JP5971047B2 - Audio signal processing apparatus, method and program - Google Patents

Description

  The present invention relates to an audio signal processing apparatus, method, and program, and can be applied to a communication apparatus or communication software that performs processing such as target audio section detection on an audio signal, such as a mobile phone or a video conference.

  Speech detection is to determine whether or not the speech signal section generated by the target speaker from the input signal (hereinafter, such speech signal is referred to as the target speech and the section is referred to as the target speech section), and the target speech is determined. This is a technique for distinguishing from a non-target voice section other than the section (in addition, a non-target voice is called a non-target voice). Based on the result of the speech detection, speech encoding processing, noise suppression processing, and the like are appropriately operated at a later stage, so that high accuracy is required for speech detection. As described in Patent Document 1, the general speech detection method is based on the assumption that the level of the target speech fluctuates and the level of the non-target speech section is steady. The instantaneous value is compared with the long-term average value, and the section in which the instantaneous value exceeds the long-term average value with a difference of a predetermined threshold or more is regarded as the target voice section.

  By the way, the non-target voice is divided into “interfering voice” which is a human voice other than the speaker and “background noise” such as office noise and road noise. Since the disturbing speech is human speech, the level fluctuation has the same behavior as that of the target speech. Therefore, in the conventional method, there is a problem that the disturbing speech section is included in the target speech section. For this reason, when this conventional method is applied to speech encoding processing, the characteristics of disturbing speech are also reflected in the encoded parameters. In addition, when this conventional method is applied to noise suppression processing, the signal in the disturbing speech section is not removed, and sufficient suppression performance cannot be obtained.

  Such a problem can be improved by changing the feature amount referred to by the target speech section detection unit to the coherence from the variation of the input speech signal level. In brief, coherence is a feature amount that means the arrival direction of an input signal. Assuming the use of mobile phones, etc., the speaker's voice (target voice) comes from the front, and the disturbing voice tends to come from other than the front. It is possible to distinguish between the target voice and the disturbing voice.

  FIG. 9 is a block diagram showing a configuration when coherence is used for the target speech segment detection function.

  Input signals s1 (n) and s2 (n) are acquired from each of the pair of microphones m_1 and m_2 via an AD converter (not shown). Note that n is an index indicating the input order of samples, and is expressed as a positive integer. In the text, it is assumed that the smaller n is the older input sample, and the larger n is the newer input sample.

The FFT unit 10 receives input signal sequences s1 (n) and s2 (n) from the microphones m_1 and m_2, and performs fast Fourier transform (or discrete Fourier transform) on the input signals s1 and s2. Thereby, the input signals s1 and s2 can be expressed in the frequency domain. In performing the Fast Fourier Transform, analysis frames FRAME1 (K) and FRAME2 (K) composed of predetermined N samples are configured and applied from the input signals s1 (n) and s2 (n). An example of constructing the analysis frame FRAME1 (K) from the input signal s1 (n) is shown in the following equation (1), and the analysis frame FRAME2 (K) is the same.

  K is an index indicating the order of frames and is expressed by a positive integer. In the text, it is assumed that the smaller the K, the older the analysis frame, and the larger, the newer the analysis frame. In the following description of the operation, it is assumed that the index representing the latest analysis frame to be analyzed is K unless otherwise specified.

  The FFT unit 10 performs fast Fourier transform processing for each analysis frame to convert the frequency domain signals X1 (f, K) and X2 (f, K) into the frequency domain signals X1 (f, K) obtained. And X2 (f, K) are given to the corresponding first directivity forming unit 11 and second directivity forming unit 12, respectively. Note that f is an index representing a frequency. X1 (f, K) is not a single value, but is composed of spectral components of a plurality of frequencies f1 to fm, as shown in equation (2). The same applies to X2 (f, K) and later-described B1 (f, K) and B2 (f, K).

X1 (f, K) = {(f1, K), (f2, K),..., (Fm, K)}
... (2)
The first directivity forming unit 11 forms a signal B1 (f, K) having strong directivity in a specific direction from the frequency domain signals X1 (f, K) and X2 (f, K), and the second directivity. The forming unit 12 forms a signal B2 (f, K) having strong directivity in a specific direction (different from the above-described specific direction) from the frequency domain signals X1 (f, K) and X2 (f, K). As a method for forming the signals B1 (f, K) and B2 (f, K) having strong directivity in a specific direction, an existing method can be applied. For example, the directivity is strong in the right direction by applying the expression (3). By applying B1 (f, K) and equation (4), B2 (f, K) having strong directivity in the left direction can be formed. In the equations (3) and (4), the frame index K is omitted because it is not involved in the calculation.

  The meaning of these equations will be described with reference to FIGS. 10 and 11 by taking the equation (3) as an example. It is assumed that sound waves arrive from the direction θ shown in FIG. 10A and are captured by a pair of microphones m_1 and m_2 that are installed at a distance l. At this time, there is a time difference until the sound wave reaches the pair of microphones m_1 and m_2. This arrival time difference τ is given by equation (5), where d = 1 × sin θ, where d is the sound path difference, and c is the sound speed.

