JP2014052585A - Sound processing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To effectively adjust a reverberation component of a sound signal by simple processing.SOLUTION: A variable setting unit 42 sets a reverberation time T. A coefficient identification unit 44 identifies a smoothing coefficient α corresponding to the reverberation time T. An intensity average unit 46 calculates an average spectrum P(k) by an index moving average of a power spectrum P(k, m) of a sound signal x(n) to which the smoothing coefficient α identified by the coefficient identification unit 44 is applied. An adjustment value calculation unit 48 calculates an adjustment value G(k, m) for suppressing a reverberation component of the sound signal x(n) according to a ratio P(k, m)/P(k) of the power spectrum P(k, m) for the average spectrum P(k). The reverberation component is suppressed by making the adjustment value G(k, m) calculated by the adjustment value calculation unit 48 act on the sound signal x(n).

Description

  The present invention relates to a technique for processing an acoustic signal, and more particularly to a technique for adjusting a reverberation component included in an acoustic signal.

  Techniques for suppressing reverberation components included in acoustic signals have been conventionally proposed. For example, in Patent Document 1, a prediction filter coefficient of a reverberation component is estimated by using a probability model of a prediction filter coefficient that estimates a reverberation component included in an acoustic signal, and a reverberation component is calculated using a prediction filter after estimation. Techniques for suppressing are disclosed. Non-Patent Document 1 discloses a technique for suppressing a reverberation component by estimating an inverse filter of a transfer function from a sound source to a sound collection point and applying the estimated inverse filter to an acoustic signal. .

JP 2009-212599 A

K. Furuya, et al. "Robust speech dereverberation using multichannel blind deconvolution with spectral subtraction", IEEE Transantions on Audio, Speech, and Language Processing, vol. 15, no. 5, p.1579-1591, 2007

  However, there is a problem that enormous calculation is required for high-precision estimation of the prediction filter coefficient of Patent Document 1 and the inverse filter of Non-Patent Document 1. In view of the above circumstances, an object of the present invention is to adjust (suppress or enhance) a reverberation component of an acoustic signal by simple processing.

In order to solve the above problems, the acoustic processing apparatus of the present invention includes a coefficient specifying unit that specifies a moving average coefficient according to reverberation time, and an intensity of an acoustic signal to which the moving average coefficient specified by the coefficient specifying unit is applied ( For example, intensity average means for calculating an average intensity (for example, average spectrum P _X (k)) by a moving average of the power spectrum P _X (k, m)), and an adjustment value for adjusting the reverberation component of the acoustic signal are average intensity. And an adjustment value calculation means for calculating the reverberation, and a reverberation adjustment means for causing the adjustment value calculated by the adjustment value calculation means to act on the acoustic signal. In the above configuration, since the adjustment value of the reverberation component is calculated according to the average intensity of the acoustic signal, the reverberation component of the acoustic signal can be adjusted by a simple process compared to the techniques of Patent Document 1 and Non-Patent Document 1. There is an advantage. In addition, since the moving average coefficient applied to the calculation of the average intensity is variably set according to the reverberation time, it is possible to effectively adjust the reverberation component as compared with the configuration in which the moving average coefficient is fixed to a predetermined value. There is also.

  In a preferred aspect of the present invention, the adjustment value calculation means includes a ratio of an intensity index that follows the intensity of the acoustic signal with higher followability than the average intensity to the average intensity (for example, the ratio of the intensity of the acoustic signal to the average intensity). ) To calculate the adjustment value. In particular, according to the configuration in which the adjustment value is calculated according to the ratio of the intensity of the acoustic signal to the average intensity, the effect that the reverberation component of the acoustic signal can be adjusted by a simple process is particularly remarkable. In addition, considering the tendency that the ratio of the intensity index to the average intensity is affected by the reverberation time, the adjustment value according to the ratio of the intensity index to the average intensity and the correction coefficient (for example, the correction coefficient A1) according to the reverberation time. A configuration for calculating is preferable. In the above aspect, since the correction coefficient corresponding to the reverberation time is applied to the calculation of the adjustment value, there is an advantage that the reverberation component of the acoustic signal can be sufficiently suppressed even when the reverberation time is long.

  In a preferred aspect of the present invention, the coefficient specifying means specifies a smoothing coefficient (for example, the smoothing coefficient α) that becomes a smaller value as the reverberation time is longer as the moving average coefficient, and the intensity averaging means is the latest of the acoustic signal. The exponential moving average of the intensity of the sound signal obtained by applying a smoothing coefficient to the intensity is calculated as the average intensity. In other words, the longer the reverberation time, the larger the smoothing coefficient (for example, the smoothing coefficient (1-α)) with respect to the past average intensity in the exponential moving average. In the above aspect, the longer the reverberation time, the smaller the smoothing coefficient for the latest intensity of the acoustic signal is set to a smaller numerical value, so that the reverberation component of the acoustic signal can be effectively adjusted.

  In a preferred aspect of the present invention, the adjustment value calculating means calculates an adjustment value for suppressing the reverberation component of the acoustic signal, and the coefficient specifying means adjusts the target sound reverberation ratio of the acoustic signal before suppression of the reverberation component. The reverberation time so as to satisfy the relationship (for example, the relationship of the approximate line L) approximately selected under the condition that the difference from the target sound reverberation ratio of the acoustic signal after suppression to which the value is applied is maximized. The moving average coefficient corresponding to is specified. In the above aspect, according to the reverberation time so as to satisfy the relationship (for example, the relationship of the approximate line) approximately selected under the condition that the difference in the target sound reverberation ratio before and after the suppression of the reverberation component is maximized. Since the moving average coefficient is specified, the effect that the reverberation component of the acoustic signal can be effectively adjusted is particularly remarkable.

  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 according to the present invention includes a coefficient specifying process that specifies a moving average coefficient according to reverberation time, and an intensity average that calculates an average intensity by a moving average of the intensity of an acoustic signal to which the moving average coefficient specified by the coefficient specifying process is applied. Processing, adjustment value calculation processing for calculating an adjustment value for adjusting the reverberation component of the acoustic signal according to the average intensity, and reverberation adjustment processing for causing the adjustment value calculated in the adjustment value calculation processing to act on the acoustic signal To run. According to the above program, the same operation and effect as the sound processing apparatus according to the present invention are realized. Note that the program of the present invention is provided in a form stored in a computer-readable recording medium and installed in the computer, or is provided in a form distributed via a communication network and installed in the computer.

1 is a block diagram of a sound processing apparatus according to a first embodiment of the present invention. It is a block diagram of an analysis processing part. It is a block diagram of an analysis device. It is a flowchart of operation | movement of an analyzer. It is a graph which shows the relationship between an average number and a suppression effect parameter | index. It is a graph which shows the relationship between reverberation time and the optimal number. It is a graph which shows the relationship between reverberation time and the numerical value of a complementary winner filter.

<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 acoustic signal x (n) to the acoustic processing device 100. The sound signal x (n) is a sample sequence (n) indicating a time waveform of sound obtained by adding sound components (initial reflection component and late reverberation component) reflected or scattered in the sound space to sound emitted from the sound source. : Sample number). For example, an acoustic signal x (n) that has been added to the existing sound such as recorded sound or synthesized sound afterwards, or recorded in an acoustic space (for example, an acoustic hall) that has a reverberant effect. An acoustic signal x (n) is preferably used. A sound collection device that collects ambient sound and generates an acoustic signal x (n), or a playback device that acquires the acoustic signal x (n) from a portable or built-in recording medium and supplies the acoustic signal x (n) to the acoustic processing device 100 Alternatively, a communication device that receives the acoustic signal x (n) from the communication network and supplies the acoustic signal x (n) to the acoustic processing device 100 may be employed as the signal supply device 12.

