JP4973287B2 - Sound processing apparatus and program - Google Patents

Description

  The present invention detects a section of a target sound on the time axis from a mixed sound of sound coming from an intended sound source (hereinafter referred to as “target sound”) and sound other than the target sound (hereinafter referred to as “interfering sound”). Regarding technology.

Conventionally, a technique for detecting a voice section on a time axis from a mixed sound of voice and noise has been proposed. For example, Patent Document 1 discloses a technique for distinguishing speech from noise based on the shape (flatness) of the frequency spectrum of the input speech. Patent Document 2 discloses a technology for distinguishing between speech and noise according to the pitch (number of zero crossings) of the input speech.
JP 2004-272052 A Japanese Patent Laid-Open No. 1-286643

  In the techniques of Patent Document 1 and Patent Document 2, both are distinguished based on the difference in acoustic characteristics between speech and noise. Therefore, when other sounds are mixed with the sound to be detected as the target sound, a voice section other than the target sound is also detected in addition to the target sound section. That is, there is a problem that only the target sound section cannot be detected with high accuracy. In view of the above circumstances, an object of the present invention is to solve the problem of detecting a section of a target sound with high accuracy even when a sound other than the target sound exists.

In order to solve the above-described problems, a sound processing apparatus according to one aspect of the present invention includes each frame of a sound signal generated by a first sound collector and a second sound collector spaced from the first sound collector. By comparing the component values at each of a plurality of frequencies with each frame of the generated sound signal, the plurality of frequencies of each frame are sorted into a dominant frequency where the target sound is dominant and an inferior frequency where the target sound is inferior. and selection means, for each of a plurality of frames, the number of dominant frequency selection means has selected for that frame (e.g., FIG. 3 and the number Ns of FIG. 7) and inferior number of frequencies (e.g., the number Ni of FIG. 3 and FIG. 7) based on the bets, the frame is provided with a section detecting means for determining whether the frame outer frame or target sound period in the target sound section.

  In the above configuration, the number of dominant frequencies and the number of inferior frequencies selected by comparing the sound signal generated by the first sound collector and the sound signal generated by the second sound collector for each frame are obtained. Since the target sound section is detected based on the target sound section, it is possible to detect the target sound section with high accuracy even when the interfering sound is a noise such as an environmental sound or a human voice. The target sound is not limited to human voice. Further, the characteristics (directivity) of the first sound collector and the second sound collector are not required in the present invention, but an omnidirectional or substantially omnidirectional microphone is particularly suitable.

  In a preferred aspect of the present invention, the section detection means determines that a frame in which a numerical value obtained by subtracting the number of inferior frequencies from the number of dominant frequencies exceeds a threshold is a frame in the section of the target sound. According to this aspect, since the target sound section is detected based on the difference value between the number of dominant frequencies and the number of inferior frequencies, it is possible to reduce the amount of processing for detecting the target sound section. is there.

  The sound processing apparatus according to a preferred aspect of the present invention is a smoothing means for smoothing a change on the time axis from one of the dominant frequency and the inferior frequency in each of the plurality of frequencies (for example, step SB4 in FIG. 7). And a first change number (for example, change number Ms_i in FIG. 7) corresponding to the number of frequencies changed from the dominant frequency to the inferior frequency by smoothing by the smoothing means, and a frequency changed from the inferior frequency to the dominant frequency by smoothing. Counting means (for example, steps SB5 to SB8 in FIG. 7) for counting a second change number (for example, the change number Mi_s in FIG. 7) corresponding to the number of the first and second intervals, Based on a first relative ratio (for example, relative ratio R1 in FIG. 7) that is a ratio of two changes and a second relative ratio (for example, relative ratio R2 in FIG. 7) that is the ratio of the first number of changes to the number of dominant frequencies. To check the target sound section. To. According to this aspect, since the section of the target sound is detected based on the number of frequencies changed from one of the dominant frequency and the inferior frequency by the smoothing by the smoothing means, the erroneous selection by the selection means is compensated. The target sound section can be detected with high accuracy.

