JP5077847B2 - Reverberation time estimation apparatus and reverberation time estimation method - Google Patents

Description

  The present invention relates to a reverberation time estimation apparatus and method for estimating reverberation time of a system, and in particular, blind estimation of reverberation time without using an original sound signal by using a frequency-sequence modulation spectrum obtained from a time-series acoustic signal. The present invention relates to a reverberation time estimation apparatus and method.

  The acoustic reverberation time is a parameter indispensable to know the reverberation characteristics of the room. In recent years, the reverberation time has been used for Fo estimation, speech recognition, and recovery of the original sound signal for an acoustic signal. In order to obtain such reverberation time, conventionally, it is necessary to measure the transfer characteristics of a room, and this transfer characteristic measurement has many restrictions such as the sound source, the condition of people and objects in the room, and the quietness of the room. Exists. In addition, since the conventional reverberation time measurement requires a relatively long time, the measurement may not be possible in an environment where the reverberation characteristics change every moment. From such a viewpoint, it is desired to develop a technique capable of measuring reverberation time in real time.

In order to obtain the reverberation time in real time, it is necessary to perform blind estimation that can estimate the reverberation time only from the observed acoustic signal without measuring the transfer characteristics of the system. Here, “system” refers to a space assumed for analysis of various events. Non-Patent Document 1 discloses a recovery method for recovering the power envelope of an original sound source signal from the power envelope of an acoustic signal affected by reverberation. In the recovery method disclosed in Non-Patent Document 1, the input / output relationship (convolution in the time domain and product in the frequency domain) of the power envelope of the sound transmitted into the room, that is, the modulation degree, the modulation frequency is a variable. A modulation transfer function expressed as a function is used. More specifically, by applying an inverse function (inverse filter) of the modulation transfer function of the system to the power envelope of the observed acoustic signal, the power envelope of the original sound (sound from a sound source without added reverberation) is obtained. Recover. In Non-Patent Document 1, when the power envelope of the acoustic signal to which reverberation is added is restored to the same shape as the power envelope of the original sound by the inverse filter, the reverberation time parameter of the inverse filter is the reverberation time of the system. It is disclosed that the reverberation time parameter of the inverse filter obtained by the process of calculating the inverse filter is estimated as the reverberation time on the assumption that the values are equal.
Furukawa, Kashiwagi, Akagi, “Recovery method of reverberant speech power envelope based on MTF”, IEICE Technical Report, IEICE, April 2002, EA2002-15, SP2002-15, p. 49-54

  However, in the reverberation sound power envelope recovery method disclosed in Non-Patent Document 1 described above, the reverberation time obtained in the process of calculating the inverse filter is the actual reverberation time until the reverberation time is around 0.5 seconds. The estimated reverberation time was in good agreement, but after that the difference between the two gradually increased and could not be said to be in good agreement.

  The present invention has been made in view of such circumstances, and an object of the present invention is to provide a reverberation time estimation device and a reverberation time estimation method capable of estimating an accurate reverberation time without measuring the transfer characteristics of the system. There is to do.

  In order to solve the above-described problem, a reverberation time estimation device according to the present invention includes a power envelope generation unit that generates a time-series power envelope corresponding to the acoustic signal based on the time-series acoustic signal to which reverberation is added. And, based on the power envelope generated by the power envelope generating means, a modulation spectrum generating means for generating a modulation spectrum of a frequency sequence, and based on the modulation spectrum generated by the modulation spectrum generating means, the acoustic signal is Reverberation time estimation means for estimating a reverberation time corresponding to a transfer function related to the reverberation characteristics of the observed system.

  In the above invention, it further comprises main modulation frequency specifying means for specifying a main modulation frequency showing a high modulation spectrum in the modulation spectrum of the frequency sequence, and the reverberation time estimation means adds the transfer function to the modulation spectrum of the frequency sequence. When the inverse transfer function is applied, the modulation spectrum at the main modulation frequency after the application is applied at the main modulation frequency of the modulation spectrum of the frequency sequence corresponding to the time-series original sound signal indicating the original sound to which no reverberation is added. It is preferable that the reverberation time corresponding to the transfer function that substantially matches the modulation spectrum is estimated.

  Further, in this case, the main modulation frequency specifying means is configured to obtain an autocorrelation function for the power envelope and to specify the reciprocal of the time shift amount at which the autocorrelation function shows a peak as the main modulation frequency. It is preferable.