  The acoustic processing apparatus 100 according to the first embodiment is a reverberation suppression apparatus that generates an acoustic signal y (n) in a time domain in which a reverberation component (late reverberation component) of an acoustic signal x (n) is suppressed. The sound emitting device 14 (for example, a speaker or headphones) reproduces a sound wave corresponding to the acoustic signal y (n) generated by the acoustic processing device 100. Illustration of a D / A converter that converts the acoustic signal y (n) from digital to analog, an amplifier that amplifies the acoustic signal y (n), and the like is omitted 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 PGM1 executed by the arithmetic processing device 22 and various data (for example, correlation coefficient C described later) 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 x (n) is stored in the storage device 24 (therefore, the signal supply device 12 is omitted) is also suitable.

  The arithmetic processing unit 22 executes a program PGM1 stored in the storage unit 24, thereby generating a plurality of functions (frequency analysis unit 32, analysis processing) for generating the acoustic signal y (n) from the acoustic signal x (n). Unit 34, reverberation adjusting unit 36, waveform synthesizing unit 38). A configuration in which each function of the arithmetic processing device 22 is distributed to a plurality of devices or a configuration in which a dedicated electronic circuit (DSP) realizes each function may be employed.

  The frequency analysis unit 32 sequentially generates a spectrum (complex spectrum) X (k, m) of the acoustic signal x (n) for each unit period (frame) on the time axis. Symbol k is a variable that designates an arbitrary frequency (band) on the frequency axis, and symbol m designates an arbitrary unit period on the time axis (a specific point in time on the time axis). Variable. For generation of the spectrum X (k, m), a known frequency analysis such as a short-time Fourier transform can be arbitrarily employed. Note that a filter bank including a plurality of bandpass filters having different passbands can also be employed as the frequency analysis unit 32.

  The analysis processing unit 34 sequentially calculates an adjustment value G (k, m) corresponding to the spectrum X (k, m) of the acoustic signal x (n) for each frequency for each unit period. The adjustment value G (k, m) in the first embodiment is a variable (gain) for suppressing the reverberation component of the acoustic signal x (n). Schematically, the adjustment value G (k, m) is set to a smaller numerical value as the reverberation component is dominant in the component of the kth frequency in the acoustic signal x (n) of the mth unit period. There is a tendency.

  The reverberation adjusting unit 36 causes the adjustment value G (k, m) calculated by the analysis processing unit 34 to act on the acoustic signal x (n). The adjustment by the reverberation adjusting unit 36 is sequentially performed for each frequency for each unit period. Specifically, the reverberation adjusting unit 36 multiplies the spectrum X (k, m) of the acoustic signal x (n) by the adjustment value G (k, m) to thereby obtain the spectrum Y (k) of the acoustic signal y (n). , m) is calculated (Y (k, m) = G (k, m) · X (k, m)). That is, the adjustment value G (k, m) corresponds to a gain (spectrum gain) for the spectrum X (k, m) of the acoustic signal x (n).

  The waveform synthesis unit 38 generates a time domain acoustic signal y (n) from the spectrum Y (k, m) generated by the reverberation adjustment unit 36 for each unit period. That is, the waveform synthesizer 38 converts the spectrum Y (k, m) of each unit period into a signal in the time domain by short-time inverse Fourier transform and connects the unit periods that follow each other to connect the acoustic signal y ( n). The acoustic signal y (n) generated by the waveform synthesizer 38 is supplied to the sound emitting device 14 and reproduced as a sound wave.

以下の数式(2)のように表現することも可能である。

数式(1)および数式(2)の記号τは、サンプルの個数で表現された時間差（ラグ）を意味し、数式(1)の記号Ｅ_n[ ]は、時間的な平均値（時間軸上の複数のサンプルにわたる平均値）を意味する。 <Examination of adjustment value G (k, m)>
The condition of the adjustment value G (k, m) will be discussed below. First, the total number of samples in the unit period after setting one unit period to the time length of the average phoneme is denoted by N _E, the voice is uncorrelated between phonemes (mutual unit period) Assuming that the autocorrelation function R _s ^(m) (τ) within one phoneme (within the m-th unit period) of the sound source signal s (n) indicating the sound emitted from the sound source is It is expressed by equation (1).

It can also be expressed as the following formula (2).

The symbol τ in Equation (1) and Equation (2) means the time difference (lag) expressed by the number of samples, and the symbol E _n [] in Equation (1) is the temporal average value (on the time axis). Mean value over multiple samples).

数式(3)の記号ｂ(μ)は、疑似乱数を意味する。例えば、独立同分布（i.i.d.）の白色雑音（平均零のガウス過程）が疑似乱数ｂ(μ)として好適である。また、記号Ｔ₆₀は残響時間を意味し、記号Ｎ₆₀は、サンプリング周波数ｆsでの残響時間Ｔ₆₀内のサンプルの総数を意味する。数式(3)から理解される通り、室内インパルス応答ｈ(μ)は、独立同分布の確率過程で表現される疑似乱数ｂ(μ)と指数減衰部ｅ^−Δμとでモデル化される。また、室内インパルス応答ｈ(μ)は、以下の数式(4)で表現される通り、空間的な平均値（発音源の複数の位置と収音点の複数の位置とにわたる集合平均）Ｅ_h[ ]に対して無相関となる。

On the other hand, a room impulse response (RIR) h (μ) is expressed by the following equation (3).

The symbol b (μ) in Equation (3) means a pseudo random number. For example, white noise with an independent same distribution (iid) (meaning Gaussian process with zero mean) is suitable as the pseudorandom number b (μ). Symbol T ₆₀ means reverberation time, and symbol N ₆₀ means the total number of samples in the reverberation time T ₆₀ at the sampling frequency fs. As understood from Equation (3), the indoor impulse response h (μ) is modeled by a pseudo random number b (μ) expressed by a stochastic process of independent and identical distribution and an exponential decay part e− ^Δμ . Further, the indoor impulse response h (μ) is a spatial average value (a set average over a plurality of positions of the sound source and a plurality of positions of the sound collection points) E _h as expressed by the following formula (4). Uncorrelated with [].

数式(5)の記号Ｅ_n,hは、時間的および空間的な平均値（期待値）を意味する。なお、数式(5)の導出では、音響信号ｘ(n)が初期反射成分ｘ_E(n)と後期残響成分ｘ_R(n)との加算で表現されるという関係（ｘ(n)＝ｘ_E(n)＋ｘ_R(n)）を利用した。数式(5)の右辺の各項について以下に検討する。 On the other hand, the autocorrelation function R _x (τ) of the acoustic signal x (n) is expressed by the following equation (5).

Symbol E _{n, h in} Equation (5) means a temporal and spatial average value (expected value). In the derivation of Equation (5), the relationship that the acoustic signal x (n) is expressed by the addition of the early reflection component x _E (n) and the late reverberation component x _R (n) (x (n) = x _E (n) + x _R (n)) was used. Each term on the right side of Equation (5) is examined below.

The first term of the formula (5) is transformed as the following formula (6).

Of the room impulse response h (μ) expressed by the above formula (3), the exponential decay part e− ^Δμ is not a stochastic process and is excluded from the calculation of the average value E _h []. In addition, since the autocorrelation function of white noise used as pseudorandom numbers (b (i), b (l)) is a Dirac delta function, the variable i matches the variable l (i = The predetermined numerical value σ _b ² is obtained only in l). Considering the above relationship, the following formula (7) is derived.