  In a further preferred aspect, the section detecting means determines that a frame in which a numerical value obtained by subtracting the second relative ratio from the first relative ratio exceeds a threshold is a frame in the target sound section. According to this aspect, since the target sound section is detected based on the difference value between the first relative ratio and the second relative ratio, it is possible to reduce the amount of processing for detecting the target sound section. is there.

The sound processing apparatus according to the present invention is realized by hardware (electronic circuit) such as a DSP (Digital Signal Processor) dedicated to each processing, and a general-purpose arithmetic processing apparatus such as a CPU (Central Processing Unit) and a program. It is also realized through collaboration with. The program according to the present invention is provided for each of a plurality of frequencies in each frame of the sound signal generated by the first sound collector and each frame of the sound signal generated by the second sound collector separated from the first sound collector. by comparing the component values, a plurality of frequencies of each frame, a distinguishing processing target sound dominant dominant frequency and the target sound is sorted into the inferior recessive frequency, for each of a plurality of frames, the at sorting process based on the number of number and recessive frequency of dominant frequencies were selected for a frame, the frame is performed and the segment detection process to determine whether the frame outer frame or target sound period in the target sound section in the computer. With the above program, the same operations and effects as the sound processing apparatus according to the present invention are exhibited. The program of the present invention is provided to a user in a form stored in a portable recording medium such as a CD-ROM and installed in a computer, or provided from a server device in a form of distribution via a network. Installed on the computer.

The present invention is also specified as a method for detecting the target sound section. In the sound processing method according to one aspect of the present invention, each frame of the sound signal generated by the first sound collector and each frame of the sound signal generated by the second sound collector separated from the first sound collector; By comparing the component values in each of the plurality of frequencies, the plurality of frequencies of each frame are sorted into the dominant frequency where the target sound is dominant and the inferior frequency where the target sound is inferior, and for each of the plurality of frames, based on the number of number and recessive frequency of dominant frequencies were selected for the target frame, the frame is to determine whether the frame outer frame or target sound period in the target sound section. Also by the above method, the effect | action and effect similar to the sound processing apparatus concerning this invention are show | played.

<A: Configuration and operation of sound processing device>
FIG. 1 is a block diagram showing the configuration of the sound processing apparatus according to the first embodiment of the present invention. The sound processing device 10 is a device that separates a target sound and an interfering sound and detects a section of the target sound on the time axis. As shown in FIG. 1, a first sound collector 31 and a second sound collector 32 are connected to the sound processing apparatus 10. Each of the first sound collector 31 and the second sound collector 32 is an omnidirectional or substantially omnidirectional microphone that collects ambient sounds in which target sound and interference sound are mixed. The first sound collector 31 generates a sound signal SA, and the second sound collector 32 generates a sound signal SB.

  The 1st sound collector 31 and the 2nd sound collector 32 are arrange | positioned at intervals. The first sound collector 31 is closer to the target sound source 51 (for example, a speaker who is using the sound processing device 10) than the second sound collector 32. The second sound collector 32 is closer to the disturbing sound source 52 than the first sound collector 31.

  As shown in FIG. 1, the sound processing apparatus 10 includes a frequency analysis unit 11, a selection unit 13, a sound source separation unit 15, a section detection unit 17, and a processing unit 19. Each element of the sound processing device 10 may be realized by an arithmetic processing device such as a CPU executing a program, or may be realized by an electronic circuit such as a DSP (Digital Signal Processor) dedicated to sound processing. Also good. A configuration in which each of the above elements is partially mounted on a separate electronic circuit is also employed.

  The frequency analysis unit 11 specifies the power spectrum PA of the sound signal SA and the power spectrum PB of the sound signal SB. More specifically, the frequency analysis unit 11 specifies the power spectrum PA by performing frequency analysis such as discrete Fourier transform on each of a plurality of frames into which the sound signal SA is divided. The frequency analysis unit 11 specifies the power spectrum PB of each frame for the sound signal SB in the same manner. Each successive frame partially overlaps on the time axis.