  In the above invention, further comprising a low-pass filter applied to the power envelope generated by the power envelope generating means, wherein the main modulation frequency specifying means is based on the power envelope output from the low-pass filter, It is preferably configured to identify the main modulation frequency.

  In the above invention, it further comprises band dividing means for dividing the acoustic signal into a plurality of channels, and channel determining means for determining a channel used for reverberation time estimation from each of the channels divided by the band dividing means. Is preferred.

  In this case, the power envelope generating means is configured to generate a power envelope for each channel band-divided by the band dividing means, and the power envelope generating means includes a power envelope generated by the power envelope generating means. And further comprising a high level part detecting means for detecting a high level part exceeding a predetermined reference value, wherein the channel determining means performs reverberation time estimation based on the high level part detected by the high level part detecting means. It is preferably configured to determine the channel to be used.

  In this case, the channel determination means determines whether or not a minute peak exists between the two high level parts detected by the high level part detecting means, and the minute peak exists. In this case, it is preferable that the channel is excluded from the channels used for estimation.

  In the above invention, the channel determination means determines whether or not a valley exists in the high level part detected by the high level part detection means. Is preferably excluded from the channels used for estimation.

  The reverberation time estimation method of the present invention includes a step of generating a time-series power envelope corresponding to the acoustic signal based on the time-series acoustic signal to which reverberation is added, and a frequency based on the generated power envelope. Generating a modulation spectrum of the sequence; and estimating a reverberation time corresponding to a transfer function related to a reverberation characteristic of the system in which the acoustic signal is observed based on the generated modulation spectrum.

  According to the reverberation time estimation apparatus and reverberation time estimation method of the present invention, it is possible to estimate an accurate reverberation time without measuring the transfer characteristics of the system.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

(Embodiment 1)
The first embodiment is a reverberation time estimation device mainly configured by a hardware digital signal processing circuit.

[Configuration of reverberation time estimation device]
FIG. 1 is a block diagram showing a configuration of a reverberation time estimation apparatus according to Embodiment 1 of the present invention. As shown in FIG. 1, a reverberation time estimation apparatus 1 according to the present embodiment includes a microphone 2 for inputting room sound, and an A / D converter for analog sound signals captured by the microphone 2. / D converter 3, digital signal processing circuit 4 that performs signal processing on the digital acoustic signal output from A / D converter 3, and processing results of digital signal processing circuit 4 are received, and reverberation time estimation processing Is provided with an arithmetic circuit 5, a memory 6, and a liquid crystal display unit 7 for displaying the reverberation time estimated by the arithmetic circuit 5.

但し、ＬＰＦ［・］は低域通過フィルタを示しており、Ｈｉｌｂｅｒｔ（・）はヒルベルト変換を示している。また、会話音声の変動リズムを考慮すると、主要な音響情報は変調周波数２０Ｈｚまでに存在することから、本実施の形態では、低域通過フィルタのカットオフ周波数を２０Ｈｚとしている。As illustrated in FIG. 1, the digital signal processing circuit 4 includes functional blocks of a power envelope generation unit 41, a low-pass filter 42, a main modulation frequency acquisition unit 43, and a normalized modulation spectrum generation unit 44. The power envelope generation unit 41 generates a power envelope signal from the digital acoustic signal output from the A / D converter 3. This power envelope signal is obtained by squaring the time envelope information of the amplitude of the acoustic signal. The power envelope signal generation unit 41 is configured to perform signal processing represented by the following equation.

However, LPF [•] indicates a low-pass filter, and Hilbert (•) indicates Hilbert transform. Considering the fluctuation rhythm of the conversational voice, the main acoustic information exists up to the modulation frequency of 20 Hz. Therefore, in this embodiment, the cutoff frequency of the low-pass filter is set to 20 Hz.