Further, in the expression (6), the portion E _n [s (ni) s (n−τ−l)] corresponding to the autocorrelation function of the sound source signal s (n) relates to the indoor impulse response h (μ). Since the portion E _h [h (i) h (l)] acts as a coefficient, it is sufficient to consider only when the variable i matches the variable l (i = 1). Therefore, Formula (6) is transformed into the following Formula (8). In the derivation of Equation (8), the variable (ni) is replaced with the variable n ′.

As described above, assuming that the speech is uncorrelated between phonemes, and taking into account the condition of Equation (2) described above, the autocorrelation function R _s (τ) in Equation (8) is converted to the autocorrelation of Equation (1). Generality is maintained even if the function R _s ^(m) (τ) is substituted. Therefore, the first term of the formula (5) is transformed as the following formula (9).

Considering that the product-sum term in Equation (9) is the sum of the geometric series, and applying the formula for the sum of the geometric series, Equation (9) corresponding to the first term in Equation (5) is This is transformed into Equation (10). In Equation (10), coefficient terms are collectively represented by the symbol (σ _b ′) ² in order to simplify the notation.

Considering that the late reverberation component x _R (n) is an acoustic component from the end point (N _E ) of the initial reflection interval to the end point of reverberation time T ₆₀ (N ₆₀ ), the second term of Equation (5) is It is transformed as Equation (11).

The average value E _h [h (i) h (l)] of Expression (11) is a significant numerical value only when the variable i and the variable l match, but the range of the variable i (N _E ≦ i ≦ N ₆₀ ) and the range of the variable l (0 ≦ l ≦ N _E −1) do not overlap each other, so the variable i and the variable l do not match. Therefore, the average value E _h [h (i) h (l)] in the equation (11) is zero. When the above equation (4) is added to the above relationship, it is understood that the second term E _{n, h} [x _R (n) x _E (n−τ)] in equation (5) is zero. The Similarly, the third term E _{n, h} [x _E (n) × _R (n−τ)] in the equation (5) is also zero.

Since the fourth term of the equation (5) corresponds only to the late reverberation component x _R (n), it is transformed as the following equation (12). In the derivation of the mathematical expression (12), as in the derivation of the mathematical expression (7), the pseudorandom numbers (b (i), b (l) are used only when the variable i matches the variable l (i = 1). The relationship that the autocorrelation function of)) becomes a predetermined numerical value σ _b ² was used.

The time length T _E (N _E samples) of the initial reflection section is sufficiently shorter than the reverberation time T ₆₀ (T _E << T ₆₀ ), and the sound source signal s (n) is a phoneme as described above. Considering that there is no correlation between them, the average value E _n [s (ni) s (n−τ−i)] of Equation (12) changes every time the variable i crosses the boundary between phonemes. Therefore, similar to the derivation of Equation (9), applying the autocorrelation function R _s ^(m) (τ) in the ^mth phoneme of the sound source signal s (n), Equation (12) is As shown in the following equation (13), it is transformed into a double product sum form (M ₆₀ = N ₆₀ / N _E ).

Similar to the derivation of Equation (10), Equation (13) is transformed into Equation (14) below by applying the formula for the sum of geometric series.

As described above, under the assumption that speech is uncorrelated between phonemes, the autocorrelation function R _s (τ) in Equation (8) is changed to the autocorrelation function R _s ^(m) (τ) in Equation (1). The relationship that it can be substituted (R _s (τ) = E _{n, h} [x _E (n) × _E (n−τ)] + E _{n, h} [x _R (n) × _R (n−τ)] ≈R _s ^(m) (τ)) and the relationship described above (particularly, Equation (10) and Equation (14)), the autocorrelation function R _x of the acoustic signal x (n) in the m-th unit period ^{(m) (τ),} the following equation including autocorrelation function R _E of the initial reflection _{segment, x} ^{(m) (τ)} and the autocorrelation function of the late reverberation interval R _R, and _x ^{(m) (τ) (} It is expressed in 15).

数式(16)から理解される通り、音響信号ｘ(n)の自己相関関数Ｒ_x ^(m)(τ)は、室内インパルス応答ｈ(μ)の指数減衰部ｅ^−Δμを加重値とする音源信号ｓ(n)の自己相関関数Ｒ_s ^(m)(τ)の線形和（加重和）として表現される。 Considering that the exponential function term e− ^2Δm′NE becomes 1 when the variable m ′ is 0 in the second term corresponding to the late reverberation section in the parentheses of the equation (15), the equation (15) is obtained. It is also possible to simplify as shown by the following formula (16).

As understood from the equation (16), the autocorrelation function R _x ^(m) (τ) of the acoustic signal x (n) is a sound source with the exponential attenuation part e ^−Δμ of the indoor impulse response h (μ) as a weight value. It is expressed as a linear sum (weighted sum) of the autocorrelation function R _s ^(m) (τ) of the signal s (n).

数式(17)および数式(18)における記号Ｆ[ ]はフーリエ変換を意味し、記号Ｐ_S(k,m)は音源信号ｓ(n)のパワースペクトルを意味する。 The Fourier transform of the autocorrelation function is the power spectrum (Wiener-Khinchin's theorem). Therefore, in consideration of Equation (15), the power spectrum P _{E, X} (k, m) of the early reflection component x _E (n) in the acoustic signal x (n) is expressed by the following Equation (17), The power spectrum P _{R, X} (k, m) of the reverberation component x _R (n) is expressed by the following equation (18).

In the equations (17) and (18), the symbol F [] means Fourier transform, and the symbol P _S (k, m) means the power spectrum of the sound source signal s (n).

数式(19)の記号Ｐ_X(k)は、音響信号ｘ(n)のパワースペクトルＰ_X(k)を複数の単位期間にわたり時間的に平均した平均スペクトル（Ｅ_m[Ｐ_X(k,m)]）を意味し、数式(19)の記号Ｐ_E,X(k)は、初期反射成分ｘ_E(n)のパワースペクトルＰ_E,X(k,m)を複数の単位期間にわたり時間的に平均した平均スペクトル（Ｅ_m［Ｐ_E,X(k,m)］）を意味する。音響信号ｘ(n)の平均スペクトルＰ_X(k)は、初期反射成分ｘ_E(n)の平均スペクトルＰ_E,X(k)と後期残響成分ｘ_R(n)の平均スペクトルＰ_R,X(k)との加算に相当する（Ｐ_X(k)＝Ｐ_E,X(k)＋Ｐ_R,X(k)）。 The spectrum X _E (k, m) (X _E (k, m) = F [x _E (n)]) of the initial reflection component x _E (n) of the acoustic signal x (n) is used as the acoustic signal x (n). The Wiener filter β (k) for recovering from the spectrum X (k, m) (X (k, m) = F [x (n)]) is expressed by the following equation (19).

Equation symbol P _X (k) of (19), the power spectrum P _X (k) temporally averaged averaged spectrum over a plurality of unit periods (E _m [P _X (k of the audio signal x (n), m )]), And the symbol P _{E, X} (k) in Equation (19) represents the power spectrum P _{E, X} (k, m) of the initial reflection component x _E (n) over time over a plurality of unit periods. Mean spectrum (E _m [P _{E, X} (k, m)]) averaged. Average spectra P _X of the acoustic signal x (n) (k), the initial reflection components x average spectrum P _E of _{E (n),} the average spectrum P _R of _X (k) and the late reverberation component x _{R (n), X} This corresponds to addition with (k) (P _X (k) = P _{E, X} (k) + P _{R, X} (k)).