τ = 1 × sin θ / c (5)
Incidentally, a signal s1 (t−τ) obtained by delaying the input signal s1 (n) by τ is the same signal as the input signal s2 (t). Therefore, the signal y (t) = s2 (t) −s1 (t−τ) taking the difference between them is a signal from which the sound coming from the θ direction is removed. As a result, the microphone arrays m_1 and m_2 have directivity characteristics as shown in FIG.

  In the above, the calculation in the time domain has been described, but the same can be said if it is performed in the frequency domain. The equations in this case are the above-described equations (3) and (4). As an example, it is assumed that the arrival direction θ is ± 90 degrees. That is, the directivity signal B1 (f) from the first directivity forming unit 11 has a strong directivity in the right direction as shown in FIG. The directivity signal B2 (f) has strong directivity in the left direction as shown in FIG.

A coherence COH is obtained by performing operations such as equations (6) and (7) in the coherence calculator 13 on the directivity signals B1 (f) and B2 (f) obtained as described above. It is done. B2 (f) ^* in the equation (6) is a conjugate complex number of B2 (f). Since the frame index K is not involved in the calculations of the expressions (6) and (7), the description of the frame index K is omitted in the expressions (6) and (7).

  As shown in FIG. 12, the target speech section detection unit 14 obtains coherence COH (K) (step S100), and compares the coherence COH (K) with the target speech section determination threshold Θ (step S101). If (K) is equal to or greater than the target speech segment determination threshold Θ, it is regarded as the target speech segment and 1.0 is substituted for the determination result variable VAD_RES (K) (step S102), and the coherence COH (K) is the target speech segment determination threshold. If it is smaller than Θ, it is regarded as a non-target speech section (interference speech, background noise section), and 0.0 is substituted for the determination result variable VAD_RES (K) (step S103), and the determination result variable VAD_RES (K) is output. (Step S104). Then, the process proceeds to the next frame (step S105). The subsequent speech encoding processing and noise suppression processing perform predetermined processing according to whether or not the target speech section is based on the result.

  Here, a brief description will be given of the background for detecting the target speech section based on the coherence COH. The concept of coherence COH can be paraphrased as the correlation between the signal coming from the right and the signal coming from the left (the above-mentioned expression (6) is an expression for calculating the correlation for a certain frequency component, and the expression (7) Calculating the average of the correlation values of the frequency components). Therefore, the case where the coherence COH is small is a case where the correlation between the two directivity signals B1 and B2 is small. Conversely, the case where the coherence COH is large can be paraphrased as a case where the correlation is large. The input signal when the correlation is small is when the input arrival direction is greatly deviated to the right or left, or when the signal is clear and has little regularity such as noise even if there is no deviation. Therefore, it can be said that the section where the coherence COH is small is a disturbing voice section or a background noise section (non-target voice section). On the other hand, when the value of the coherence COH is large, it can be said that there is no bias in the arrival direction, and therefore the input signal comes from the front. Now, since it is assumed that the target speech comes from the front, it can be said that it is the target speech section when the coherence COH is large.

  By the way, even if the coherence is the target speech section, the coherence value is small in the small amplitude part, so even the speech arriving from the front may be erroneously determined as the non-target speech section.

  In response to such a problem, the inventor of the present application proposes to add a hangover provision unit and hold the determination result that the target speech section is a predetermined time in a prior application (Japanese Patent Application No. 2012-151563). ing.

  FIG. 13 shows a configuration in which a hangover giving unit is added, and FIG. 14 shows an operation of the target speech section detection / hangover giving unit 15 in FIG.

  The configuration of FIG. 13 differs from the configuration of FIG. 9 in that a target speech segment detection / hangover provision unit 15 is provided instead of the target speech segment detection unit 14.

  When receiving the coherence COH (K) (step S100), the target speech section detection / hangover giving unit 15 compares the coherence COH (K) with the hangover counter initialization threshold Ψ (step S200). If the coherence COH (K) is equal to or greater than the threshold Ψ, it is determined that the target speech section is present, and 1.0 is assigned to the determination result variable VAD_RES (K), and the counter initial value LENGTH is substituted for the hangover counter counter. (Step S201). On the other hand, if the coherence COH (K) is smaller than the threshold Ψ, the target speech segment detection / hangover provision unit 15 compares the coherence COH (K) with the target speech segment determination threshold Φ (provided that Ψ> Φ) (step S202). ). If the coherence COH (K) is equal to or greater than the threshold Φ, only 1.0 is substituted into the determination result variable VAD_RES (K) without operating the hangover counter counter (step S203). If the coherence COH (K) is smaller than the threshold Φ, it is determined whether or not the hangover counter counter is positive (step S204). If the hangover counter counter is positive, even if the coherence COH (K) is small, it is determined as the target speech section, 1.0 is substituted for the determination result variable VAD_RES (K), and the hangover counter counter is decremented by one. To do. On the other hand, if the coherence COH (K) is smaller than the threshold Φ and the hangover counter counter is 0 or less, it is determined as a non-target speech section, and 0.0 is substituted for the determination result variable VAD_RES (K) (step) S206). Thereafter, the determination result variable VAD_RES (K) is output to the subsequent stage (step S104), and the process proceeds to the next frame (step S105).