A low-pass filter 42 is provided before the main modulation frequency acquisition unit 43, and a power envelope signal from which high-frequency components have been removed by the low-pass filter 42 is given to the main modulation frequency acquisition unit 43. The main modulation frequency acquisition unit 43 obtains an autocorrelation function of the power envelope signal, and determines a time shift amount (shift time) between the power envelope and the autocorrelation function as the main modulation frequency f _md . In the present embodiment, since the main modulation frequency f _md is a relatively low modulation frequency, it is considered that the main modulation frequency f _md appears up to 2 Hz. Therefore, the cutoff frequency of the low-pass filter 42 is set to 2 Hz. In this way, by providing the low-pass filter 42 in the previous stage of the main modulation frequency acquisition unit 43, erroneous detection of the main modulation frequency f _md can be prevented. In the present embodiment, the number of modulation frequencies acquired as the main modulation frequency is one. However, the present invention is not limited to this, and the number of acquisitions of the main modulation frequency may be plural. However, from the viewpoint of circuit complexity and cost increase, it is better that the number of acquisitions of the main modulation frequency is small, and it is particularly preferable to use one or two. In the present embodiment, the cut-off frequency of the low-pass filter 42 is 2 Hz. However, the present invention is not limited to this, and other cut-off frequencies such as 5 Hz or 10 Hz may be employed. However, if the cut-off frequency is set to a high value, the power envelope may contain a lot of high frequency components of noise, so it is preferable to set the value to 10 Hz or less.

  The normalized modulation spectrum generation unit 44 includes an FFT circuit, and generates a modulation spectrum signal that is a frequency series digital signal from a power envelope signal that is a time series digital signal. Further, the normalized modulation spectrum generation unit 44 is configured to generate a modulation spectrum signal in which the size of the modulation spectrum is normalized with the modulation spectrum of the DC component (the modulation spectrum when the modulation frequency is 0 Hz).

  Data of main modulation frequency and modulation spectrum output from the digital signal processing circuit 4 is given to the arithmetic circuit 5. The arithmetic circuit 5 is a processor dedicated to the reverberation time estimation process configured by an FPGA or an ASIC. The arithmetic circuit 5 uses a modulation transfer function (hereinafter referred to as MTF) that expresses the degree of modulation in the room (the relationship between the input and output of the power envelope of the sound transmitted into the room) as a function with the modulation frequency as a variable. And estimate the reverberation time of the room.

但し、ｆ_ｍは変調周波数を、ｈ（ｔ）は室内インパルス応答を示す。また、室内音響学でよく使われる統計近似のインパルス応答は、次式で表される。

但し、Ｔ_Ｒは残響時間を、ｎ（ｔ）は白色雑音を示す。上記の式（３）を用いれば、ＭＴＦは、ある残響時間Ｔ_Ｒにおいて、変調周波数ｆ_ｍに対する入出力パワーの割合として表される。室内が拡散音場である場合、減衰過程が指数関数的であることを、波動論を用いて示すことができる。この場合、ＭＴＦの大きさ（変調度）は、次式で示される。

図２は、式（４）のＭＴＦ（変調度）を示すグラフである。図２において、横軸は変調周波数を、縦軸はＭＴＦ（変調度）を示している。例えば、変調周波数ｆ_ｍ＝１０ＨｚのときのＭＴＦは、図２に示すように０．４０２である。これは、入力パワーエンベロープにある変調周波数ｆ_ｍ＝１０Ｈｚの成分の変調度が１であるとすると、その出力パワーエンベロープの同変調周波数における成分の変調度は０．４０２に減衰するということを意味する。この減衰特性は、ある意味で変調周波数に対する低域通過フィルタ特性と解釈することができる。Here, the MTF will be described in detail. MTF is defined by the following equation as the Fourier transform of the square of the impulse response when the characteristics of the transmission line can be described as a linear passive system.

However, _{f m} is the modulation frequency, h (t) denotes the room impulse response. The impulse response of statistical approximation often used in room acoustics is expressed by the following equation.

However, _{T R} is the reverberation time, n (t) denotes white noise. By using the above equation (3), MTF, in some reverberation time _{T R,} expressed as a percentage of the output power with respect to the modulation frequency _{f m.} If the room is a diffuse sound field, it can be shown using wave theory that the decay process is exponential. In this case, the magnitude (modulation degree) of the MTF is expressed by the following equation.

FIG. 2 is a graph showing the MTF (modulation degree) of Equation (4). In FIG. 2, the horizontal axis indicates the modulation frequency, and the vertical axis indicates the MTF (modulation degree). For example, the MTF when the modulation frequency f _m = 10 Hz is 0.402 as shown in FIG. This means that if the modulation degree of the component of the modulation frequency f _m = 10 Hz in the input power envelope is 1, the modulation degree of the component at the modulation frequency of the output power envelope is attenuated to 0.402. To do. This attenuation characteristic can be interpreted in a sense as a low-pass filter characteristic with respect to the modulation frequency.