数式(20)の記号ｍ1'は、非発話区間Ｈ0が開始してからのフレームの個数を意味し、記号Ｐar_(m1')[ ]は部分和を意味する。 Next, the late reverberation component x _R (n) is examined. The acoustic signal x (n) is divided into an utterance section H1 and a non-utterance section H0 on the time axis. The utterance section H1 is a section where the sound source signal s (n) exists, and the non-speak section H0 is a section where the sound source signal s (n) does not exist (late reverberation component derived from the sound source signal s (n) in the utterance section H1). x _R (n) exists). The late reverberation component x _R (n) is observed mainly in the non-speech interval H0. The power spectrum P _X (k, m) of the acoustic signal x (n) in the non-speaking section H0 is a portion of the power spectrum P _S ^(H1) (k, m) of the sound source signal s (n) in the speaking section H1. It is expressed as a sum, and at the same time, it corresponds to a partial sum of the power spectrum P _{R, X} ^(H1) (k, m) of the late reverberation component. That is, the following formula (20) is established.

The symbol m1 ′ in the equation (20) means the number of frames from the start of the non-speech interval H0, and the symbol Par _{(m1 ′)} [] means a partial sum.

Ratio of power spectrum P _X (k, m) of one unit period to average spectrum P _X (k) of acoustic signal x (n) (hereinafter referred to as “intensity ratio”) P _X (k, m) / P _X Consider (k). Considering the relationship of the formula (20), the intensity ratio P _X (k, m) / P _X (k) is expressed by the following formula (21).

The partial sum Par _{(m1 ′)} [P _{R, X} ^(H1) (k, m)] of Equation (21) is the power spectrum P _{R, X} ^(H1) (k, m ⁾ of the late reverberation component x _R (n). ) When the relationship (Par _{(m1 ′)} [PR _{, X} ^(H1) (k, m)] ≦ PR _{, X} ^(H1) (k, m)) is applied to Equation (21) The following formula (22) is derived.

Further, the average value E _m [P _{R, X} (k, m) / P _X (k)] of the intensity ratio P _{R, X} (k, m) / P _X (k) corresponding to the right side of the equation (22). Is expressed as the following Expression (23) by applying the relationship of Expression (19).

The following formula (24) is derived from the formula (22) and the formula (23).

The Wiener filter β (k) in Expression (19) emphasizes the spectrum X _E (k, m) of the initial reflection component x _E (n) in the spectrum X (k, m) of the acoustic signal x (n). Act on. Therefore, the right side {1-β (k)} of the equation (24) represents the spectrum X of the late reverberation component x _R (n) from the spectrum X (k, m) of the acoustic signal x (n) in the non-speech interval H0. _It acts as a filter that suppresses _R (k, m) (hereinafter referred to as a “complementary winner filter” in consideration of being a complementary winner filter). Therefore, as understood from Equation (24), the late reverberation component is obtained by multiplying the spectrum X (k, m) of the acoustic signal x (n) by the intensity ratio P _X (k, m) / P _X (k). x _R (n) is suppressed. The power spectrum P _X (k, m) the power spectrum P _E (k, m) of the initial reflection components x _E (n) and the late reverberation component x _R (n at the inside speech segment H1 acoustic signal x (n) ) In the power spectrum P _R (k, m) (P _X (k, m) = P _E (k, m) + P _R (k, m)), the intensity ratio P _X (k, m) The average value (expected value) of m) / P _X (k) is 1, and the acoustic signal x (n) is not suppressed.

数式(25)から理解される通り、解析処理部３４は、強度比Ｐ_X(k,m)/Ｐ_X(k)が１以上である場合に調整値Ｇ(k,m)を１（最大値）に設定し、強度比Ｐ_X(k,m)/Ｐ_X(k)が１を下回る場合に調整値Ｇ(k,m)を強度比Ｐ_X(k,m)/Ｐ_X(k)に設定する。なお、調整値Ｇ(k,m)の最大値を１以外の数値に設定した構成や、調整値Ｇ(k,m)の最小値を所定の正数に設定した構成も採用され得る。 Considering the above knowledge, the analysis processing unit 34 of the first embodiment calculates the adjustment value G (k, m) for reverberation component suppression by the calculation of the following equation (25).

As understood from the equation (25), the analysis processing unit 34 sets the adjustment value G (k, m) to 1 (maximum) when the intensity ratio P _X (k, m) / P _X (k) is 1 or more. Value), and when the intensity ratio P _X (k, m) / P _X (k) is less than 1, the adjustment value G (k, m) is set to the intensity ratio P _X (k, m) / P _X (k ). A configuration in which the maximum value of the adjustment value G (k, m) is set to a numerical value other than 1 or a configuration in which the minimum value of the adjustment value G (k, m) is set to a predetermined positive number may be employed.

Incidentally, the average spectrum P _X (k) of the acoustic signal x (n) is expressed by the following formula (26) derived using the above formula (16). In the derivation of Equation (26), the formula of the sum of the geometric series was applied to the sum of the exponential function terms e ^−2Δm′NE as in the derivation of Equation (10).

Since the reverberation time T ₆₀ (the number of samples N ₆₀ ) is the time length until the room impulse response h (μ) is attenuated by 60 dB, the exponential function term e ^{−2ΔN60 in} the equation (26) is sufficiently ^larger than 1. (E ^−2ΔN60 << 1). Therefore, Equation (26) is approximated by Equation (27) below.

As understood from the equation (27), the average spectrum P _X (k) applied to the calculation of the adjustment value G (k, m) depends on the reverberation time T ₆₀ (N ₆₀ ). That is, assuming that the adjustment value G (k, m) is calculated using the average (simple moving average) of the power spectra P _X (k, m) over M unit periods as the average spectrum P _X (k), the adjustment is performed. In order to effectively suppress the reverberation component of the acoustic signal x (n) using the value G (k, m), the power spectrum P _X (k, m) applied to the calculation of the average spectrum P _X (k). Must be set to an appropriate numerical value corresponding to the reverberation time T ₆₀ of the acoustic signal x (n).

The storage device 24 in FIG. 1 holds the relationship between the reverberation time T ₆₀ and the average number M that are selected in advance so that the reverberation component is effectively suppressed. Specifically, the storage device 24 stores a coefficient (hereinafter referred to as “correlation coefficient”) C that defines an approximate line (regression line) that approximates the relationship between the reverberation time T ₆₀ and the average number M. For example, the gradient of a primary regression line that approximates the relationship between the reverberation time T ₆₀ and the average number M is stored in the storage device 24 as a correlation coefficient (regression coefficient) C.

  FIG. 2 is a block diagram of the analysis processing unit 34 of the first embodiment. As shown in FIG. 2, the analysis processing unit 34 according to the first embodiment includes a variable setting unit 42, a coefficient specifying unit 44, an intensity average unit 46, and an adjustment value calculation unit 48.

数式(28)の記号Ｐ_X ^(m)(k)は、第ｍ番目の単位期間に対応する平均スペクトルＰ_X(k)を意味する。記号αは平滑化係数（忘却係数）であり、１未満の正数に設定される（０＜α＜１）。具体的には、平滑化係数αは、過去の平均スペクトルＰ_X ^(m-1)(k)に対する最新（現在）のパワースペクトルＰ_X(k,m)の加重値（各単位期間のパワースペクトルＰ_X(k,m)の平滑化の時定数）に相当する。 The intensity averaging unit 46 is a power spectrum P _X (k, m) (P _X (k, m) of each unit period corresponding to the spectrum X (k, m) of the acoustic signal x (n) calculated by the frequency analysis unit 32. ) = | X (k, m) | ² ) The average spectrum (average intensity of each frequency) P _X (k) is sequentially calculated every unit period. In the above description, the simple moving average of the average number M is mentioned, but in the first embodiment, the exponential moving average of the power spectrum P _X (k, m) expressed by the following formula (28) is averaged for convenience. Calculated as the spectrum P _X (k). That is, the intensity average unit 46 corresponds to an IIR (Infinite Impulse Response) type low-pass filter.