JP-A-7-181991

  By the way, as shown in FIG. 15, if the direction of arrival of the target speech is close to the front, the coherence COH (K) becomes a large value as a whole, but the coherence COH (K) becomes a small value as it shifts to the side. Become. Therefore, when the arrival direction of the disturbing voice is lateral, the coherence COH (K) often falls below the threshold Φ, the hangover counter counter is decremented many times, and rapidly becomes a negative value. The hangover effect cannot be obtained in the middle of the voice section. On the other hand, when the disturbing voice comes from the front, the coherence COH (K) often exceeds the threshold Φ, so the hangover counter counter does not decrement and the hangover effect until after the target voice section ends. Therefore, the non-target speech section is erroneously determined as the target speech section.

  Therefore, an audio signal processing apparatus, method, and program capable of effectively operating the hangover imparting function without being affected by the arrival direction of the disturbing voice is desired.

  According to a first aspect of the present invention, in an audio signal processing device that separates a target audio segment and a non-target audio segment from an input audio signal, (1) a delay subtraction process is performed on the input audio signal, thereby A first directivity forming unit for forming a first directivity signal having a directivity characteristic having a blind spot, and (2) performing a delay subtraction process on the input audio signal, thereby Uses a second directivity forming section for forming a second directivity signal having a directivity characteristic having a blind spot in a different second predetermined orientation, and (3) using the first and second directivity signals. A coherence calculation unit for obtaining coherence, and (4) comparing the coherence with the target speech segment determination threshold value, and whether the input speech signal is a target speech segment arriving from the target direction or other non-target speech Determine whether it is a section and Hangover is greater than the target speech segment determination threshold, and the result of the comparison using the target speech segment determination threshold hangs even if the determination result changes from the target speech segment to the non-target speech segment. The target speech segment detection / hangover imparting unit that continues the determination result of the target speech segment by the over length, and (5) a feature amount that represents the switching frequency of the coherence change tendency in the non-target speech segment is obtained and obtained. And a hangover length control unit for controlling the hangover length applied by the target speech segment detection / hangover giving unit according to the feature amount.

  According to a second aspect of the present invention, there is provided an audio signal processing method for separating a target audio segment and a non-target audio segment from an input audio signal. (1) The first directivity forming unit performs a delay subtraction process on an input audio signal. As a result, a first directivity signal having a directivity characteristic having a blind spot in the first predetermined direction is formed. (2) The second directivity forming unit performs a delay subtraction process on the input audio signal. Thus, a second directivity signal having a directivity characteristic having a blind spot in a second predetermined orientation different from the first predetermined orientation is formed, and (3) a coherence calculation unit is configured to (4) The target speech segment detection / hangover imparting unit compares the coherence with the target speech segment determination threshold value, and the input speech signal arrives from the target direction. The target voice segment or other It is determined whether it is a target speech segment, and the coherence is compared with a hangover provision threshold value that is larger than the target speech segment determination threshold value. Even if the target speech section is changed, the determination result of the target speech section is continued by the hangover length. (5) The hangover length control unit represents the switching frequency of the coherence change tendency in the non-target speech section. The hangover length applied by the target speech segment detection / hangover imparting unit is controlled according to the obtained feature amount.

  The audio signal processing program according to the third aspect of the present invention is the first directivity in which the computer has (1) delayed directivity processing applied to the input audio signal to give a directivity characteristic having a blind spot in the first predetermined direction. A first directivity forming unit that forms a signal; and (2) performing a delay subtraction process on the input audio signal, thereby providing a directivity characteristic having a blind spot in a second predetermined direction different from the first predetermined direction. A second directivity forming unit that forms the given second directivity signal, (3) a coherence calculation unit that obtains coherence using the first and second directivity signals, and (4) the coherence The target speech segment determination threshold is compared to determine whether the input speech signal is a target speech segment arriving from the target orientation or any other non-target speech segment, and the coherence and the target speech segment are determined. C that is greater than the judgment threshold Compared with the threshold value for the gover assignment, even if the result of the comparison using the target speech segment determination threshold changes from the target speech segment to the non-target speech segment, the determination result of the target speech segment is continued by the hangover length. A target speech segment detection / hangover imparting unit, and (5) a feature amount representing a switching frequency of a coherence change tendency in the non-target speech segment is obtained, and the target speech segment detection is performed according to the obtained feature amount -It is made to function as a hangover length control part which controls the said hangover length which a hangover provision part applies.