Next, the principle of reverberation time estimation in this embodiment will be described. FIG. 3A is a graph showing a power envelope of an acoustic signal to which reverberation is not added, and FIG. 3B is a graph showing a modulation spectrum thereof. 4A is a graph showing a power envelope of an acoustic signal when reverberation is added to the sound shown in FIG. 3A, and FIG. 4B is a graph showing a modulation spectrum thereof. Figure 3A, the primary modulation frequency of the acoustic signal in Fig. 4A are both 5 Hz, reverberation time T _R of the reverberation added to the sound signal of Figure 4A is 2.0 seconds. By applying an MTF inverse filter to the modulation spectrum of FIG. 4B, the value at the main modulation frequency (5 Hz) of the modulation spectrum can be returned to the value of the same main modulation frequency of the modulation spectrum of FIG. An appropriate reverberation time can be estimated based on the parameters of the inverse filter.

  As shown in FIGS. 3B and 4B, it can be seen that there is no influence of reverberation in the vicinity of the modulation frequency of 0 Hz (DC component). Further, since the MTF indicates the ratio (ratio) of input / output power and the normalized Fourier transform of the power envelope of the system is the MTF, the modulation spectrum of the original sound is added to the MTF of the system when reverberation is added. It turns out that it decreases according to. Therefore, in the present embodiment, based on the fact that the power of the main modulation frequency is often sufficiently close to the power near 0 Hz in the acoustic modulation spectrum with no reverberation added, the observed acoustic modulation spectrum of 0 Hz , The power at the main modulation frequency of the original acoustic modulation spectrum is the main frequency of the observed acoustic modulation spectrum. An MTF that decreases to the power at is determined, and the reverberation time parameter of the MTF is estimated as the reverberation time of the system. In other words, when the reverberation time estimation apparatus according to the present embodiment applies the MTF inverse filter to the observed acoustic modulation spectrum, the power at the main modulation frequency of the modulation spectrum is the original power. An inverse characteristic corresponding to the MTF is estimated that approximately matches the power at the main modulation frequency of the acoustic modulation spectrum (ie, the power at 0 Hz of the observed acoustic modulation spectrum). The broken line in FIG. 4B shows the MTF that reduces the power at the main modulation frequency of the original acoustic modulation spectrum to the power at the main frequency of the observed acoustic modulation spectrum, and the reverberation time corresponding to this MTF. Presumed.

ここで、“＊”は畳み込みを、ｅ_ｘ（ｔ）及びｅ_ｈ（ｔ）はそれぞれｘ（ｔ）及びｈ（ｔ）のエンベロープを、ｎ_１（ｔ）及びｎ_２（ｔ）は相互に無関係な白色雑音を、ａは振幅項を示している。ｎ_１（ｔ）とｎ_２（ｔ）との相互独立性から、残響信号とそのパワーエンベロープとの間には次の関係があることが知られている。

但し、＜・＞は集合平均を表す。この関係から、ｅ^２ _ｙ（ｔ）はｅ^２ _ｘ（ｔ）とｅ^２ _ｈ（ｔ）との畳み込みで得られることが分かる。Based on such a concept, the arithmetic circuit 5 is specifically configured to execute the following processing. The memory 6 stores the inverse characteristics of the MTF having the reverberation time as a parameter as function data that can be processed by the arithmetic circuit 5. The function data of the inverse filter (inverse characteristic) stored in the memory 6 is prepared for a plurality of reverberation times given in advance as estimation candidates. Here, the inverse filter will be described in more detail. Based on the concept of MTF, an acoustic signal of a sound source to which no reverberation is added (hereinafter referred to as a sound source signal) x (t), an indoor impulse response h (t), and an observed acoustic signal obtained as a convolution thereof ( Hereinafter, the relationship of y (t) (referred to as a reverberation signal) is modeled as follows.

Where “*” is the convolution, e _x (t) and e _h (t) are the envelopes of x (t) and h (t), respectively, and n ₁ (t) and n ₂ (t) are Irrelevant white noise, a indicates the amplitude term. From the mutual independence between n ₁ (t) and n ₂ (t), it is known that the following relationship exists between the reverberation signal and its power envelope.

