signal processing apparatus according to claim 1, wherein the feature amount calculation unit calculates the feature amount based on the target signal and the noise signal. The acoustic signal processing apparatus according to claim 3, further comprising a candidate signal generation unit that removes the noise signal from the target signal. The feature quantity calculation unit calculates the feature quantity based on the target signal in a first target direction and the target signal in a second target direction different from the first target direction. The acoustic signal processing device according to claim 1 or 2. The acoustic signal processing apparatus according to claim 1, wherein the feature amount includes a signal-to-noise ratio. The feature amount calculation unit calculates the feature amount based on a frequency component of the target signal equal to or lower than a first frequency corresponding to an installation interval of the plurality of microphones. The acoustic signal processing device according to item 1. The acoustic signal processing device according to claim 1, wherein the feature amount calculation unit calculates a plurality of the feature amounts for each of the target directions. The direction selection unit averages the current feature amount when the change amount of the current feature amount with respect to the previous feature amount is less than a predetermined threshold, and outputs the output based on the averaged feature amount. The acoustic signal processing device according to any one of claims 1 to 8, wherein a direction is selected. Generating a target signal, which is an acoustic signal corresponding to a sound arriving from a target direction, for a plurality of the target directions based on the multi-channel acoustic signals output by the plurality of microphones;
Calculating a feature quantity of each target direction based on the target signal of each target direction;
Selecting an output direction from among the plurality of target directions based on the feature amount of each target direction;
An acoustic signal processing method. Generating a target signal, which is an acoustic signal corresponding to a sound arriving from a target direction, for a plurality of the target directions based on the multi-channel acoustic signals output by the plurality of microphones;
Calculating a feature quantity of each target direction based on the target signal of each target direction;
Selecting an output direction from among the plurality of target directions based on the feature amount of each target direction;
A program that causes a computer to execute.

Images