  According to the audio signal processing apparatus, method, and program of the present invention, the hangover provision function can be used effectively without being affected by the arrival direction of the disturbing voice.

It is a block diagram which shows the structure of the audio | voice signal processing apparatus which concerns on 1st Embodiment. It is a block diagram which shows the detailed structure of the hangover length control part of FIG. It is explanatory drawing which shows the structural example of the memory | storage part of FIG. It is a flowchart which shows the operation | movement of the hangover length control part of FIG. It is a block diagram which shows the structure of the audio | voice signal processing apparatus which concerns on 2nd Embodiment. It is a block diagram which shows the internal structure of the hangover counter initialization threshold value control part in the audio | voice signal processing apparatus of 2nd Embodiment. It is explanatory drawing which shows the structural example of the initialization threshold value memory | storage part of FIG. It is a flowchart which shows operation | movement of the hangover counter initialization threshold value control part in the audio | voice signal processing apparatus of 2nd Embodiment. It is a block diagram which shows the structure in the case of using coherence for a target audio | voice detection function. It is explanatory drawing which shows the property of the directivity signal from the directivity formation part of FIG. It is explanatory drawing which shows the characteristic of the directivity by the two directivity formation parts of FIG. It is a flowchart which shows the process of the target audio | voice area detection part of FIG. It is a block diagram which shows the structure which added the hangover provision part to the structure of FIG. It is a flowchart which shows the operation | movement of the target speech area detection and hangover provision part of FIG. It is explanatory drawing which shows the relationship between the arrival direction of disturbance audio | voice, and the magnitude of coherence.

(A) First Embodiment Hereinafter, a first embodiment of an audio signal processing apparatus, method, and program according to the present invention will be described with reference to the drawings.

  The audio signal processing apparatus, method, and program according to the first embodiment are intended to obtain a hangover effect without excess or deficiency by appropriately controlling the hangover length according to the arrival direction of the disturbing voice.

(A-1) Configuration of the First Embodiment FIG. 1 is a block diagram showing the configuration of the audio signal processing device according to the first embodiment. Is shown. Here, the part excluding the pair of microphones m_1 and m_2 can be realized as software (audio signal processing program) executed by the CPU, but can be functionally represented in FIG.

  In FIG. 1, an audio signal processing apparatus 1 according to the first embodiment includes microphones m_1 and m_2, an FFT unit 10, a first directivity forming unit 11, a second directivity forming unit 12, and a coherence similar to those of the prior application configuration. In addition to the calculation unit 13 and the target speech section detection / hangover provision unit 15, a hangover length control unit 16 is provided.

  Here, the microphones m_1, m_2, the FFT unit 10, the first directivity forming unit 11, the second directivity forming unit 12, the coherence calculation unit 13, and the target speech segment detection / hangover providing unit 15 Since the same function is carried out, the function description is abbreviate | omitted.

  The hangover length control unit 16 is provided to solve the above-described problem. Based on the feature amount obtained from the coherence COH (K), the hangover length control unit 16 sets an appropriate hangover length LENGTH according to the arrival direction of the disturbing speech. It can be set. The hangover length control unit 16 estimates the arrival direction in the non-target voice section based on the coherence COH (K) and the target voice section determination result variable VAD_RES (K), and sets the hangover length LENGTH.

  The target speech segment detection / hangover provision unit 15 of the first embodiment performs a hangover provision operation by applying the hangover length LENGTH set by the hangover length control unit 16.

  FIG. 2 is a block diagram showing a detailed configuration of the hangover length control unit 16. In FIG. 2, the hangover length control unit 16 includes a coherence / speech section detection result receiving unit 21, a feature amount calculation control unit 22, a hangover length collation control unit 23, a non-target speech section feature amount calculation unit 24, a hangover length. A collation unit 25, a storage unit 26, and a hangover length transmission unit 27 are included.

  The coherence / speech interval detection result receiving unit 21 receives the coherence COH (K) from the coherence calculation unit and the determination result variable VAD_RES (K) from the target speech segment detection / hangover giving unit 15.

  The feature quantity calculation control unit 22 refers to the determination result variable VAD_RES (K) to recognize whether or not it is a non-target speech section, and hangover length matching control unit 23, non-target speech section feature quantity calculation unit 24, and hangover The long verification unit 25 is operated or is not operated.

  The non-target speech section feature quantity calculation unit 24 calculates the difference DIFF (K) between the coherence COH (K) in the current frame and the COH (K-1) in the previous frame (the difference DIFF (K) is the time difference). If it is divided by 1, it can be expressed as a differential coefficient), and the product of the difference DIFF (K) and the differential coefficient DIFF (K-1) of the immediately preceding interval is calculated, and the product is a negative value If so (meaning that the coherence change trend has decreased from an increase or has changed from a decrease to an increase), a variable (hereinafter referred to as a zero cross value) zero_cross representing the number of positive and negative inversions of the differential coefficient is increased. If the product is a positive value, the zero cross value zero_cross is not updated but is used as it is.

