JP5454330B2 - Sound processor - Google Patents

Description

  The present invention relates to a technique for suppressing a noise component included in an acoustic signal.

  Conventionally, a technique for suppressing a noise component from an acoustic signal of a mixed sound of a target sound component and a noise component has been proposed. For example, in Patent Document 1, a component having the minimum intensity is selected from a low-frequency component and an average component of each low-frequency component in each of a plurality of acoustic signals, and synthesized with a high-frequency component of each acoustic signal. A technique for generating a noise suppression signal in which wind noise is suppressed is disclosed.

Japanese Patent No. 4356670

  However, in the technique of Patent Document 1, since the component used for generating the noise suppression signal is selected based on the intensity only, for example, when the intensity of the target sound component is small compared to the wind noise, the target sound component May be removed. In addition, when the average component of multiple acoustic signals is selected as the low-frequency component of the noise suppression signal, the waveform of the target sound component changes significantly during the generation of the noise suppression signal. There is also a problem that it is not reproduced. In view of the above circumstances, an object of the present invention is to suppress a noise component of an acoustic signal with high accuracy.

  In order to solve the above-described problems, the acoustic processing device according to the first aspect of the present invention provides, for each of the first acoustic signal and the second acoustic signal collected in parallel, component values for each frequency of the acoustic signal. Is a non-negative matrix factorization of an observation matrix (for example, the observation matrix Vi in FIG. 2), and a plurality of bases (for example, the base Ci in FIG. 4) indicating component values for different frequencies of the acoustic signal. [1] to Ci [K]) including a base matrix (for example, the base matrix Wi in FIG. 4) and a plurality of weight sequences (for example, the weight sequence Ei [1 in FIG. 5) each indicating a time series of weight values of each base. ] To Ei [K]), matrix decomposition means for generating a coefficient matrix (for example, coefficient matrix Hi in FIG. 5), and a base matrix of the second acoustic signal among a plurality of bases of the base matrix of the first acoustic signal. A base having a high correlation with the base is a noise base corresponding to the noise component of the first acoustic signal (for example, in FIG. 4). Noise specifying means for specifying as noise base Ci_noise), each base other than the noise base in the base matrix of the first acoustic signal and a weight sequence corresponding to each base in the coefficient matrix of the first acoustic signal, Coefficient sequence generating means for generating a processing coefficient sequence (for example, processing coefficient sequence Gi in FIG. 1) in which coefficient values for each frequency are set so that the noise component of the first acoustic signal is suppressed, and processing on the first acoustic signal Noise suppression means for operating the coefficient sequence.

  In the above configuration, each observation matrix of the first acoustic signal and the second acoustic signal is decomposed into a base matrix and a coefficient matrix, and the base matrix of the second acoustic signal among the plurality of bases of the base matrix of the first acoustic signal. A processing coefficient sequence is generated after removing noise bases having a high correlation with the bases. Therefore, even when the intensity of the target sound component of the first acoustic signal is lower than the noise component, the noise component can be suppressed with high accuracy. In addition, since the basis other than the noise basis is maintained in the basis matrix, there is an advantage that the noise component can be suppressed while maintaining the waveform of the target sound component faithfully.

  In a preferred example of the first aspect of the present invention, the matrix decomposing means generates a base matrix and a coefficient matrix for the component of the first band (for example, band BL in FIG. 2) of the first acoustic signal, and the noise suppressing means is The processing coefficient sequence is applied to the component of the first band in the first acoustic signal, the component after processing by the noise suppression means, and the second band on the high frequency side compared to the first band of the first acoustic signal. Waveform synthesizing means for synthesizing the components in the band (for example, band BH in FIG. 2) is provided. In the above aspect, the processing coefficient sequence is generated and applied to the first band component of the first acoustic signal, and is synthesized with the component in the second band after being processed by the noise suppressing means. Therefore, it is possible to reduce the processing load as compared with the configuration in which the generation and application of the processing coefficient sequence is performed for the entire band of the first acoustic signal. Further, by setting the first band so as to include the frequency of the noise component, there is an advantage that a processing coefficient sequence capable of suppressing the noise component with high accuracy can be generated.

  The acoustic processing apparatus according to a preferred example of the first aspect of the present invention includes phase difference calculating means for calculating a phase difference (for example, phase difference ΔP [tn] in FIG. 6) between the first acoustic signal and the second acoustic signal. Then, the coefficient sequence generation means variably sets each coefficient value of the processing coefficient sequence according to the phase difference between the first acoustic signal and the second acoustic signal. For example, each coefficient value of the processing coefficient sequence is variable so that the effect of noise suppression by the processing coefficient sequence increases as the phase difference between the first acoustic signal and the second acoustic signal increases (the noise component is dominant). Is set. According to the above embodiment, since the phase difference between the first acoustic signal and the second acoustic signal is reflected in the processing coefficient sequence, the noise component is sufficiently suppressed as compared with the configuration in which the phase difference is not reflected in the processing coefficient sequence. There is an advantage that you can.