  FIG. 2 is a schematic diagram of the power spectra PA and PB. As shown in the figure, the power spectrums PA and PB are expressed by K frequency bins corresponding to different frequencies F (F1 to FK) (K is a natural number of 2 or more). One frequency bin corresponding to the power spectrum PA is data including power at a frequency F (hereinafter referred to as “target frequency”) F corresponding to the frequency bin in the power spectrum PA.

  1 compares the power spectra PA and PB specified by the frequency analysis unit 11 to select all target frequencies F corresponding to the respective frequency bins at frequencies (hereinafter, “the target sound is dominant). A distinction is made between f s (referred to as “dominant frequency”) and f i where the target sound is inferior (hereinafter referred to as “dominant frequency”) fi. That is, as shown in FIG. 2, the selection unit 13 compares the power at the same frequency for the power spectra PA and PB for all the target frequencies F1 to FK, and selects the target frequency F with the high power of the power spectrum PA as the dominant frequency. While selecting fs, the target frequency F having a large power spectrum PB power is selected as the inferior frequency fi. Then, the selection unit 13 sets selection data (flag) D that designates which of the target frequencies F1 to FK of one frame is selected as the dominant frequency fs or the inferior frequency fi. That is, the sorting unit 13 sets the sorting data D to “1” for the target frequency F sorted to the dominant frequency fs, and sets the sorting data D to “0” for the target frequency F sorted to the inferior frequency fi. .

  The sound source separation unit 15 is a means for separating the target sound and the disturbing sound based on the result of selection by the selection unit 13. The sound source separation unit 15 of this embodiment generates a power spectrum QA in which the target sound is emphasized and a power spectrum QB in which the interference sound is emphasized (that is, the target sound is suppressed) from the power spectra PA and PB. More specifically, the sound source separation unit 15 arranges the frequency bins of the target frequency F (D = 1) selected by the selection unit 13 to the dominant frequency fs in the power spectrum PA along the frequency axis, thereby arranging the target sound. The power spectrum QA is generated, and the power spectrum QB of the interference sound is generated by arranging the frequency bins of the target frequency F (D = 0) selected by the selection unit 13 as the inferior frequency fi in the power spectrum PB. The frequency bin of the target frequency F selected as the inferior frequency fi in the power spectrum PA and the frequency bin of the target frequency F selected as the dominant frequency fs in the power spectrum PB are discarded. A configuration in which the sound source separation unit 15 generates and outputs only one of the power spectra QA and QB is also employed.

  The section detection unit 17 in FIG. 1 is means for detecting a section where the target sound exists on the time axis (hereinafter referred to as “target sound section”) based on the result of selection by the selection unit 13. More specifically, in the section detection unit 17, the number Ns of dominant frequencies fs (number of frequency bins of the power spectrum QA) is relative to the number Ni of inferior frequencies fi (number of frequency bins of the power spectrum QB). Many frames are determined as frames within the target sound section, and frames having a relatively small number Ns of dominant frequencies fs relative to the number Ni of inferior frequencies fi are determined as frames outside the target sound section. That is, the target sound section is defined in units of frames.

  FIG. 3 is a flowchart showing the contents of processing by the section detection unit 17. The process of FIG. 3 is executed each time the selecting unit 13 classifies the K target frequencies F (F1 to FK) into the dominant frequency fs and the inferior frequency fi for one frame. As illustrated in FIG. 3, the section detection unit 17 includes the total number of the target frequencies F selected by the selection unit 13 as the dominant frequency fs among the K target frequencies F (the total number of the selection data D set to “1”). ) Ns is counted (step SA1). Further, the section detection unit 17 counts the total number of target frequencies F selected as the inferior frequency fi among the K target frequencies F (the total number of selection data D set to “0”) Ni (step SA2). ). The sum of the number Ns of dominant frequencies fs and the number Ni of inferior frequencies fi is the total number K of target frequencies F (K = Ns + Ni).

  The section detection unit 17 calculates the index value A (A = Ns−Ni) by subtracting the number Ni calculated in step SA2 from the number Ns calculated in step SA1 (step SA3). Next, the section detection unit 17 determines whether or not the index value A calculated in step SA3 exceeds a predetermined threshold value TH (step SA4). The threshold value TH is set to zero, for example.