  The hangover length collation control unit 23 increases the counter variable timer that detects the collation period, and when the predetermined value LIMIT is reached, the hangover length collation unit 25 performs collation.

  As illustrated in FIG. 3, the storage unit 25 stores a hangover length LENGTH in association with the range of the zero cross value zero_cross. In the example of FIG. 3, when the zero cross value zero_cross is greater than or equal to A and less than B, α is associated with the hangover length LENGTH. When the zero cross value zero_cross is greater than or equal to B and less than C, β (> α ) And zero cross value zero_cross is C or more and less than D, γ (> β) is associated as the hangover length LENGTH. With this relationship, if the number of zero crossings of the differential coefficient is large (disturbance sound comes from the side), the hangover length LENGTH is increased and the number of zero crossings of the differential coefficient is small (disturbance from the front). Hangover length LENGTH can be shortened.

  Under the control of the hangover length collation control unit 23, the hangover length collation unit 25 collates the storage unit 26 using the zero cross value zero_cross obtained by the non-target speech section feature quantity calculation unit 24 as a key, and determines the hangover length LENGTH. To get.

  The hangover length transmission unit 27 gives the hangover length LENGTH obtained by the hangover length verification unit 25 to the target speech section detection / hangover provision unit 16.

(A-2) Operation of the First Embodiment Next, the operation of the audio signal processing device 1 of the first embodiment will be described in the order of overall operation and detailed operation in the hangover length control unit 16 with reference to the drawings. explain.

  The signals s1 (n) and s2 (n) input from the pair of microphones m_1 and m_2 are respectively converted from the time domain to the frequency domain signals X1 (f, K) and X2 (f, K) by the FFT unit 10. After that, directivity signals B1 (f, K) and B2 (f, K) having a blind spot in a predetermined direction are generated by the first and second directivity forming units 11 and 12, respectively. Then, the coherence calculation unit 13 applies the directivity signals B1 (f, K) and B2 (f, K) to execute the calculations of the equations (6) and (7), and the coherence COH (K) is calculated. Calculated.

  In the hangover length control unit 16, the hangover length LENGTH is determined by a method described later, and is given to the target speech section detection / hangover giving unit 15. The target speech segment detection / hangover giving unit 15 determines whether the target speech segment is a target speech segment or a non-target speech segment based on the coherence COH (K). The determination result variable VAD_RES (K) formed in this way is output to the subsequent stage.

  Next, the operation of the hangover length control unit 16 will be described. FIG. 4 is a flowchart showing the operation of the hangover length control unit 16.

  The hangover length control unit 16 estimates the arrival direction based on the positive and negative fluctuations of the differential coefficient of the coherence COH (K) in the non-target speech section (hereinafter referred to as “zero cross”), and determines the hangover length LENGTH. Set.

  Prior to the description of the operation of the hangover length control unit 16, the significance of providing the hangover length control unit 16 will be described. As described above, the strength of the correlation of coherence changes depending on the arrival direction, the correlation is stably strong near the front, and the fluctuation of the coherence is relatively gentle. On the other hand, if the direction of arrival deviates sideways, the correlation weakens and the coherence fluctuations are random and intense. When the direction of arrival of jamming speech is closer to the front, the fluctuation of the differential coefficient is gentle and the number of zero crosses is small because there are few positive and negative fluctuations. The number of zero crosses increases. Based on the above behavior, the hangover length can be controlled according to the arrival direction.

  In FIG. 4, when processing proceeds to a new frame, first, coherence COH (K) is received from the coherence calculation unit 13, and a determination result variable VAD_RES (K) is received from the target speech section detection / hangover provision unit 15 (step S300). . Then, it is determined whether or not the determination result variable VAD_RES (K) is a value representing a non-target speech section (step S301).

  If the determination result variable VAD_RES (K) is a value representing the target speech section, the counter variable timer and the differential coefficient DIFF that define the verification period are initialized (step S302), and the variable K that defines the frame is incremented by one. Then (step S303), the processing shifts to the next frame.

  On the other hand, when the received determination result variable VAD_RES (K) is a value representing the non-target speech section, the coherence COH (K) in the current frame and the COH (K−1) in the immediately preceding frame. The difference DIFF (K) is calculated (step S304), and then the product of the difference DIFF (K) and the differential coefficient DIFF (K-1) of the immediately preceding section is calculated to determine whether the product is positive or negative ( Step S305). If the product is a negative value, the zero cross value zero_cross is incremented by 1 (step S306). If the product is a positive value, the zero cross value zero_cross is not updated and is used as it is (step S307).