  An acoustic processing apparatus according to a preferred example of the first aspect of the present invention includes intensity difference calculation means for calculating an intensity difference (amplitude difference or power difference) between a first acoustic signal and a second acoustic signal, and generates a coefficient sequence. The means variably sets each coefficient value of the processing coefficient sequence according to the intensity difference between the first acoustic signal and the second acoustic signal (for example, the intensity difference ΔA [tn] in FIG. 6). For example, each coefficient value of the processing coefficient sequence is variable so that the effect of noise suppression by the processing coefficient sequence increases as the intensity difference between the first acoustic signal and the second acoustic signal increases (the noise component is dominant). Is set. According to the above embodiment, since the intensity difference between the first acoustic signal and the second acoustic signal is reflected in the processing coefficient sequence, the noise component is sufficiently suppressed as compared with the configuration in which the intensity difference is not reflected in the processing coefficient sequence. There is an advantage that you can.

  The acoustic processing device according to the second aspect of the present invention provides an observation matrix having, as elements, a time series of component values for each frequency of the acoustic signal for each of the first acoustic signal and the second acoustic signal collected in parallel. A plurality of bases (for example, bases Ci [1] to Ci [K] in FIG. 4) indicating component values for different frequencies of different components of the acoustic signal in non-negative matrix factorization of the observation matrix Vi in FIG. And a plurality of weight sequences (for example, weight sequences Ei [1] to Ei [K] in FIG. 5) each indicating a time series of weight values of the respective bases. Matrix decomposition means (for example, the matrix decomposition unit 42 in FIG. 7) for generating a coefficient matrix (for example, the coefficient matrix Hi in FIG. 5), and the base matrix of the second acoustic signal among the plurality of bases of the base matrix of the first acoustic signal A basis having a high correlation with the basis of the noise basis corresponding to the noise component of the first acoustic signal (for example, FIG. 4). Noise specifying means (for example, the noise specifying unit 44 in FIG. 7) specifying each of the bases other than the noise base and the coefficient matrix of the first acoustic signal among the base matrix of the first acoustic signal. Suppression processing means (for example, a suppression processing unit 48 in FIG. 7) that generates an acoustic signal in which the noise component of the first acoustic signal is suppressed by multiplication with a weight sequence corresponding to a base. With the above configuration, the same effect as the first aspect is realized. Further, since the generation and application of the processing coefficient sequence are omitted, there is an advantage that the configuration and processing of the sound processing device are simplified as compared with the first mode.

  Note that the scope of application of the present invention (the first aspect and the second aspect) is not limited to a configuration for processing two systems of acoustic signals. That is, even in a configuration for processing three or more systems of acoustic signals, a configuration that satisfies the requirements of the present invention when focusing on two specific systems of acoustic signals is naturally included in the scope of the present invention.

  The acoustic processing device according to each of the above aspects is realized by hardware (electronic circuit) such as a DSP (Digital Signal Processor) dedicated to processing of an acoustic signal, or a general-purpose calculation such as a CPU (Central Processing Unit). This is also realized by cooperation between the processing device and the program. The program of the present invention exemplified below is provided to the user in a form stored in a computer-readable recording medium and installed in the computer, and is also provided from the server device in the form of distribution via a communication network. Installed on the computer.

  The program according to the first aspect of the present invention provides a non-negative observation matrix for each of the first acoustic signal and the second acoustic signal collected in parallel, the element being a time series of component values for each frequency of the acoustic signal. In matrix factorization, a base matrix including a plurality of bases indicating component values for each frequency of different components of the acoustic signal, and a coefficient matrix including a plurality of weight sequences each indicating a time series of weight values of each base, A base having a high correlation with a base of the base matrix of the second acoustic signal among a plurality of bases of the base matrix of the first acoustic signal and a noise base corresponding to the noise component of the first acoustic signal First acoustic signal in accordance with the noise identification processing specified as follows, each basis other than the noise basis in the basis matrix of the first acoustic signal, and the weight sequence corresponding to each basis in the coefficient matrix of the first acoustic signal So that the noise component of A coefficient sequence generation process for generating a processing coefficient sequence coefficient value for each number is set to execute a noise suppression process for applying a processing coefficient sequence to the first acoustic signal to the computer. According to the above program, the same operation and effect as the sound processing apparatus according to the first aspect of the present invention are realized.

  The program according to the second aspect of the present invention provides an observation matrix (for example, an element having a time series of component values for each frequency of the acoustic signal as an element for each of the first acoustic signal and the second acoustic signal collected in parallel. The non-negative matrix factorization of the observation matrix Vi) in FIG. 2 includes a plurality of bases (for example, the bases Ci [1] to Ci [K] in FIG. 4) indicating component values for different frequencies of the acoustic signal. A coefficient matrix including a base matrix (for example, the base matrix Wi in FIG. 4) and a plurality of weight sequences (for example, the weight sequences Ei [1] to Ei [K] in FIG. 5) each indicating a time series of weight values of each base. (For example, the coefficient matrix Hi in FIG. 5) and a base having a high correlation with the base of the base matrix of the second acoustic signal among the plurality of bases of the base matrix of the first acoustic signal. The noise base corresponding to the noise component of the acoustic signal (for example, the noise base Ci_noise in FIG. 4) is specified. Noise of the first acoustic signal is obtained by performing noise identification processing and multiplication of each basis other than the noise basis in the basis matrix of the first acoustic signal and a weight sequence corresponding to each basis in the coefficient matrix of the first acoustic signal. And causing the computer to execute a suppression process for generating an acoustic signal in which the component is suppressed. According to the above program, the same operation and effect as the sound processing apparatus according to the second aspect of the present invention are realized.