  If the result of step SA4 is affirmative (A> TH), the section detection unit 17 determines that the frame being processed at this stage is a frame within the target sound section (step SA5). On the other hand, if the result of step SA4 is negative (A ≦ TH), the section detection unit 17 determines that the current frame is a frame outside the target sound section (step SA6). Then, the section detection unit 17 notifies the processing unit 19 of the determination result of step SA5 or SA6 (step SA7).

  FIG. 4 is a schematic diagram showing a result of processing by the section detection unit 17. In the figure, the horizontal axis indicates time (number of frames), and the vertical axis indicates the number Ns of dominant frequencies fs and the number Ni of inferior frequencies fi. In FIG. 4, the target sound section (section hatched) detected by the section detection unit 17 is shown along the horizontal axis. As shown in the figure, a section in which the number Ns of dominant frequencies fs exceeds the number Ni of inferior frequencies fi is detected as a target sound section.

  The processing unit 19 in FIG. 1 is a unit that executes predetermined processing based on the power spectra QA and QB generated by the sound source separation unit 15 and the target sound section detected by the section detection unit 17. The processing unit 19 of the present embodiment executes a voice recognition process. In other words, the processing unit 19 generates a sound signal in the time domain (that is, a signal indicating the waveform of the target sound) by performing inverse discrete Fourier transform on the power spectrum QA, and the target sound section detected by the section detection unit 17 Speech recognition processing is executed for the components in the target.

  As described above, in the present embodiment, the number Ns of dominant frequencies fs selected based on the comparison between the sound signal SA generated by the first sound collector 31 and the sound signal SB generated by the second sound collector 32, and the inferior frequency. Since the target sound section is detected based on the number Ni of fi, the target sound section is detected with high accuracy even when the interfering sound is noise such as environmental sound as well as human speech. Therefore, it is possible to improve the accuracy of the voice recognition processing by the processing unit 19. Since the target sound is emphasized (separation between the target sound and the interference sound) by the sound source separation unit 15 in addition to the high-precision detection of the target sound section, the accuracy of the speech recognition processing for the target sound is improved. This is particularly noticeable.

  Further, since the result of the processing by the selection unit 13 is shared by the target sound enhancement in the sound source separation 15 and the detection of the target sound interval in the section detection unit 17, the enhancement of the target sound and the detection of the target sound section are separate. There is also an advantage that the processing amount in the sound processing apparatus 10 is reduced as compared with the configuration executed on the basis.

<B: Second Embodiment>
Next, the sound processing apparatus 10 according to the second embodiment of the present invention will be described. The configuration and operation of the sound processing apparatus 10 according to this embodiment are the same as those in the first embodiment except for the section detection unit 17. Therefore, hereinafter, the operation of the section detection unit 17 will be described mainly, and description of elements other than the section detection unit 17 will be omitted as appropriate.

  The result of the selection unit 13 selecting the target frequency F into the dominant frequency fs and the inferior frequency fi is based on the superiority or inferiority of the actual target sound and the interference sound due to the noise suddenly generated in the sound signals SA and SB. There may be a momentary reversal. For example, when the power at one target frequency F of the power spectrum PA exceeds the power spectrum PB due to the noise of the sound signal SA or SB in the jth frame, as shown in FIG. ) Although the target frequency F is the inferior frequency fi in the frames up to the (j + 1) th and subsequent frames, the target frequency F instantaneously becomes the dominant frequency fs in the jth frame. Selected. In other words, the target frequency F that should be selected as the inferior frequency fi for the jth frame is erroneously selected as the dominant frequency fs due to noise. Similarly, when the power at one target frequency F of the power spectrum PB instantaneously exceeds the power spectrum PA due to noise in the jth frame, as shown in FIG. In this frame, the target frequency F to be selected as the dominant frequency fs is erroneously selected as the inferior frequency fi. The section detection unit 17 of the present embodiment detects the target sound section by compensating for the erroneous selection as described above.