  If the determination result variable VAD_RES (K) is a value representing the target speech section, the counter variable timer for detecting the verification period is increased (step S308), and it is determined whether or not the predetermined value LIMIT has been reached. (Step S309). If the counter variable timer after the increase does not reach the predetermined value LIMIT, the variable K defining the frame is incremented by 1 (step S303), and the process proceeds to the next frame.

  If the counter variable timer is incremented by 1 per frame and the predetermined value LIMIT is set to 20, the predetermined value LIMIT is reached every 20 frames, and the hangover length verification operation described later is performed. Executed.

  When the incremented counter variable timer reaches a predetermined value LIMIT, it is compared with the stored contents of the storage unit 26, and the hangover length LENGTH corresponding to the zero cross value zero_cross at that time is obtained to detect the target speech section and give the hangover. The counter variable timer is initialized (step S310). Thereafter, the variable K defining the frame is incremented by 1 (step S303), and the process proceeds to the next frame.

(A-3) Effect of Embodiment As described above, according to the first embodiment, the hangover length is set to an appropriate value according to the arrival direction of the disturbing speech, so that the hangover without excess or deficiency is achieved. There is an effect.

  As a result, by applying the audio signal processing apparatus of the first embodiment to a communication apparatus such as a video conference system or a mobile phone, it is possible to expect improvement in call sound quality.

(B) Second Embodiment Next, a second embodiment of the audio signal processing apparatus, method and program according to the present invention will be described with reference to the drawings.

  The second embodiment is intended to solve the following inconveniences in addition to the inconveniences solved by the first embodiment.

  When background noise is superimposed, the coherence in the target speech section behaves such that the maximum value decreases and the minimum value increases. The reason why the maximum value is small is that the coherence value is calculated by reflecting the influence of background noise whose waveform regularity is lower than that of the audio signal. Also, the minimum value is large because the background noise is more regular than the silent case although the regularity is low.

  For this reason, depending on the background noise, the frequency at which the coherence COH (K) exceeds the hangover counter initialization threshold Ψ is reduced, and the hangover cannot be sufficiently applied, resulting in erroneous determination in the target speech section determination.

  In the second embodiment, in order to eliminate such inconvenience, the hangover counter initialization threshold Ψ is set by using the feature that the variation of the coherence COH (K) in the target speech section becomes small when background noise is superimposed. The adaptive control is performed according to the variation of the coherence COH (K) in the target speech section. In the second embodiment, dispersion is used as an index representing variation in coherence COH (K).

(B-1) Configuration of Second Embodiment FIG. 5 is a block diagram showing a configuration of an audio signal processing device according to the second embodiment, which is the same as FIG. 1 according to the first embodiment described above. Corresponding parts are denoted by the same reference numerals.

  In FIG. 5, the audio signal processing device 1 </ b> A according to the second embodiment includes a hangover counter initialization threshold value control unit 17 in addition to the same configuration as that of the first embodiment. A functional description of the same configuration as that of the first embodiment is omitted.

  The hangover counter initialization threshold value controller 17 calculates the coherence variance in the target speech section based on the coherence COH (K) and the determination result variable VAD_RES (K), and based on the calculated variance, the hangover counter An initialization threshold value Ψ is determined and set in the target speech segment detection / hangover provision unit 15.

  As described above, when background noise is superimposed, the maximum value of coherence in the target speech section becomes small and the minimum value becomes large. Therefore, it can be said that the coherence variance becomes small in the target speech section. Therefore, it is possible to determine that the background noise is not superimposed if the coherence variance in the target speech section is large, and that the background noise is superimposed if the variance is small. Therefore, if the hangover counter initialization threshold Ψ is controlled in accordance with the coherence variance value in the target speech section, erroneous determination of the target speech section when background noise is superimposed can be improved.

  FIG. 6 is a block diagram showing an internal configuration of the hangover counter initialization threshold value control unit 17.

  In FIG. 6, the hangover counter initialization threshold control unit 17 includes a coherence / determination result reception unit 31, a threshold update control unit 32, a variance calculation unit 33, a hangover counter initialization threshold verification unit 34, and an initialization threshold storage unit 35. And a hangover counter initialization threshold transmission unit 36.

  The coherence / determination result receiving unit 31 receives the coherence COH (K) from the coherence calculation unit 13 and the determination result variable VAD_RES (K) from the target speech section detection / hangover providing unit 15.

  The threshold update control unit 32 refers to the determination result variable VAD_RES (K) to determine whether or not it is the target speech segment, and only in the target speech segment, the variance calculation unit 33, the hangover counter initialization threshold collation unit 34, and the initial value The activation threshold storage unit 35 functions effectively. The threshold update control unit 32 maintains the immediately previous hangover counter initialization threshold in the non-target speech section.