1 is a block diagram of a sound processing apparatus according to a first embodiment of the present invention. It is explanatory drawing of an observation matrix. It is a block diagram of a characteristic analysis part. It is explanatory drawing of a base matrix. It is explanatory drawing of a coefficient matrix. It is a block diagram of the characteristic analysis part in 2nd Embodiment. It is a block diagram of the noise suppression part in 3rd Embodiment.

<A: First Embodiment>
FIG. 1 is a block diagram of a sound processing apparatus 100 according to the first embodiment of the present invention. As shown in FIG. 1, a signal supply device 12 and a sound emitting device 14 are connected to the sound processing device 100.

  The signal supply device 12 supplies the sound processing device 100 with the stereo-type sound signal s1 and sound signal s2 collected in parallel (simultaneously) at different positions. Each acoustic signal si (i = 1, 2) is a time-domain signal representing a sound pressure waveform of a mixed sound of a target sound component and a noise component. In FIG. 1, a plurality of sound collecting devices (for example, omnidirectional stereo microphones) 122 that are spaced apart from each other are illustrated as the signal supply device 12. However, a playback device that reads out each acoustic signal si from a portable or built-in recording medium and supplies it to the acoustic processing device 100, or a communication device that receives each acoustic signal si from a communication network and supplies it to the acoustic processing device 100. The signal supply device 12 can also be employed.

  The sound processing apparatus 100 generates a stereo sound signal q1 and a sound signal q2 from the sound signal s1 and the sound signal s2. Each acoustic signal qi is a time domain signal obtained by suppressing a noise component (emphasizing a target sound component) from the acoustic signal si. The sound emitting device 14 (for example, a stereo speaker or stereo headphones) radiates sound waves corresponding to the acoustic signal q1 and the acoustic signal q2 generated by the acoustic processing device 100. Note that an A / D converter that converts the acoustic signal si from analog to digital and a D / A converter that converts the acoustic signal qi from digital to analog are not shown for convenience.

  As shown in FIG. 1, the sound processing device 100 is realized by a computer system including an arithmetic processing device 22 and a storage device 24. The storage device 24 stores a program PG executed by the arithmetic processing device 22 and various data used by the arithmetic processing device 22. A known recording medium such as a semiconductor recording medium or a magnetic recording medium or a combination of a plurality of types of recording media can be arbitrarily employed as the storage device 24. A configuration in which the acoustic signal s1 and the acoustic signal s2 are stored in the storage device 24 (therefore, the signal supply device 12 can be omitted) is also suitable.

  The arithmetic processing unit 22 has a plurality of functions (frequency analysis unit 32, characteristic analysis unit 34A, noise suppression unit 36, etc.) for generating the acoustic signal qi from the acoustic signal si by executing the program PG stored in the storage device 24. A waveform synthesis unit 38) is realized. A configuration in which each function of the arithmetic processing unit 22 is distributed over a plurality of integrated circuits, or a configuration in which a dedicated electronic circuit (DSP) realizes each function may be employed.

  1 sequentially generates a spectrum S1 of the acoustic signal s1 and a spectrum S2 of the acoustic signal s2 for each unit period (frame) on the time axis. As shown in FIG. 2, each spectrum Si is a power spectrum in which a plurality of component values (power) xi corresponding to different frequencies (f1, f2,..., FM,...) On the frequency axis are arranged. . For the generation of the spectrum Xi, known frequency analysis such as short-time Fourier transform can be arbitrarily employed.

  The spectrum Si of each acoustic signal si is divided into a spectrum Xi in the band BL and a spectrum XHi in the band BH. The band BL is set so as to include the frequency of the noise component. In the present embodiment, wind noise is assumed as a noise component. Wind noise is a noise component generated when air itself flows and directly collides with the diaphragm of the sound collecting device 122. The vibration frequency of the diaphragm due to the air collision is lower than the frequency of the sound wave propagating to the diaphragm as air vibration (sound pressure change). Specifically, the frequency of the wind noise is dominated by a low frequency component of 1.5 kHz or less, for example. Considering the above tendency, the band BL is set to a frequency band of 1.5 kHz or less including M (M is a natural number) frequencies f1 to fM. The band BH is a high frequency band (for example, 1.5 kHz or more) compared to the band BL.

  As shown in FIG. 2, the time series (that is, the spectrogram) of the spectrum Si generated by the frequency analysis unit 32 is divided on the time axis for each analysis period TA composed of N unit periods t1 to tN. The analysis period TA is set to a long time of about several tens of seconds, for example. M rows × N columns in which M component values xi [f1, tn] to xi [fM, tn] of the spectrum Xi in the band BL are arranged in time series for N unit periods t1 to tN in the analysis period TA. The observation matrix Vi is defined for each analysis period TA. That is, the n-th column (n = 1 to N) of the observation matrix Vi has M component values xi [f1, tn] to xi [fM, of the spectrum Xi in the n-th unit period tn in the analysis period TA. tn], the m-th row (m = 1 to M) of the observation matrix Vi is the component value xi [fm, t1] at the frequency fm over N unit periods t1 to tN in the analysis period TA. This corresponds to a time series of ~ xi [fm, tN]. Since the component value xi [fm, tn] of the spectrum Xi corresponds to power (non-negative value), the observation matrix Vi is a non-negative matrix (matrix not including negative numbers).