  FIG. 7 is a flowchart showing the contents of the operation of the section detection unit 17 according to this embodiment. For each of “2W + 1” frames from the (jW) th to the (j + W) th, the sorting unit 13 sorts the K target frequencies F1 to FK into the dominant frequency fs and the inferior frequency fi. Each time, the process of FIG. 7 is executed for the j-th frame located at the center on the time axis (W is a natural number).

  For the j-th frame, the section detecting unit 17 similarly to steps SA1 and SA2 in FIG. 3, the number Ns of dominant frequencies fs among the K target frequencies F (step SB1) and the total number of inferior frequencies fi. Ni (step SB2) is counted. In addition, the section detection unit 17 initializes a variable k for identifying any of the K target frequencies F1 to FK to “1” and initializes the number of changes Ms_i and Mi_s to “0” ( Step SB3).

  The section detection unit 17 sequentially executes the processing from step SB4 to step SB8 for the target frequency Fk, and then adds “1” to the variable k (step SB9). Further, the section detection unit 17 determines whether or not the variable k after addition in step SB9 exceeds the total number K of the target frequencies F (step SB10). If the result of step SB10 is negative, the section detection unit 17 executes the processing from step SB4 to step SB8 for the target frequency Fk corresponding to the updated variable k, and the result of step SB10 is positive. In step SB11, the process proceeds. That is, the processing from step SB4 to step SB8 is sequentially repeated for each of the K target frequencies F1 to FK.

  In step SB4, the section detection unit 17 changes the dominant frequency fs and the inferior frequency fi on the time axis from one to the other of the target frequency Fk from the (jW) th to the (j + W) th “j”. Smooth over 2W + 1 "frames. More specifically, the section detection unit 17 selects the selection data D (D (jW) to D (j + W) set for the target frequency Fk of each frame from the (jW) th to the (j + W) th frame. )) Is smoothed by a median filter. That is, the sorting data D located at the center ((k + 1) th) when the sorting data D (jW) to D (j + W) are arranged in order of magnitude is the dominant frequency of the jth frame. Calculated as updated data D (j) after updating in Fk.

  According to the smoothing using the median filter, fluctuations in the selection data D continuous over up to W frames within the range of “2W + 1” frames are removed. For example, the selection data D (j) of the j-th frame in FIG. 5 is changed from “1 (dominant frequency fs)” to “0 (inferior frequency fi)” by smoothing as shown by an arrow in FIG. . Further, the selection data D (j) of the jth frame in FIG. 6 changes from “0 (inferior frequency fi)” to “1 (dominant frequency fs)” by smoothing.

  When step SB4 is executed, the section detection unit 17 determines whether or not the selection data D (j) has changed from “1” to “0” due to smoothing (step SB5). When the result of step SB5 is affirmative, the section detection unit 17 adds “1” to the change number Ms_i (step SB6). On the other hand, if the result of step SB5 is negative, the section detection unit 17 proceeds to step SB7 without passing through step SB6.

  The selection data D (j) of the target frequency Fk, which was erroneously selected as the dominant frequency fs due to the instantaneous noise, although it should actually be selected as the inferior frequency fi, is indicated by an arrow in FIG. As described above, the smoothing in step SB4 changes from “1” to “0”. Therefore, the number of changes Ms_i at the stage where the processing from step SB4 to step SB8 is repeated K times corresponds to the number of target frequencies Fk misselected (the number of misselections) as the dominant frequency fs in one frame. .

  Next, the section detection unit 17 determines whether or not the selection data D (j) has changed from “0” to “1” by smoothing (step SB7), and only when the result of step SB7 is positive. Then, “1” is added to the number of changes Mi_s (step SB8). Actually, the selection data D (j) of the target frequency Fk, which has been erroneously selected as the inferior frequency fi in spite of being the dominant frequency fs, is smoothed from “0” to “1” as shown in FIG. Change. Accordingly, the number of changes Mi_s at the stage where the processing from step SB4 to step SB8 is repeated K times corresponds to the number of target frequencies Fk misselected (number of misselections).