  The variance calculator 33 calculates a variance variance (K) of coherence in the target speech section. Here, the time difference to the oldest sample may vary, but the variance may be calculated using the coherence of a predetermined number of samples, and the number of samples may vary, but the samples within a predetermined period The variance may be calculated using

  The initialization threshold value storage unit 35 stores the coherence variance variation range and the value of the hangover counter initialization threshold value Ψ in association with each other. FIG. 7 is an explanatory diagram illustrating a configuration example of the initialization threshold value storage unit 35. The range where the variance variation is P or more and less than Q is associated with δ as the value of the hangover counter initialization threshold Ψ, and the range where the variance variation is Q or more and less than R is ε ( > Δ), and the range where the variance variation is greater than or equal to R and less than S is associated with ζ (> ε) as the value of the hangover counter initialization threshold Ψ.

  With the above magnitude relationship, when the variance is small (background noise is superimposed), the hangover counter initialization threshold Ψ can be reduced, and the hangover effect in the target speech period can be reduced. Can be prevented from being damaged.

  The hangover counter initialization threshold value collation unit 34 collates the initialization threshold value storage unit 35 using the variance variation (K) of the coherence calculated by the variance calculation unit 33 as a key, and the range to which the value of the variance variation (K) belongs The value of the hangover counter initialization threshold Ψ associated with is extracted.

  The hangover counter initialization threshold value transmission unit 36 receives the value of the hangover counter initialization threshold value Ψ obtained by the hangover counter initialization threshold value comparison unit 34, or the hangover counter initialization threshold value Ψ for the immediately previous (K-1) frame. Is transmitted to the target speech segment detection / hangover assigning unit 15.

  The target speech section detection / hangover provision unit 15 of the second embodiment executes the hangover provision function by applying the hangover counter initialization threshold Ψ (K) from the hangover counter initialization threshold control unit 17. To do.

(B-2) Operation of Second Embodiment Next, the detailed operation of the hangover counter initialization threshold value collating unit 34 in the audio signal processing device 1A of the second embodiment will be described. FIG. 8 is a flowchart showing the operation of the hangover counter initialization threshold value control unit 17.

  The coherence COH (K) from the coherence calculation unit 13 and the determination result variable VAD_RES (K) from the target speech section detection / hangover provision unit 15 are received by the coherence / determination result reception unit 31 (step S400). Then, the threshold update control unit 32 refers to the determination result variable VAD_RES (K), and determines whether or not the target speech section is reached (step S401). In other words, this determination is a determination as to whether the target speech section is a review of the hangover counter initialization threshold Ψ or a non-target speech section in which the previous hangover counter initialization threshold Ψ is continued (applied).

  If the determination result variable VAD_RES (K) is a value indicating that it is the target speech section, the variance calculation unit 33 also uses the input coherence COH (K), and the variance variance of the coherence in the target speech section. (K) is calculated (step S402). Then, the hangover counter initialization threshold value collating unit 34 obtains the hangover counter initialization threshold value Ψ (K) corresponding to the calculated variance variation (K) from the initialization threshold value storage unit 35 (step S403).

  On the other hand, if the determination result variable VAD_RES (K) is a value indicating that it is the target speech section, the threshold update control unit 22 applied immediately before, but this time the hangover counter initialization threshold of the frame It is set as Ψ (K) (step S404).

  As described above, when the hangover counter initialization threshold value Ψ (K) for the current frame determined by the parameter K is obtained, the hangover counter initialization threshold value transmission unit 36 performs the target speech section detection / hangover provision unit 15. (Step S405), and then the process proceeds to the next frame (Step S406).

(B-3) Effects of Second Embodiment According to the second embodiment, the following effects can be obtained in addition to the same effects as those of the first embodiment.

  According to the second embodiment, the hangover counter initialization threshold can be set to an appropriate value according to the superimposition of background noise on the target speech, so that a hangover effect without excess or deficiency can be obtained.

  That is, according to the second embodiment, it is possible to achieve a hangover effect without excess or deficiency regardless of the arrival direction of the disturbing speech and regardless of the degree of the background noise superimposed on the target speech.

(C) Other Embodiments In the first embodiment, the hangover length is set based on the zero cross of the differential coefficient of the instantaneous value of coherence in the non-target speech section. The hangover length may be set based on another feature amount representing the frequency of change in coherence.

  In the first embodiment, the table configuration storage unit is used to obtain the hangover length corresponding to the zero-cross value. However, the method for obtaining the corresponding hangover length from the zero-cross value is a conversion. The method is not limited to a method using a table, and may be a method using a conversion function, for example.

  In each of the above embodiments, the processing that was processed with the frequency domain signal may be performed with the time domain signal if possible, and conversely, the processing that was processed with the time domain signal is possible. In this case, processing may be performed using a frequency domain signal.

  In each of the above embodiments, a case has been described in which a signal captured by a pair of microphones is immediately processed. However, the audio signal to be processed of the present invention is not limited to this. For example, the present invention can be applied to processing a pair of audio signals read from a recording medium, and the present invention can also be applied to processing a pair of audio signals transmitted from the opposite device. Can be applied.