  The characteristic analysis unit 34A in FIG. 1 generates a processing coefficient sequence Gi (G1, G2) for each processing period TA by analyzing the observation matrix Vi. The processing coefficient sequence Gi is a matrix of M rows × N columns in which coefficient values gi [f1, t1] to gi [fM, tN] are arranged. The coefficient value gi [fm, tn] corresponds to a gain (spectrum gain) with respect to the component value xi [fm, tn] at the frequency fm in the spectrum Xi of the unit period tn. It is variably set according to the characteristics of the signal si (wind noise intensity). Specifically, the coefficient value gi [fm, tn] is set to a smaller numerical value as the wind noise becomes dominant in the component of the frequency fm in the acoustic signal si within the unit period tn.

  The noise suppression unit 36 applies the processing coefficient sequence Gi generated by the characteristic analysis unit 34A to the observation matrix Vi of the acoustic signal si, so that the N spectra Yi corresponding to each of the N unit periods t1 to tN. A time series (a spectrogram within the analysis period TA) is sequentially generated for each analysis period TA. The spectrum Yi corresponding to the unit period tn is a power spectrum in which M component values (powers) yi [f1, tn] to yi [fM, tn] corresponding to different frequencies fm are arranged.

  The component value yi [fm, tn] at the frequency fm in the spectrum Yi of the unit period tn is observed with the coefficient value gi [fm, tn] corresponding to the component value yi [fm, tn] in the processing coefficient sequence Gi. The matrix Vi is set to a multiplication value with the component value xi [fm, tn] corresponding to the component value yi [fm, tn] (yi [fm, tn] = gi [fm, tn] × xi [fm , tn]). As described above, the coefficient value gi [fm, tn] is set to a smaller value as the wind noise becomes more dominant. Therefore, the spectrum Yi after processing by the noise suppression unit 36 is obtained from the spectrum Xi of the acoustic signal si. This corresponds to the suppressed spectrum (the spectrum of the band BL component of the acoustic signal qi). That is, the noise suppression unit 36 functions as an element that suppresses wind noise from each of the acoustic signal s1 and the acoustic signal s2.

  The waveform synthesizer 38 in FIG. 1 generates a time-domain acoustic signal qi from the spectrum Yi (band BL) generated by the noise suppressor 36 every unit period tn and the spectrum XHi (band BH) generated by the frequency analyzer 32. Generate. Specifically, the waveform synthesizer 38 generates a spectrum Qi for each unit period tn by combining (adding) the spectrum Yi of the band BL and the spectrum XHi of the band BH, and the amplitude spectrum and the acoustic signal corresponding to the spectrum Qi. A time-domain signal is generated by inverse Fourier transform using the phase spectrum of si, and is connected to each other in the preceding and following unit periods tn to generate an acoustic signal qi.

  FIG. 3 is a block diagram of the characteristic analysis unit 34A. As shown in FIG. 3, the characteristic analysis unit 34A of the first embodiment includes a matrix decomposition unit 42, a noise specifying unit 44, and a coefficient sequence generation unit 46. For each of the observation matrix V1 of the acoustic signal s1 and the observation matrix V2 of the acoustic signal s2, the matrix decomposition unit 42 uses a non-negative matrix factorization (NMF) of the observation matrix Vi to form a basis matrix Wi (W1,. W2) and a coefficient matrix Hi (H1, H2) are generated. The base matrix Wi (FIG. 4) is a non-negative matrix of M rows × K columns in which component values wi [f1,1] to wi [fM, K] are arranged, and the coefficient matrix Hi (FIG. 5) is a weight value hi. A non-negative matrix of K rows × N columns in which [1, t1] to hi [K, tN] are arranged (K is a natural number). The base matrix Wi and the coefficient matrix Hi are calculated so that the product of the base matrix Wi and the coefficient matrix Hi approximates the observation matrix Vi (Vi≈WiHi).

  As shown in FIG. 4, the base matrix Wi is composed of K codebooks Ci [1] to Ci [K]. The basis Ci [k] of the k-th column (k = 1 to K) of the basis matrix Wi is the frequency of the k-th component among the K types of components estimated to constitute the acoustic signal si within the analysis period TA. This corresponds to a sequence (power spectrum) of component values wi [f1, k] to wi [fM, k] at f1 to fM. On the other hand, the coefficient matrix Hi is composed of K weight sequences (excitation) Ei [1] to Ei [K] as shown in FIG. The weight series Ei [k] in the k-th row of the coefficient matrix Hi is a weight value hi [k, t1] to hi [k, for each unit period tn for the component expressed by the basis Ci [k] of the basis matrix Wi. This corresponds to a time series of tN] (time change of each component value wi [fm, k] of the basis Ci [k]). As understood from the above definition, the spectrum Xi in the unit period tn is applied with K weight values hi [1, tn] to hi [K, tn] corresponding to the unit period tn in the coefficient matrix Hi. It is approximated by a weighted sum of K bases Ci [1] to Ci [K] (Xi≈hi [1, tn] × Ci [1] + hi [2, tn] × Ci [2] + …… + hi [ K, tn] × Ci [K]).