Symbol P _X ^(m) (k) in Expression (28) means an average spectrum P _X (k) corresponding to the m-th unit period. The symbol α is a smoothing coefficient (forgetting coefficient), and is set to a positive number less than 1 (0 <α <1). Specifically, the smoothing coefficient α is a weighted value of the latest (present) power spectrum P _X (k, m) with respect to the past average spectrum P _X ^(m−1) (k) (power spectrum of each unit period). This corresponds to a smoothing time constant of P _X (k, m).

Variable setting unit 42 (reverberation time of e.g. acoustic space acoustic signal x (n) is pronounced and From) reverberation time envisaged in the acoustic signal x (n) to set the T ₆₀ variably. For example, the variable setting unit 42 specifies the reverberation time T ₆₀ by analyzing an acoustic signal recorded when a predetermined measurement signal is emitted into the acoustic space. For example, a time stretched signal (TSP: Time Stretched Pulse) whose frequency changes continuously in time is suitably used as the measurement signal. For the measurement of the reverberation time T ₆₀ , for example, a known technique such as an impulse integration method (Schroeder method) can be arbitrarily used. It is also possible to calculate the reverberation time T ₆₀ from the acoustic signal x (n) using known multi-step linear prediction. For multi-step linear prediction, see, for example, K. Kinoshita, et al., “Suppression of late reverberation effect on speech signal using long-term multiple-step linear prediction”, IEEE Transactions on Audio, Speech and Language processing, 17 (4 ), p.534-545, 2009. Further, a configuration in which the variable setting unit 42 acquires the reverberation time T ₆₀ specified by the user through an operation on the input device (not shown) may be employed.

数式(29)の記号Ｎ_Eは、音響信号ｘ(n)の単位期間内のサンプルの総数であり、記号ｆsは音響信号ｘ(n)のサンプリング周波数である。例えば単位期間内のサンプルの総数Ｎ_Eを５１２個とし、サンプリング周波数を１６ｋＨzとした場合、平滑化係数αと平均個数Ｍとの関係は以下の数式(30)で表現される。強度平均部４６は、係数特定部４４が残響時間Ｔ₆₀に応じて設定した平滑化係数αを適用した数式(28)の演算で平均スペクトルＰ_X(k)を算定する。数式(29)や数式(30)の平滑化係数αは、平均個数Ｍの単純移動平均に近似する数式(28)の指数移動平均の平滑化係数に相当する。したがって、平均個数Ｍが大きいほど最新のパワースペクトルＰ_X(k,m)に対する平滑化係数αは小さい数値となる。換言すると、平均個数Ｍが大きいほど、過去の平均スペクトルＰ_X ^(m-1)(k)に対する平滑化係数(１−α)は大きい数値となる。

The coefficient specifying unit 44 specifies the smoothing coefficient α according to the reverberation time T ₆₀ set by the variable setting unit 42. Specifically, the coefficient specifying unit 44 specifies the average number M corresponding to the reverberation time T ₆₀ based on the relationship defined by the correlation coefficient C in the storage device 24, and the smoothing corresponding to the average number M. The conversion factor α is calculated by the following equation (29).

Symbol N _E of equation (29) is the total number of samples in the unit period of the audio signal x (n), the symbol fs is the sampling frequency of the audio signal x (n). For example the total number N _E of the sample in the unit period and 512, when the sampling frequency is 16 kHz, the relationship between the average number M and the smoothing coefficient α is expressed by the following equation (30). The intensity average unit 46 calculates the average spectrum P _X (k) by the calculation of Expression (28) using the smoothing coefficient α set by the coefficient specifying unit 44 according to the reverberation time T ₆₀ . The smoothing coefficient α in Expression (29) or Expression (30) corresponds to the smoothing coefficient of the exponential moving average in Expression (28) that approximates the average number M of simple moving averages. Therefore, the smoothing coefficient α for the latest power spectrum P _X (k, m) becomes smaller as the average number M is larger. In other words, the smoothing coefficient (1-α) with respect to the past average spectrum P _X ^(m−1) (k) becomes a larger numerical value as the average number M is larger.

The adjustment value calculation unit 48 in FIG. 2 calculates each frequency by the calculation of Expression (25) using the average spectrum P _X (k) (P _X ^(m) (k)) calculated by the intensity average unit 46 for each unit period. The adjustment value G (k, m) is sequentially calculated for each unit period. The adjustment value G (k, m) calculated by the adjustment value calculation unit 48 is the reverberation component suppression processing by the reverberation adjustment unit 36 of FIG. 1 (the adjustment value G (for the spectrum X (k, m) of the acoustic signal x (n)). applied to multiplication of k, m).

In the first embodiment described above, the adjustment value G (k, m) is calculated according to the average spectrum P _X (k) of the power spectrum P _X (k, m) of the acoustic signal x (n). The reverberation component of the acoustic signal x (n) is simpler than the technique of Patent Document 1 for estimating the prediction filter coefficient of the reverberation component and the technique of Non-Patent Document 1 for generating the inverse filter by estimating the transfer function. Can be suppressed. Further, since the smoothing coefficient α (average number M) applied to the calculation of the average spectrum P _X (k) is variably set according to the reverberation time T ₆₀ , the smoothing coefficient α is fixed to a predetermined value. As compared with the above, there is an advantage that the reverberation component of the acoustic signal x (n) is effectively suppressed.

<Relationship between reverberation time T ₆₀ and average number M>
A suitable relationship between the reverberation time T ₆₀ and the average number M will be described in detail below. FIG. 3 is a block diagram of an analysis apparatus 200 that analyzes the correlation between the reverberation time T ₆₀ and the average number M. The analysis apparatus 200 in FIG. 3 is an information processing apparatus that analyzes the correlation between the reverberation time T ₆₀ and the average number M and specifies the above-described correlation coefficient C. Like the sound processing apparatus 100 in FIG. This is realized by a computer system including an arithmetic processing device 72 and a storage device 74. The storage device 74 stores a program PGM2 executed by the arithmetic processing device 72 and various data used by the arithmetic processing device 72. For example, 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 74.

The arithmetic processing unit 72 functions as a correlation analysis unit 76 that analyzes the relationship between the reverberation time T ₆₀ and the average number M by executing the program PGM2 stored in the storage device 74. In FIG. 3, the analysis device 200 separate from the sound processing device 100 is illustrated, but the correlation analysis unit 76 may be mounted on the sound processing device 100.

FIG. 4 is a flowchart of an analysis process in which the correlation analysis unit 76 analyzes a preferable relationship between the reverberation time T ₆₀ and the average number M. For example, the analysis process is executed in response to an instruction from the user. When the analysis process is started, the correlation analysis unit 76 sets the reverberation time T ₆₀ to a predetermined value (a numerical value that can be set by the variable setting unit 42) (S11), and a pseudo-random number b that defines the indoor impulse response h (μ). (μ) (specifically, a seed value defining the pseudo random number b (μ)) is set (S12). In addition, the correlation analysis unit 76 sets the average number M to a predetermined value (candidate value) (S13).

The correlation analysis unit 76 sets the acoustic signal x (n) (S14). Specifically, as expressed by the following mathematical formula (31), the correlation analysis unit 76 performs the indoor impulse response h (μ) corresponding to the pseudo random number b (μ) set in step S12 (Mathematical formula (3)). Is convolved with a predetermined sound source signal s (n) prepared in advance to generate an acoustic signal x (n).

  Further, the correlation analysis unit 76 generates the acoustic signal y (n) by the reverberation suppression process for the acoustic signal x (n) set in step S14 (S15). The content of the reverberation suppression process is the same as the above-described process executed by the sound processing apparatus 100 (analysis processing unit 34, reverberation adjustment unit 36). That is, the correlation analysis unit 76 adjusts the adjustment value G (k, m) calculated from the acoustic signal x (n) by the calculation of Expression (25) using the average number M (smoothing coefficient α) set in Step S13. Is applied to the acoustic signal x (n) to generate the acoustic signal y (n).