  DESCRIPTION OF SYMBOLS 1 ... Audio | voice signal processing apparatus, m_1, m_2 ... Microphone, 10 ... FFT part, 11 ... 1st directivity formation part, 12 ... 2nd directivity formation part, 13 ... Coherence calculation part, 15 ... Target audio | voice area detection * Hangover giving unit, 16 ... Hangover length control unit, 17 ... Hangover counter initialization threshold value control unit, 21 ... Coherence / speech interval detection result receiving unit, 22 ... Feature quantity calculation control unit, 23 ... Hangover length collation control , 24... Non-target speech section feature amount calculation unit, 25... Hangover length collation unit, 26... Storage unit, 27... Hangover length transmission unit, 31 ... coherence / determination result reception unit, 32. 33 ... dispersion calculation unit, 34 ... hangover counter initialization threshold value comparison unit, 35 ... initialization threshold value storage unit, 36 ... hangover counter initialization threshold value transmission unit.

Claims (4)

In an audio signal processing apparatus that separates a target voice section and a non-target voice section from an input voice signal,
A first directivity forming unit that forms a first directivity signal having a directivity characteristic having a blind spot in a first predetermined direction by performing a delay subtraction process on the input audio signal;
By applying a delay subtraction process to the input audio signal, the second directivity forming a second directivity signal having a directivity characteristic having a blind spot in a second predetermined direction different from the first predetermined direction. Forming part;
A coherence calculator that obtains coherence using the first and second directional signals;
The coherence is compared with a target speech segment determination threshold to determine whether the input speech signal is a target speech segment arriving from a target direction or any other non-target speech segment, and the coherence and the above Even if the determination result by comparison using the target voice segment determination threshold is changed from the target voice segment to the non-target voice segment, the hangover length is larger than the target voice segment determination threshold. A target voice section detection / hangover giving unit for continuing the determination result of the target voice section;
A feature amount indicating the switching frequency of the coherence change tendency in the non-target speech section is obtained, and the hangover length applied by the target speech section detection / hangover providing unit is controlled according to the obtained feature amount. An audio signal processing device comprising: a hangover length control unit.   The hangover length control unit processes the input audio signal in units of frames for each predetermined period. In the non-target audio section, the difference between the coherence values of the current frame and the immediately preceding frame is taken to obtain the coherence differential coefficient. Calculate and determine whether the product of this differential coefficient and the differential coefficient in the previous frame is positive or negative, if negative, increase the counter variable that measures the number of positive and negative fluctuations of the differential coefficient, and if the product is positive, The audio signal processing apparatus according to claim 1, wherein a counter variable obtained by a method of maintaining the counter variable is applied as the feature amount. In an audio signal processing method for separating a target voice section and a non-target voice section from an input voice signal,
The first directivity forming unit forms a first directivity signal having a directivity characteristic having a blind spot in a first predetermined direction by performing a delay subtraction process on the input audio signal,
The second directivity forming unit performs delay subtraction processing on the input audio signal, thereby providing a second directivity having a directivity characteristic having a blind spot in a second predetermined direction different from the first predetermined direction. Form a signal,
The coherence calculation unit obtains coherence using the first and second directional signals,
The target speech segment detection / hangover assigning unit compares the coherence with the target speech segment determination threshold, and the input speech signal is a target speech segment arriving from the target direction or other non-target speech segment And comparing the coherence with a hangover provision threshold value greater than the target speech segment determination threshold value, and the determination result by the comparison using the target speech segment determination threshold value is from the target speech segment to the non-target speech segment. Even if it changes to, the judgment result of the target voice section is continued by the hangover length,
The hangover length control unit obtains a feature value indicating the switching frequency of the coherence change tendency in the non-target speech section, and the target speech section detection / hangover provision unit applies according to the obtained feature value. An audio signal processing method characterized by controlling the hangover length. Computer
A first directivity forming unit that forms a first directivity signal having a directivity characteristic having a blind spot in a first predetermined direction by performing a delay subtraction process on the input audio signal;
By applying a delay subtraction process to the input audio signal, the second directivity forming a second directivity signal having a directivity characteristic having a blind spot in a second predetermined direction different from the first predetermined direction. Forming part;
A coherence calculator that obtains coherence using the first and second directional signals;
The coherence is compared with a target speech segment determination threshold to determine whether the input speech signal is a target speech segment arriving from a target direction or any other non-target speech segment, and the coherence and the above Even if the determination result by comparison using the target voice segment determination threshold is changed from the target voice segment to the non-target voice segment, the hangover length is larger than the target voice segment determination threshold. A target voice section detection / hangover giving unit for continuing the determination result of the target voice section;
A feature amount indicating the switching frequency of the coherence change tendency in the non-target speech section is obtained, and the hangover length applied by the target speech section detection / hangover providing unit is controlled according to the obtained feature amount. An audio signal processing program that functions as a hangover length control unit.

Images