  A known method is arbitrarily employed for non-negative matrix factorization of the observation matrix Vi. For example, a method of sequentially updating (iteratively calculating) the base matrix Wi and the coefficient matrix Hi so as to minimize the difference (for example, distance) between the product of the base matrix Wi and the coefficient matrix Hi and the observation matrix Vi is preferable. Adopted. The initial value of the base matrix Wi applied to the iterative calculation (the initial value of the component value wi [fm, k]) is set to a random number, for example. For example, the initial values of M component values wi [f1, k] to wi [fM, k] of each base Ci [k] are simulated so as to simulate the spectrum of wind noise (frequency characteristics that attenuate as the frequency increases). The set configuration is also suitable.

  For each of the base matrix W1 of the acoustic signal s1 and the base matrix W2 of the acoustic signal s2, the noise specifying unit 44 in FIG. 3 wind noise among the K bases Ci [1] to Ci [K] of the base matrix Wi. One base (hereinafter referred to as “noise base”) Ci_noise corresponding to is specified. Since wind noise is generated due to the turbulent flow of air colliding with the sound collecting device 122, the instantaneous frequency of the wind noise included in each of the acoustic signals s1 and s2 collected at different positions. The characteristics are statistically independent of each other. However, the long-term frequency characteristics of wind noise are more likely to be maintained at the same characteristics regardless of the position of sound collection, as compared to voice and the like. That is, the frequency characteristics of wind noise over a long period such as the analysis period TA tend to be similar between the acoustic signal s1 and the acoustic signal s2.

  Considering the above tendency, the noise specifying unit 44 uses the basis matrix W1 (basis C1 [1] to C1 [K]) of the acoustic signal s1 and the basis matrix W2 (basis C2 [1] to C2 [) of the acoustic signal s2. K]), each base Ci [k] (C1 [k1], C2 [k2]) having a high correlation with each other is specified as a noise base Ci_noise from each of the base matrix W1 and the base matrix W2. For example, the correlation between the basis C1 [k] and the basis C2 [k] for all combinations of selecting one basis C1 [k] of the basis matrix W1 and one basis C2 [k] of the basis matrix W2 An index (correlation index) indicating the degree of the correlation is calculated, and each of the combinations of the base C1 [k1] and the base C2 [k2] (the values of the variable k1 and the variable k2) having the maximum degree of correlation indicated by the correlation index The noise base Ci_noise (C1_noise, C2_noise) is extracted. As a correlation index between the base C1 [k] and the base C2 [k], for example, a distance (Euclidean distance) or an inner product is preferably employed.

  For each of the acoustic signal s1 and the acoustic signal s2, the coefficient sequence generator 46 of FIG. 3 removes the noise base Ci_noise from the basis matrix Wi of the acoustic signal si as shown in FIG. A processing coefficient sequence Gi (G1, G2) is obtained from the matrix HAi of columns and the matrix HAi of (K-1) rows × N columns excluding the weight sequence Ei_noise corresponding to the noise basis Ci_noise from the coefficient matrix Hi as shown in FIG. ) Is generated.

  First, the coefficient sequence generation unit 46 excludes the noise basis Ci_noise from the matrix WAi (M rows × (K−1) columns) and the weight sequence Ei_noise from the matrix HAi ((K−1) rows × N). Matrix VAi is calculated by multiplication with (column). The matrix VAi is an M row × N column matrix in which element values va [f1, t1] to va [fM, tN] are arranged. A sequence of M element values va [f1, tn] to va [fM, tn] located in the nth column of the matrix VAi corresponds to a power spectrum in which wind noise is suppressed from the spectrum Xi of the unit period tn.

Second, the coefficient sequence generator 46 calculates the coefficient value gi [fm, tn] of the processing coefficient sequence Gi by the calculation of the following formula (A). The symbol v [fm, tn] in the formula (A) is the m-th row in the matrix of M rows × N columns obtained by multiplying the base matrix Wi before the noise base Ci_noise is excluded and the coefficient matrix Hi before the weight sequence Ei_noise is excluded. This means the element value in the n-th column and corresponds to the estimated value of the component value xi [fm, tn] of the spectrum Xi.
gi [fm, tn] = va [fm, tn] / v [fm, tn] (A)
The reason why the element value va [fm, tn] is divided by the element value v [fm, tn] in the formula (A) is that the coefficient value gi [fm, tn] is set to a numerical value in the range of 0 to 1 This is to limit. The processing coefficient sequence G1 and the processing coefficient sequence G2 generated by the above procedure are applied to the suppression of wind noise by the noise suppression unit 36 of FIG.

  As understood from the above description, the element value va [fm, tn] becomes the element value v [fm, tn] as the wind noise prevails in the frequency fm component of the acoustic signal si within the unit period tn. On the other hand, the coefficient value gi [fm, tn] is a small numerical value because it decreases. On the other hand, when the component of the frequency fm of the acoustic signal si does not include wind noise, the element value va [fm, tn] is substantially the same value as the element value v [fm, tn] (not affected by the exclusion of the noise base Ci_noise). Therefore, the coefficient value gi [fm, tn] is a numerical value close to 1. Therefore, as described above, the noise suppression unit 36 causes the processing coefficient sequence Gi to act on the observation matrix Vi, thereby generating the acoustic signal qi in which the wind noise is suppressed from the acoustic signal si.