  When the processes from step SB4 to SB8 are repeated K times (step SB10: YES), the section detection unit 17 calculates the relative ratios R1 and R2 (step SB11). The relative ratio R1 is a ratio (R1 = Mi_s / Ni) between the number of changes Mi_s counted in step SB8 and the number Ni of inferior frequencies fi counted in step SB2. The relative ratio R2 is a ratio (R2 = Ms_i / Ns) between the number of changes Ms_i counted in step SB6 and the number Ns of dominant frequencies fs counted in step SB1. Then, the section detection unit 17 calculates the index value B (B = R1-R2) by subtracting the relative ratio R2 from the relative ratio R1 (step SB12).

  Next, the section detection unit 17 determines whether or not the index value B calculated in step SB12 exceeds a predetermined threshold value TH (step SB13). The threshold value TH is set to zero, for example. If the result of step SB13 is affirmative (B> TH), the section detection unit 17 determines that the frame being processed at this stage is a frame within the target sound section (step SB14). On the other hand, when the result of step SB13 is negative (B ≦ TH), the section detection unit 17 determines that the current stage frame is a frame outside the target sound section (step SB15). Then, the section detection unit 17 notifies the processing unit 19 of the result of determination in step SB14 or step SB15, and ends the process of FIG. 7 (step SB16).

  FIG. 8 is a schematic diagram showing a result of processing by the section detection unit 17 of the present embodiment. In the figure, the vertical axis represents the numerical values of the relative ratios R1 and R2, and the horizontal axis represents time (the number of frames). A target sound section in which the relative ratio R1 exceeds the relative ratio R2 is shown along the horizontal axis. The sound signals SA and SB assumed in FIG. 8 have the same waveforms as those illustrated in FIG. As shown in FIG. 8, according to the present embodiment, the target sound section can be detected with high accuracy even when the interfering sound is human speech, as in the first embodiment. Moreover, in the present embodiment, since the target sound section is detected in consideration of the influence of misselection, there is an advantage that the target sound section can be detected with higher accuracy than in the first embodiment.

<C: Modification>
Various modifications can be made to the above embodiment. An example of a specific modification is as follows. In addition, you may combine each following aspect suitably.

(1) Modification 1
The method of determining the target sound section (determining whether each frame is within the target sound section) based on the result of selection by the selection unit 13 is appropriately changed. For example, in the first embodiment, a configuration is adopted in which the target sound section is detected based on the ratio of the number Ns of dominant frequencies fs and the number Ni of inferior frequencies fi. That is, the section detection unit 17 calculates the ratio of the number Ns to the number Ni as the index value A (A = Ns / Ni) in step SA3 in FIG. The threshold value TH in step SA4 in FIG. 3 is set to “1”, for example. With the above configuration, the same effect as that of the first embodiment can be obtained.

  Similarly, in the second embodiment, a configuration is adopted in which the target sound section is detected based on the ratio between the relative ratios R1 and R2. For example, the section detector 17 calculates the ratio of the relative ratio R1 to the relative ratio R2 as an index value B (B = R1 / R2) in step SB12 of FIG. The threshold value TH in step SB13 in FIG. 7 is set to “1”, for example. With the above configuration, the same effect as in the second embodiment can be obtained. As can be understood from the above examples, the section detection unit 17 according to a preferred aspect of the present invention is a means for detecting a target sound section based on the number Ns of dominant frequencies fs and the number Ni of inferior frequencies fi. It ’s enough.

(2) Modification 2
In the above configuration, the configuration in which the number Ns and Ni are counted for all the bands of the power spectrums PA and PB specified by the frequency analysis unit 11 is illustrated, but a specific band ( For example, the numbers Ns and Ni may be counted only for the target frequency F within the band from 300 Hz to 4000 Hz to which human voice mainly belongs.

A configuration is also employed in which the number Ns or Ni is counted by weighting each band to which the target frequency F belongs. For example, when n1 dominant frequencies fs belong to the band B1 and n2 dominant frequencies fs belong to the band B2 different from the band B1 in one frame, the weight value α1 and the band B2 corresponding to the band B1 are set. The corresponding weight value α2 is set separately, and the number Ns is calculated by the following equation (a).
Ns = n1 × α1 + n2 × α2 (a)
According to the above configuration, it is possible to detect a target sound section belonging to a specific band with particularly high accuracy.