The correlation analysis unit 76 generates the target sound signal t (n) (S16). The target sound signal t (n) means an acoustic signal to be extracted from the acoustic signal x (n). Specifically, as expressed by the following formula (32), the acoustic signal x ( This corresponds to the initial reflection component x _E (n) (that is, the acoustic signal y (n) when the dereverberation suppression process is complete) within the initial reflection interval (number of samples N _E ) among n).

The correlation analyzing unit 76 calculates an index of the reverberation component suppression effect (hereinafter referred to as “suppression effect index”) δ (S17). The suppression effect index δ is the purpose of the acoustic signal y (n) after suppression of the reverberation component using the target sound reverberation ratio TIRx and the adjustment value G (k, m) of the acoustic signal x (n) before suppression of the reverberation component. The difference from the sound reverberation ratio TIRy (δ = TIRy−TIRx). The target sound reverberation ratio TIR (Target-to-Interference Ratio) means the intensity ratio between the target sound component (Target) and the reverberation component (Interference). Specifically, the correlation analysis unit 76 is calculated by the following equation (33) using the acoustic signal x (n) generated in step S14 and the target sound signal t (n) generated in step S16. Target sound calculated by the following equation (34) using the target sound reverberation ratio TIRx, the acoustic signal y (n) generated in step S15, and the target sound signal t (n) generated in step S16. The suppression effect index δ is calculated from the reverberation ratio TIRy.

  The symbol l in the equations (33) and (34) means the unit period number, and the symbol N is set to a predetermined positive number. As understood from the above description, the suppression effect index δ becomes larger as the suppression effect of the reverberation component is larger (the target sound reverberation ratio TIRy is larger than the target sound reverberation ratio TIRx).

  When the suppression effect index δ is calculated according to the above procedure, the correlation analysis unit 76 determines whether or not the suppression effect index δ is calculated for all numerical values of the average number M (S18). When the determination result is negative (S18: NO), the correlation analysis unit 76 sets the average number M to a value different from the current value (S13), and calculates the suppression effect index δ for the changed average number M. (S14-S17).

  When the suppression effect index δ is calculated for all values of the average number M (S18: YES), the relationship between the average number M and the suppression effect index δ is specified as illustrated in FIG. As can be understood from the description with reference to Equation (27) and FIG. 5, the suppression effect index δ (the reverberation component suppression effect) changes according to the average number M. The correlation analysis unit 76 specifies a numerical value (hereinafter referred to as “optimum number”) M0 that maximizes the suppression effect index δ among a plurality of average numerical values M (S19).

The correlation analysis unit 76 determines whether or not the optimum number M0 has been specified for all types of pseudo-random numbers b (μ) prepared in advance (S20). When the determination result is negative (S20: NO), the correlation analysis unit 76 changes the pseudorandom number b (μ) (S12), and calculates the optimum number M0 for the changed pseudorandom number b (μ) (S13). To S19). On the other hand, when identifying the optimum number M0 for all types of pseudo-random number b (μ) (S20: YES ), the correlation analysis unit 76, whether to identify the optimal number M0 for all the numerical values of reverberation time T ₆₀ Determine (S21). When the determination is negative (S21: NO), the correlation analysis unit 76 sets the reverberation time T ₆₀ to separate numbers currently (S11), the reverberation time T ₆₀ after changing the pseudo-random number b (mu ) The optimum number M0 for each is specified (S12 to S20).

When the optimum number M0 of each pseudorandom number b (μ) is calculated for all the values of the reverberation time T ₆₀ (S21: YES), the relationship between the reverberation time T ₆₀ and the average number M (optimum number M0) as shown in FIG. Is specified for each pseudo-random number b (μ). The circles in FIG. 6 are the optimum number M0 experimentally selected so that the suppression effect index becomes the maximum from the result of actually measuring the indoor impulse response. It can be confirmed in FIG. 6 that even when the indoor impulse response h (μ) is modeled by the equation (3), the same result as that obtained by actually measuring the indoor impulse response is observed.

As shown in FIG. 6, the correlation analysis unit 76 specifies an approximate line L that approximates the correlation between the reverberation time T ₆₀ and the average number M (optimum number M0) over a plurality of pseudo-random numbers b (μ) (S22). ). For example, a primary regression line that approximates the correlation between the reverberation time T ₆₀ and the average number M is specified as the approximate line L. As described above, the approximate line L expresses an approximate correlation between the reverberation time T ₆₀ and the average number M selected under the condition that the suppression effect index δ is maximized. Specifically, as is understood from FIG. 6, the approximation line L defines an approximate relation between reverberation time T ₆₀ as the average number M as the reverberation time T ₆₀ is longer to increase the average number M . As described above with reference to Equation (29), the smoothing coefficient α for the latest power spectrum P _X (k, m) becomes smaller as the average number M increases, so that the longer the reverberation time T _{60, the} smoother it becomes. It can also be said that the approximate line L defines the relationship between the reverberation time T ₆₀ and the smoothing coefficient α so that the coefficient α decreases. In other words, the smoothing coefficient (1-α) for the past average spectrum P _X ^(m−1) (k) increases as the reverberation time T ₆₀ is longer (average number M is larger).

The variable that defines the approximate line L specified by the correlation analysis unit 76 in the above procedure is stored as the correlation coefficient C in the storage device 24 of the sound processing apparatus 100, and the smoothing coefficient α (average number M) by the coefficient specifying unit 44 is stored. Applied to the settings. For example, when the approximate line L is expressed by a linear regression line as illustrated above, the gradient of the approximate line L is stored in the storage device 24 as the correlation coefficient C. The coefficient specifying unit 44 of the sound processing apparatus 100 specifies the average number M having the relationship of the approximate line L defined by the correlation coefficient C with respect to the reverberation time T ₆₀ set by the variable setting unit 42.

As understood from the above description, the average number M corresponding to the smoothing coefficient α set by the coefficient specifying unit 44 and the reverberation time T ₆₀ set by the variable setting unit 42 are approximate lines specified by the correlation analyzing unit 76. L relationship. That is, the average number M (smoothing coefficient α) and the reverberation time T ₆₀ satisfy the approximate relationship (approximate line L) selected under the condition that the suppression effect index δ is maximized. Therefore, the above-described effect that the reverberation component of the acoustic signal x (n) can be effectively suppressed is particularly remarkable.

Second Embodiment
A second embodiment of the present invention will be described below. In addition, about the element which an effect | action and function are the same as that of 1st Embodiment in each form illustrated below, the reference | standard referred by description of 1st Embodiment is diverted, and each detailed description is abbreviate | omitted suitably.

The complementary winner filter {1-β (k)} of the above-described equation (24) is expressed by the following equation (35) by applying the equations (19) and (27).

FIG. 7 is a graph depicting the relationship of Equation (35) between the reverberation time T ₆₀ and the numerical value (gain) of the complementary Wiener filter {1-β (k)}. It can be seen from FIG. 7 that the value of the complementary Wiener filter {1-β (k)} increases (the suppression effect of the reverberation component decreases) as the reverberation time T ₆₀ is longer. The intensity ratio P _X (k, m) / P _X (k) that is the basis of the adjustment value G (k, m) is expressed by the complementary Wiener filter {1-β (k )} Since the following values are set, in the first embodiment in which the adjustment value G (k, m) is calculated by the equation (25), the reverberation due to the adjustment value G (k, m) increases as the reverberation time T ₆₀ increases. There is a tendency for the component suppression effect to decrease (intensity ratio P _X (k, m) / P _X (k) increases).