  As described above, in the first embodiment, the observation matrix Vi of the acoustic signal si is decomposed into the base matrix Wi and the coefficient matrix Hi, and the matrix WAi and the weight sequence Ei_noise in which the noise base Ci_noise is excluded from the base matrix Wi are obtained. A processing coefficient sequence Gi is generated using the matrix HAi excluded from the coefficient matrix Hi. Therefore, even when the intensity of the target sound component of the acoustic signal si is lower than that of the wind noise, it is possible to suppress the wind noise with high accuracy. Further, since each basis Ci [k] other than the noise basis Ci_noise in the basis matrix Wi and each weight series Ei [k] other than the weight series Ei_noise in the coefficient matrix Hi are maintained, the target sound component of the acoustic signal si is maintained. There is also an advantage that an acoustic signal qi can be generated in which the waveform of is maintained faithfully.

  As a method for identifying the noise base Ci_noise from the base matrix Wi, for example, a configuration in which a model created in advance so as to simulate the frequency characteristics of wind noise is compared with each base Ci [k] of the base matrix Wi. Can be employed. However, in a configuration using a wind noise model, wind noise having a frequency characteristic different from that of a model prepared in advance may not be sufficiently suppressed. On the other hand, in the first embodiment, each base Ci [k] having a high correlation between the base matrix W1 and the base matrix W2 is specified as a noise base Ci_noise, so that it is compared with a configuration using a wind noise model. There is an advantage that wind noise with various characteristics can be sufficiently suppressed.

  Further, since the generation and application of the processing coefficient sequence Gi is selectively executed for the band BL including wind noise (the band BH is omitted), it is compared with the case where the entire band of the acoustic signal si is to be processed. Thus, there is an advantage that the processing load of the arithmetic processing unit 22 can be reduced while maintaining the accuracy of suppression of wind noise.

<B: Second Embodiment>
Next, a second embodiment of the present invention will be described. In addition, about the element which an effect | action and function are equivalent to 1st Embodiment in each following illustration, the code | symbol referred by the above description is diverted and each detailed description is abbreviate | omitted suitably.

  The target sound component that arrives from the front direction with respect to the two sound collecting devices 122 of the signal supply device 12 reaches each sound collecting device 122 with substantially the same intensity (amplitude) without generating a phase difference. On the other hand, since wind noise is caused by air turbulence as described above, the possibility of reaching each sound collecting device 122 with the same phase and the same amplitude is low. Therefore, the more the wind noise becomes dominant in the acoustic signal s1 and the acoustic signal s2, the more the phase difference and intensity difference between the two tend to increase. In consideration of the above tendency, in the present embodiment, each coefficient value gi [fm, tn] of the processing coefficient sequence Gi is corrected according to the phase difference or intensity difference between the acoustic signal s1 and the acoustic signal s2.

  The sound processing apparatus 100 of the second embodiment has a configuration in which the characteristic analysis unit 34A of the first embodiment is replaced with a characteristic analysis unit 34B of FIG. As shown in FIG. 6, the characteristic analysis unit 34B includes a phase difference calculation unit 52 and an intensity difference calculation in addition to the same elements (matrix decomposition unit 42, noise identification unit 44, coefficient sequence generation unit 46) as the characteristic analysis unit 34A. Part 54. The components of the band BM of each acoustic signal si (s1, s2) are supplied to the phase difference calculation unit 52 and the intensity difference calculation unit 54. The band BM is set so as to include the frequency of the wind noise and the frequency of the main target sound component. For example, the band BM is set to a frequency band of 4 kHz or less (that is, a band including the band BL).

  The phase difference calculator 52 calculates the phase difference ΔP [tn] between the acoustic signal s1 and the acoustic signal s2 for each unit period tn. The phase difference ΔP [tn] is, for example, a representative value (for example, an average value) of the phase difference at each frequency in the band BM. Similarly, the intensity difference calculation unit 54 calculates an intensity difference (for example, an amplitude difference or a power difference) ΔA [tn] between the acoustic signal s1 and the acoustic signal s2 for each unit period tn.

  The coefficient sequence generator 46 calculates the coefficient value γi [fm, tn] calculated by the same method as the coefficient value gi [fm, tn] of the first embodiment (the coefficient value gi [fm, tn] of the first embodiment). Corresponding to the phase difference ΔP [tn] calculated by the phase difference calculation unit 52 and the intensity difference ΔA [tn] calculated by the intensity difference calculation unit 54, thereby correcting the coefficient value gi of the processing coefficient sequence Gi. Calculate [fm, tn].

  First, the coefficient sequence generator 46 sets a coefficient value ξp [tn] for each unit period tn according to the phase difference ΔP [tn] calculated by the phase difference calculator 52. Specifically, the coefficient value ξp [tn] is 0 so that the coefficient value ξp [tn] becomes smaller as the phase difference ΔP [tn] is larger (that is, the wind noise is dominant in the unit period tn). It is set variably within the range above and below 1. Secondly, the coefficient sequence generator 46 sets the coefficient value ξa [tn] for each unit period tn according to the intensity difference ΔA [tn] calculated by the intensity difference calculator 54. Specifically, the coefficient value ξa [tn] is 0 so that the coefficient value ξa [tn] becomes smaller as the intensity difference ΔA [tn] is larger (that is, the wind noise is dominant in the unit period tn). It is set variably within the range above and below 1.