(3) Modification 3
In each of the above embodiments, the configuration in which the threshold value TH is a fixed value is exemplified, but a configuration in which the threshold value TH is variably controlled is also employed. For example, the threshold TH is increased as the volume around the first sound collector 31 and the second sound collector 32 is increased, or the threshold TH is increased as the number of sound collectors connected to the sound processing device 10 is increased. The configuration to be adopted is adopted.

(4) Modification 4
The configuration in which the frame in which the index value A of the first embodiment and the index value B of the second embodiment exceed the threshold value TH is exemplified as the target sound section, but the frame of which the index value A or the index value B exceeds the threshold value TH A configuration is also employed in which a section wider than the section is detected as the target sound section. For example, a configuration in which a predetermined number of frames before the frame in which the index value B (or index value A) starts to exceed the threshold value TH is used as the starting point of the target sound interval, or the index value B (or index value A) exceeds the threshold value TH. A configuration is adopted in which a predetermined number of frames after the frame that starts to fall below the end point of the target sound section.

(5) Modification 5
In the second embodiment, the means for smoothing the variation of the selected data D over a plurality of frames is not limited to the median filter. For example, a Kalman filter may be used instead of the median filter. A configuration is also employed in which the moving average of the selection data D in a plurality of frames is the selection data D (j) of the jth frame.

(6) Modification 6
As a method of distinguishing the plurality of target frequencies F into the dominant frequency fs and the inferior frequency fi, a known technique can be arbitrarily adopted instead of the method exemplified in each of the above embodiments. For example, the technique disclosed in Japanese Patent Application Laid-Open No. 2006-197552 may be adopted for selecting the dominant frequency fs and the inferior frequency fi. More specifically, the first sound collector 31 and the second sound collector 32 are arranged in a direction perpendicular to the direction in which the target sound arrives. The selection unit 13 includes a power spectrum PA obtained by frequency analysis of the difference between the sound signal SA generated by the first sound collector 31 and the sound signal SB generated by the second sound collector 32, and a signal obtained by delaying the sound signal SA. The power spectrum PB obtained by frequency analysis of the difference from the sound signal SB is compared. The sorting unit 13 sorts the target frequency F whose power spectrum PA is smaller than the power spectrum PB into the dominant frequency fs and also selects the target frequency F whose power spectrum PB is smaller than the power spectrum PA as the inferior frequency. Select fi. The same effect can be obtained by the above method.

  Further, in each of the above embodiments, the configuration in which the power of each target frequency F in each of the sound signals SA and SB is compared is exemplified, but the component value (for comparison between the dominant frequency fs and the inferior frequency fi) ( The acoustic feature amount is not limited to power. For example, a configuration in which the target frequency F is selected as the dominant frequency fs and the inferior frequency fi by comparing the phases at the target frequencies F for each frame of the sound signal SA and each frame of the sound signal SB is also employed.

(7) Modification 7
The process executed by the processing unit 19 using the detection result by the section detection unit 17 is not limited to the voice recognition process. For example, the processing unit 19 may execute a process of reducing the volume of a section (mainly disturbing sound) other than the target sound section of the sound indicated by the power spectrum QA.

  Further, the processing unit 19 may execute a process (for example, a spectral subtraction method) for suppressing the interference sound from the sound in which the target sound and the interference sound are mixed. In the spectral subtraction method, the target sound is subtracted from the power spectrum QA (or PA) of the target sound by subtracting the product of the power spectrum QB in which the disturbing sound is dominant and a predetermined coefficient (hereinafter referred to as “suppression coefficient”). A power spectrum is generated. The processing unit 19 changes the suppression coefficient between the inside and the outside of the target sound section detected by the section detecting unit 17. For example, the suppression coefficient is increased on the inner side of the target sound section than on the outer side of the target sound section. When the disturbance sound outside the target sound section is excessively suppressed, the sound may be unnatural. As described above, according to the configuration in which the suppression coefficient is changed between the inside and outside of the target sound, the interference sound is effectively removed within the target sound section, and the natural sound in which the interference sound is appropriately suppressed outside the target sound section. There is an advantage that sound can be generated.