強度比Ｐ_X(k,m)/Ｐ_X(k)が１を下回る場合、調整値Ｇ(k,m)は、第１実施形態と同様の強度比Ｐ_X(k,m)/Ｐ_X(k)と補正係数Ａ1および補正係数Ａ2とに応じた数値に設定される。補正係数Ａ1は、残響時間Ｔ₆₀に応じた強度比Ｐ_X(k,m)/Ｐ_X(k)の変動を補償するための係数であり、変数設定部４２が設定した残響時間Ｔ₆₀に応じて可変に設定される。具体的には、残響時間Ｔ₆₀が長いほど強度比Ｐ_X(k,m)/Ｐ_X(k)が増加するという傾向が補償されるように、残響時間Ｔ₆₀が長いほど補正係数Ａ1は減少する。例えば、以下の数式(37)で表現されるように、残響時間Ｔ₆₀を指数の分母に含む指数関数が補正係数Ａ1として好適に採用される。数式(37)の記号Ｔ_Eは、単位期間の時間長（サンプル数Ｎ_E）を意味する。

Considering the above tendency, the adjustment value calculation unit 48 of the second embodiment calculates the adjustment value G (k, m) of each frequency by the following expression (36) instead of the above expression (25). Calculated for each unit period.

When the intensity ratio P _X (k, m) / P _X (k) is less than 1, the adjustment value G (k, m) is the same as the intensity ratio P _X (k, m) / P _{X in} the first embodiment. The value is set according to (k), the correction coefficient A1, and the correction coefficient A2. Correction factor A1, the intensity ratio P _X (k, m) corresponding to the reverberation time T ₆₀ is a factor to compensate for variations in / P _X (k), the reverberation time T ₆₀ the variable setting unit 42 has set It is variably set accordingly. Specifically, as the reverberation time T ₆₀ is longer intensity ratio P _X (k, m) as tendency / P _X (k) is increased is compensated, as reverberation time T ₆₀ is greater correction factor A1 is Decrease. For example, as represented by the following equation (37), an exponential function including reverberation time T ₆₀ in the denominator of the exponent is preferably employed as a correction coefficient A1. The symbol T _{E in} Equation (37) means the time length of the unit period (number of samples N _E ).

  On the other hand, the correction coefficient A2 is set to an arbitrary positive number less than 1 (0 <A2 <1). For example, the correction coefficient A2 is variably set according to an instruction from the user to the input device (not shown). The smaller the correction coefficient A2, the stronger the reverberation component suppression effect. Note that the correction coefficient A1 or the correction coefficient A2 can be omitted from the equation (36).

In the second embodiment, the same effect as in the first embodiment is realized. In the second embodiment, since the correction coefficient A1 corresponding to the reverberation time T ₆₀ is applied to the calculation of the adjustment value G (k, m), even when the reverberation time T ₆₀ is long acoustic signal x (n) There is an advantage that the reverberation component can be sufficiently suppressed.

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

係数特定部４４は、変数設定部４２が設定した残響時間Ｔ₆₀に対し、相関解析部７６が特定した近似線Ｌの関係を充足するように平均個数Ｍを設定する。強度平均部４６は、係数特定部４４が設定した平均個数Ｍを適用した数式(38)の演算で平均スペクトルＰ_X(k)を算定する。なお、平均個数ＭのパワースペクトルＰ_X(k,m)の各々を加重する加重移動平均で平均スペクトルＰ_X(k)を算定することも可能である。 (1) In the above-described embodiments, the exponential moving average of the power spectrum P _X (k, m) of the acoustic signal x (n) is calculated as the average spectrum P _X (k). As described above, the intensity average unit 46 can calculate a simple moving average of the power spectra P _X (k, m) over the average number M as the average spectrum P _X (k). That is, the intensity averaging unit 46 corresponds to a FIR (Finite Impulse Response) type low-pass filter.

The coefficient specifying unit 44 sets the average number M so as to satisfy the relationship of the approximate line L specified by the correlation analysis unit 76 with respect to the reverberation time T ₆₀ set by the variable setting unit 42. The intensity average unit 46 calculates the average spectrum P _X (k) by the calculation of Expression (38) using the average number M set by the coefficient specifying unit 44. It is also possible to calculate the average spectrum P _X (k) by a weighted moving average that weights each of the average number M of power spectra P _X (k, m).

As understood from the above description, the intensity average unit 46 is a moving average of the intensity (power spectrum P _X (k, m)) of the acoustic signal x (n) to which the moving average coefficient specified by the coefficient specifying unit 44 is applied. Are included as elements for calculating the average intensity (average spectrum P _X (k)), and the moving average coefficient is the smoothing coefficient α applied to the exponential moving average (formula (28)) and the simple moving average (formula (38 )) And the average number M applied to the weighted moving average.

(2) In each of the above-described embodiments, the relationship between the reverberation time T ₆₀ and the average number M is defined by the approximate line L (correlation coefficient C), but the power spectrum P _X (k, m) of the acoustic signal x (n). ) Is calculated as an average spectrum P _X (k), the relationship between the reverberation time T ₆₀ and the smoothing coefficient α is defined by an approximate line L (correlation coefficient C). It is also possible. The coefficient specifying unit 44 specifies the smoothing coefficient α directly (that is, without setting the average number M) so as to satisfy the relationship of the approximate line L with respect to the reverberation time T ₆₀ set by the variable setting unit 42. .

(3) In each of the above embodiments, the average number M (smoothing coefficient α) corresponding to the reverberation time T ₆₀ is specified using the correlation coefficient C (approximate line L) stored in the storage device 24. The method for specifying the average number M according to the reverberation time T ₆₀ is not limited to the above examples. For example, a reference table in which each value of the reverberation time T ₆₀ is associated with each value of the average number M (smoothing coefficient α) is stored in the storage device 24, and is referenced to the reverberation time T ₆₀ set by the variable setting unit 42. A configuration in which the coefficient specifying unit 44 sets the corresponding average number M in the table is also employed. The numerical values of the reverberation time T ₆₀ and the average number M corresponding to each other in the lookup table are in a relationship defined by the approximate line L.

(4) The acoustic processing device 100 of each embodiment described above can be mounted on each terminal device in a voice call system that transmits and receives acoustic signals between a plurality of terminal devices. The acoustic processing device 100 of each terminal device generates an acoustic signal y (n) by reverberation suppression processing on the acoustic signal x (n) generated by the sound collection device (signal supply device 12) and transmits the acoustic signal y (n) to the counterpart terminal device. To do. The reverberation time T ₆₀ is set by the variable setting unit 42 by analyzing an acoustic signal recorded when sound corresponding to a measurement signal such as a time extension signal is radiated from the terminal device. According to the above configuration, there is an advantage that a clear voice in which a reverberation component given in an acoustic space where each terminal device exists is suppressed can be exchanged between the terminal devices. The above-described acoustic processing apparatus 100 can also be applied to a recording apparatus that records sound in an acoustic space. The sound processing device 100 in the recording device generates a sound signal y (n) by the reverberation suppression process for the sound signal x (n) generated by the sound collecting device and stores the sound signal y (n) in the storage device 24.