  Third, the coefficient sequence generation unit 46 adjusts the coefficient value γi [fm, tn] calculated by the calculation of the formula (A) according to the coefficient value ξp [tn] and the coefficient value ξa [tn]. Each coefficient value gi [fm, tn] (gi [f1, t1] to gi [fM, tN]) of the coefficient sequence Gi is calculated. For example, the coefficient sequence generation unit 46 calculates a multiplication value of the coefficient value γi [fm, tn], the coefficient value ξp [tn], and the coefficient value ξa [tn] as the coefficient value gi [fm, tn] (gi [ fm, tn] = γi [fm, tn] × ξp [tn] × ξa [tn]). As understood from the above description, the coefficient value gi [gm, tn] is smaller as the coefficient value ξp [tn] and the coefficient value ξa [tn] are smaller (that is, the wind noise is dominant in the unit period tn). It becomes a numerical value. Therefore, according to the second embodiment, there is an advantage that wind noise can be sufficiently suppressed as compared with the first embodiment.

  In the above example, the phase difference ΔP [tn] and the intensity difference ΔA [tn] are calculated for each unit period tn, but the phase difference ΔP [fm, tn] for each frequency fm and the intensity difference ΔA [for each frequency fm. A configuration in which fm, tn] is calculated for each unit period tn is also suitable. The coefficient sequence generator 46 corresponds to the coefficient value ξp [fm, tn] and the intensity difference ΔA [fm, tn] corresponding to the phase difference ΔP [fm, tn] for each combination of the frequency fm and the unit period tn. The coefficient value ξa [fm, tn] is calculated. The coefficient value gi [fm, tn] of the processing coefficient sequence Gi is calculated as a multiplication value of the coefficient value γi [fm, tn], the coefficient value ξp [fm, tn], and the coefficient value ξa [fm, tn], for example. .

<C: Third Embodiment>
Next, a third embodiment of the present invention will be described. In the first embodiment, the time series of the spectrum Yi is generated by applying the processing coefficient sequence Gi (G1, G2) corresponding to the matrix VAi after the exclusion of the noise basis Ci_noise to the observation matrix Vi of the acoustic signal si. As described above, the sequence of M element values va [f1, tn] to va [fM, tn] located in the nth column of the matrix VAi has a power spectrum in which wind noise is suppressed from the spectrum Xi of the unit period tn. Equivalent to. Therefore, in the third embodiment, the matrix VAi after removal of the noise basis Ci_noise is used as a time series of the spectrum Yi. That is, the generation of the processing coefficient sequence Gi and the application to the observation matrix Vi are omitted.

  The sound processing apparatus 100 of the third embodiment has a configuration in which the characteristic analysis unit 34A and the noise suppression unit 36 of the first embodiment are replaced with a noise suppression unit 60 of FIG. As shown in FIG. 7, the noise suppression unit 60 includes a matrix decomposition unit 42, a noise identification unit 44, and a suppression processing unit 48. The matrix decomposing unit 42 and the noise specifying unit 44 are the same as those in the first embodiment. That is, the matrix decomposition unit 42 generates the base matrix Wi (W1, W2) and the coefficient matrix Hi (H1, H2) by non-negative matrix factorization of each observation matrix Vi, and the noise specifying unit 44 stores the base matrix W1. Noise base C1_noise and noise base C2_noise in the base matrix W2 are specified.

  For each of the acoustic signal s1 and acoustic signal s2, the suppression processing unit 48 applies an M-row × (K−1) -column matrix WAi obtained by excluding the noise basis Ci_noise from the basis matrix Wi of the acoustic signal si and the noise basis Ci_noise. The matrix VAi of M rows × N columns calculated by multiplying the corresponding weight sequence Ei_noise from the coefficient matrix Hi by the multiplication with the (K-1) rows × N columns matrix HAi is obtained from the spectrum Yi after the suppression of wind noise. A time series (spectrogram) is sequentially generated for each analysis period TA. That is, a sequence of M element values va [f1, tn] to va [fM, tn] located in the nth column of the matrix VAi is supplied to the waveform synthesis unit 38 as the spectrum Yi of the unit period tn. Other operations are the same as those in the first embodiment.

  In the third embodiment, a time series (matrix VAi) of the spectrum Yi is generated by multiplying the matrix WAi excluding the noise basis Ci_noise from the base matrix Wi and the matrix HAi excluding the weight sequence Ei_noise from the coefficient matrix Hi. Therefore, the same effect as the first embodiment is realized. Further, since calculation of the processing coefficient sequence Gi (formula (A)) and application to the observation matrix Vi are omitted, there is an advantage that the configuration and processing of the sound processing apparatus 100 are simplified as compared with the first embodiment. is there.

<D: Modification>
Each of the above forms can be variously modified. Specific modifications are exemplified below. Two or more aspects arbitrarily selected from the following examples can be appropriately combined.

(1) Modification 1
In each of the above embodiments, the acoustic signal si is divided into the band BL and the band BH, but the band division of the acoustic signal si may be omitted. For example, a configuration in which the entire band of the acoustic signal si is processed by the characteristic analysis unit 34 (34A, 34B) or the noise suppression unit 36 is also employed. However, since the intensity of the wind noise decreases in the higher band (for example, the band BH), in the configuration in which the band division of the acoustic signal si is omitted, the independent basis Ci [k] of the wind noise is obtained by non-negative matrix factorization. It becomes difficult to extract with high accuracy. Therefore, when the frequency band of the noise component to be suppressed is known in advance, only the frequency band (band BL) including the noise component is used for the characteristic analysis unit 34 (34A, 34B) or the noise suppression unit 36. The above-described configuration that is the object of the processing by is particularly suitable.