(8) Modification 8
You may combine the well-known method for detecting the area where an audio | voice exists in each of the above forms. For example, a pitch detector for detecting the pitch of the sound signal SA is added to the sound processing apparatus 10 of FIG. The section detection unit 17 notifies the processing unit 19 of a section in which the pitch detection unit has detected the pitch among the target sound sections detected in the processes of FIGS. 3 and 7 as a new target sound section. While a clear pitch is specified for human voices and musical instrument performance sounds, no clear pitch is detected for noise. Therefore, according to the above configuration, it is possible to detect the target sound section including the human voice and the performance sound of the musical instrument with high accuracy.

It is a block diagram which shows the structure of the sound processing apparatus which concerns on 1st Embodiment of this invention. It is a conceptual diagram for demonstrating selection with a dominant frequency and an inferior frequency. It is a flowchart which shows operation | movement of an area detection part. It is a conceptual diagram which shows the relationship between the number Ns and Ni, and a target sound area. It is a conceptual diagram for demonstrating the misselection by the selection part. It is a conceptual diagram for demonstrating the misselection by the selection part. It is a flowchart which shows operation | movement of an area detection part. It is a conceptual diagram which shows the relationship between relative ratio R1 and R2 and a target sound area.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Sound processing apparatus, 11 ... Frequency analysis part, 13 ... Sorting part, 15 ... Sound source separation part, 17 ... Section detection part, 19 ... Processing part.

Claims (6)

Comparing component values at each of a plurality of frequencies between each frame of the sound signal generated by the first sound collector and each frame of the sound signal generated by the second sound collector separated from the first sound collector. And a selection means for selecting the plurality of frequencies of each frame into a dominant frequency in which the target sound is dominant and an inferior frequency in which the target sound is inferior,
For each of a plurality of frames, said sorting means based on the number of number and recessive frequency of dominant frequencies were selected for the target frame, the frame is to determine whether the frame outer frame or target sound period in the target sound section A sound processing device comprising section detection means. The sound processing apparatus according to claim 1, wherein the section detection unit determines that a frame in which a value obtained by subtracting the number of inferior frequencies from the number of dominant frequencies exceeds a threshold is a frame in a section of the target sound. Smoothing means for smoothing a change on the time axis from one of the dominant frequency and the inferior frequency in each of the plurality of frequencies;
A first change number corresponding to the number of frequencies changed from the dominant frequency to the inferior frequency by smoothing by the smoothing means, and a second change number corresponding to the number of frequencies changed from the inferior frequency to the dominant frequency by the smoothing. And counting means for counting
The section detecting means includes
A target sound section is detected based on a first relative ratio, which is a ratio of the second change number to the number of inferior frequencies, and a second relative ratio, which is a ratio of the first change number to the number of dominant frequencies. The sound processing apparatus according to claim 1. The sound processing apparatus according to claim 3, wherein the section detection unit determines that a frame in which a numerical value obtained by subtracting the second relative ratio from the first relative ratio exceeds a threshold value is a frame in the target sound section.   The section detection means weights and adds the number of dominant frequencies in each of a plurality of bands by a weight value individually set for each band, and individually calculates the number of inferior frequencies in each of a plurality of bands for each band. Weighted addition with the set weight value
  The sound processing apparatus according to any one of claims 1 to 4. On the computer,
Comparing component values at each of a plurality of frequencies between each frame of the sound signal generated by the first sound collector and each frame of the sound signal generated by the second sound collector separated from the first sound collector. And a sorting process for sorting the plurality of frequencies of each frame into a dominant frequency where the target sound is dominant and an inferior frequency where the target sound is inferior,
For each of a plurality of frames, the sorting processing based on the number of number and recessive frequency of dominant frequencies were selected for that frame, the frame to determine the frame outer frame or target sound period in the target sound section A program that executes section detection processing and.

Images