(5) In each embodiment described above, the intensity ratio P _X (k, m) / of the power spectrum P _X (k, m) of one unit period to the average spectrum P _X (k) of the acoustic signal x (n). Although P _X (k) is exemplified as the adjustment value G (k, m), the method for calculating the adjustment value G (k, m) is changed as appropriate. For example, the moving average coefficient (smoothing coefficient α, average number M) at the time of calculation of the average spectrum P _X (k) is averaged by the moving average of the power spectrum P _X (k, m) to which a moving average coefficient is applied. The spectrum Q (k) is calculated, and the intensity ratio Q (k) / P _X (k) of the average spectrum Q (k) to the average spectrum P _X (k) is calculated as the adjustment value G (k, m) (ie, It is also possible to replace the power spectrum P _X (k, m) in Equation (25) with the average spectrum Q (k). The average spectrum Q (k) is a simple moving average of the power spectrum P _X (k, m) over a number of unit periods less than the average number M applied to the calculation of the average spectrum P _X (k), or the average spectrum P It is an exponential moving average of the power spectrum P _X (k, m) to which a smoothing coefficient exceeding the smoothing coefficient α applied to the calculation of _X (k) is applied.

Power spectrum P _X (k, m) in each embodiment described above and illustrated average spectrum Q (k) In this modification, mean spectrum P _X (k) power spectrum to high trackability compared to P _X ( It is included as an intensity index Λ (k) following the time change of k, m). That is, in a preferred embodiment of the present invention, the relative ratio Λ () of the intensity index Λ (k) (P _X (k, m) or Q (k)) to the average spectrum P _X (k) of the acoustic signal x (n). k) / P _X (k) is calculated as the adjustment value G (k, m).

In each of the above-described embodiments, the adjustment value G (k, m) for suppressing the reverberation component of the acoustic signal x (n) is exemplified. However, the present invention is also applied to the case where the reverberation component of the acoustic signal x (n) is emphasized. Applies. For example, an adjustment value {λ−G (k, m)} obtained by subtracting the adjustment value G (k, m) calculated by Expression (25) from a predetermined value λ (for example, 1) is applied to the acoustic signal x (n). By doing so, it is possible to generate an acoustic signal y (n) that emphasizes the reverberation component. As understood from the above description, the adjustment value calculation unit 48 sets the adjustment value for adjusting (suppressing or enhancing) the reverberation component of the acoustic signal x (n) to the average intensity (average spectrum P _X (k)). It is included as an element to be calculated accordingly.

(6) In each embodiment described above, has been set a common reverberation time T ₆₀ over the entire frequency, it is possible to individually set the reverberation time T ₆₀ for each predetermined band. The coefficient specifying unit 44 individually sets a moving average coefficient (smoothing coefficient α, average number M) corresponding to the reverberation time T ₆₀ of each band for each band. In the above configuration, a configuration in which an approximate correlation (approximate line L) between the reverberation time T _{60 at} which the suppression effect index δ is maximum and the moving average coefficient is individually prepared for each band is preferable.

(7) In each of the above embodiments, the sound processing apparatus 100 is exemplified as a single apparatus. However, a part of the functions of the sound processing apparatus 100 is shared with an external device (for example, a web server) that can communicate with the sound processing apparatus 100. It is also possible. For example, the external device calculates the reverberation time T ₆₀ in response to a request from the sound processing device 100 and specifies the moving average coefficient (smoothing coefficient α, average number M) according to the reverberation time T ₆₀ to perform acoustic processing. Notify the device 100. In the sound processing apparatus 100, the coefficient specifying unit 44 obtains a moving average coefficient from an external device, and the intensity average unit 46 calculates an average spectrum P _X (k) based on the moving average to which the moving average coefficient is applied. Therefore, the variable setting unit 42 is omitted from the sound processing apparatus 100.

How external device to specify reverberation time T ₆₀ is arbitrary. For example, the acoustic characteristics of the acoustic space designated by the user (for example, the average sound absorption coefficient, spatial size, sound source, and listening point position) are notified from the acoustic processing device 100 to the external device, and the external device is a known method such as a mirror image method. The reverberation evaluation technique employs a configuration for calculating the reverberation time T ₆₀ from the acoustic characteristics of the acoustic space. It is also possible to transmit an acoustic signal recorded at the time of sound emission according to the measurement signal from the acoustic processing device 100 to the external device, and the external device calculates the reverberation time T ₆₀ from the acoustic signal. As understood from the above description, the coefficient specifying unit 44 is included as an element for specifying the moving average coefficient (smoothing coefficient α and average number M) according to the reverberation time T ₆₀ , and calculates the moving average coefficient. It does not matter whether the coefficient specifying unit 44 executes the coefficient specifying unit 44 or acquires the moving average coefficient calculated by the external device.

(8) In each of the above-described embodiments, the acoustic processing device 100 including the analysis processing unit 34 and the reverberation adjustment unit 36 has been exemplified. However, the present invention can be applied to an adjustment value calculation device that calculates the adjustment value G (k, m). Can be specified. The adjustment value calculation device is configured to include the analysis processing unit 34 in each of the above-described embodiments, and the presence or absence of the reverberation adjustment unit 36 is not questioned. The adjustment value G (k, m) calculated by the adjustment value calculation device is provided to the acoustic processing device including the reverberation adjustment unit 36 and applied to the reverberation suppression processing of the acoustic signal x (n).

(9) In addition to a reverberant component in a narrow sense caused by reflection or scattering in an acoustic space, a reverberant component (resonance component) such as a musical performance of a musical instrument can be implied by the reverberant component. Specifically, the present invention can also be applied to the adjustment of the resonance component (bottle sound, box sound) of a stringed instrument such as a violin or the like by the sound board of a keyboard instrument such as a piano as in the above-described embodiments. Is possible. That is, the reverberation component of the present invention means a component that attenuates with time (attenuation component).

DESCRIPTION OF SYMBOLS 100 ... Acoustic processing device, 12 ... Signal supply device, 14 ... Sound emission device, 22 ... Arithmetic processing device, 24 ... Memory | storage device, 32 ... Frequency analysis part, 34 ... Analysis processing part, 36 ... ... reverberation adjustment part, 38 ... waveform synthesis part, 42 ... variable setting part, 44 ... coefficient specifying part, 46 ... intensity average part, 48 ... adjustment value calculation part, 200 ... analysis device, 72 ... Arithmetic processing device, 74... Storage device, 76.

Claims (5)

Coefficient specifying means for specifying a moving average coefficient according to the reverberation time;
Intensity average means for calculating an average intensity by a moving average of the intensity of the acoustic signal applied with the moving average coefficient specified by the coefficient specifying means;
Adjustment value calculation means for calculating an adjustment value for adjusting the reverberation component of the acoustic signal according to the average intensity;
Reverberation adjusting means for causing the adjustment value calculated by the adjustment value calculating means to act on the acoustic signal. 2. The acoustic processing according to claim 1, wherein the adjustment value calculation unit calculates the adjustment value according to a ratio of an intensity index that follows the intensity of the acoustic signal with higher followability than the average intensity to the average intensity. apparatus. The adjustment value calculation means is based on the ratio of the intensity index that follows the intensity of the acoustic signal with high followability compared to the average intensity to the average intensity and the correction coefficient corresponding to the reverberation time. The sound processing apparatus according to claim 2, wherein the adjustment value is calculated. The coefficient specifying means specifies a smoothing coefficient that becomes a smaller numerical value as the reverberation time is longer as the moving average coefficient,
The sound processing apparatus according to claim 2 or 3, wherein the intensity averaging means calculates an exponential moving average of the intensity of the acoustic signal obtained by applying the smoothing coefficient to the latest intensity of the acoustic signal as the average intensity. The adjustment value calculation means calculates an adjustment value for suppressing a reverberation component of the acoustic signal,
The coefficient specifying unit is configured on the condition that the difference between the target sound reverberation ratio of the acoustic signal before suppression of the reverberation component and the target sound reverberation ratio of the acoustic signal after suppression applied with the adjustment value is maximized. The sound processing apparatus according to claim 1, wherein a moving average coefficient corresponding to the reverberation time is specified so as to satisfy an approximately selected relationship.

Images