(2) Modification 2
In each of the above embodiments, the processing coefficient sequence Gi is generated for each analysis period TA of the acoustic signal si, but the division of the analysis period TA is omitted. For example, a configuration in which the processing coefficient sequence Gi is generated using the time series of the spectrum Xi for each unit period tn over the entire interval of the acoustic signal si as one observation matrix Vi can be employed.

(3) Modification 3
In each of the above embodiments, the configuration in which the spectrum Si is the power spectrum is illustrated, but a configuration in which the spectrum Si is the amplitude spectrum can also be employed. Since the coefficient value gi [fm, tn] calculated by the above formula (A) is a gain with respect to power, in the configuration in which the spectrum Si is an amplitude spectrum, the square root (va [fm, tn] of the right side of the formula (A) is used. ] / V [fm, tn]) ^1/2 is calculated as the coefficient value gi [fm, tn].

(4) Modification 4
In each of the above embodiments, the acoustic signal qi is generated by multiplying each component value xi [fm, tn] of the acoustic signal si by each coefficient value gi [fm, tn] of the processing coefficient sequence Gi. The method of causing Gi to act on the acoustic signal si is appropriately changed. For example, a configuration in which the coefficient value gi [fm, tn] is added to each component value xi [fm, tn] of the acoustic signal si may be employed. Contrary to the examples in the above embodiments, in the configuration in which the processing coefficient sequence Gi is generated so that the coefficient value gi [fm, tn] becomes larger as the wind noise becomes more dominant, the component value xi [ A configuration in which fm, tn] is divided or subtracted by the coefficient value gi [fm, tn] may be employed.

(5) Modification 5
In the above embodiments, two systems of acoustic signals qi (q1, q2) are generated, but the above embodiments can be similarly applied when only one system (monaural format) of acoustic signals q1 is generated. For example, the coefficient sequence generation unit 46 generates only one processing coefficient sequence G1 corresponding to the observation matrix V1, and the noise suppression unit 36 generates only the time series of the spectrum Y1 by applying the processing coefficient sequence G1 to the observation matrix V1. To do. In the above configuration, the acoustic signal s2 is used to specify the noise base C1_noise from the basis matrix W1 of the acoustic signal s1.

(6) Modification 6
The processing of the arithmetic processing unit 22 (generation and application of the processing coefficient sequence Gi) is executed in real time in parallel with the supply of the acoustic signal s1 and the acoustic signal s2, and each acoustic signal qi is sequentially generated for each processing. Can be regenerated. However, a configuration (batch process) in which generation of the acoustic signal qi is started after generation and application of the processing coefficient sequence Gi for the acoustic signal s1 and the acoustic signal s2 prepared in advance is completed.

DESCRIPTION OF SYMBOLS 100 ... Sound processing device, 12 ... Signal supply device, 14 ... Sound emission device, 22 ... Arithmetic processing device, 24 ... Memory | storage device, 32 ... Frequency analysis part, 34A, 34B ... Characteristic analysis part, 36 …… Noise suppression unit, 38 …… Waveform synthesis unit, 42 …… Matrix decomposition unit, 44 …… Noise identification unit, 46 …… Coefficient sequence generation unit, 52 …… Phase difference calculation unit, 54 …… Intensity difference calculation Department.

Claims (4)

For each of the first acoustic signal and the second acoustic signal collected in parallel, the acoustic signals are different by non-negative matrix factorization of an observation matrix having a time series of component values for each frequency of the acoustic signal as elements. Matrix decomposition means for generating a base matrix including a plurality of bases indicating component values for each frequency of the component, and a coefficient matrix including a plurality of weight sequences each indicating a time series of weight values of each base;
A base having a high correlation with the base of the base matrix of the second acoustic signal among the plurality of bases of the base matrix of the first acoustic signal is identified as a noise base corresponding to a noise component of the first acoustic signal Noise identification means to
According to each basis other than the noise basis among the basis matrix of the first acoustic signal and a weight sequence corresponding to each basis among the coefficient matrix of the first acoustic signal, the first acoustic signal of the first acoustic signal Coefficient sequence generation means for generating a processing coefficient sequence in which coefficient values for each frequency are set so that noise components are suppressed;
Noise suppression means comprising: noise suppression means for applying the processing coefficient sequence to the first acoustic signal. The matrix decomposition means generates the base matrix and the coefficient matrix for the first band component of the first acoustic signal,
The noise suppression means causes the processing coefficient sequence to act on the component of the first band in the first acoustic signal,
The waveform synthesizing unit that synthesizes the component after processing by the noise suppression unit and the component in the second band on the high frequency side compared to the first band in the first acoustic signal. Sound processing device. Comprising phase difference calculating means for calculating a phase difference between the first acoustic signal and the second acoustic signal;
The acoustic processing apparatus according to claim 2, wherein the coefficient sequence generation unit variably sets each coefficient value of the processing coefficient sequence in accordance with a phase difference between the first acoustic signal and the second acoustic signal. Comprising intensity difference calculating means for calculating an intensity difference between the first acoustic signal and the second acoustic signal;
The acoustic processing device according to claim 2 or 3, wherein the coefficient sequence generation means variably sets each coefficient value of the processing coefficient sequence according to an intensity difference between the first acoustic signal and the second acoustic signal. .

Images