However, <•> represents a set average. From this relationship, it can be seen that e ² _y (t) is obtained by convolution of e ² _x (t) and e ² _h (t).

次に、パワーエンベロープに関する入出力の関係は、Ｚ［ｅ^２ _ｙ（ｔ）］／Ｚ［ｅ^２ _ｘ（ｔ）］＝Ｚ［ｅ^２ _ｈ（ｔ）］となることから、音源信号の変調スペクトルＺ［ｅ^２ _ｘ（ｔ）］は、次式のように求めることができる。

但し、ｆ_ｓはサンプリング周波数である。ここで、式（１２）の第２項（式（１１））がちようど逆フィルタの特性（残響が積分処理であるのに対して、回復は微分処理に相当する）を表しており、１次のＩＩＲフィルタで実現できることが分かる。この式（１２）の逆フィルタを表す関数データが、メモリ６に複数の残響時間の分だけ記憶されている。演算回路５は、メモリ６から各残響時間の逆フィルタを読み出し、デジタル信号処理回路４から受け付けた主要変調周波数及び変調スペクトルに対して各残響時間の逆フィルタを適用して、各残響時間に対応する音源信号の主要変調周波数における変調スペクトルを算出するように構成されている。Next, consider the relationship in the frequency domain. Since the relational expressions (5) to (10) are actually used in discrete time, z conversion is used here as frequency conversion. When the z conversion is Z [•], the modulation transfer function in the z region of Expression (8) is expressed as the following expression.

Next, since the input / output relationship regarding the power envelope is Z [e ² _y (t)] / Z [e ² _x (t)] = Z [e ² _h (t)], the modulation of the sound source signal is performed. The spectrum Z [e ² _x (t)] can be obtained as follows.

Where f _s is the sampling frequency. Here, the second term of equation (12) (equation (11)) is similar to the characteristics of the inverse filter (reverberation is an integration process, whereas recovery corresponds to a differentiation process). It can be seen that this can be realized with the following IIR filter. Function data representing the inverse filter of equation (12) is stored in the memory 6 for a plurality of reverberation times. The arithmetic circuit 5 reads out the inverse filter of each reverberation time from the memory 6 and applies the inverse filter of each reverberation time to the main modulation frequency and modulation spectrum received from the digital signal processing circuit 4 to cope with each reverberation time. The modulation spectrum at the main modulation frequency of the sound source signal is calculated.

  In addition, the arithmetic circuit 5 selects one of the modulation spectra at the main modulation frequency of the sound source signal corresponding to each reverberation time thus obtained, the magnitude closest to 0 dB (1 in terms of modulation degree). The reverberation time is estimated as the reverberation time of the room. This principle will be described below.

ここで、ｆ_ｍ＝ｆ_ｍｄのときの変調スペクトル及び変調伝達関数ＭＴＦの関係を考える。残響時間Ｔ_Ｒを完全に推定することができたとすると、主要変調周波数ｆ_ｍｄにおける音源信号の変調スペクトルと回復された信号の変調スペクトルとは一致する。また、ｆ_ｍ＝０のときの変調伝達関数ｍ（０）はあらゆるＴ_Ｒに対して１であるから、ｆ_ｍ＝０における音源信号と残響信号の変調スペクトルも常に一致する。これらの条件から、次式が導かれる。

これを式（１３）に代入すると、次式が導かれる。

ここで、変調伝達関数ｍ（ｆ_ｍ）はｆ_ｍの関数であるが、残響時間パラメータＴ_Ｒも見掛け上変数とみることができる。ブラインド推定では、Ｔ_Ｒが不明であるため、式（４）のｍ（ｆ_ｍ）をｍ（ｆ_ｍ，Ｔ_Ｒ）と表せば、求めるべき残響時間Ｔ_Ｒは、次式で表すことができる。

この式により、残響時間のブラインド推定が可能となる。Each modulation spectrum and E _{(f m).} Here, when m (f _m ) = E _h (f _m ) is taken into consideration and the convolution on the right side of equation (10) is expressed in (logarithmic) modulation spectrum expression, the following equation is obtained.

Now _consider the relationship between the modulation spectrum and the modulation transfer function MTF of the time of _f m ₌ f _md. When it was possible to completely estimate the reverberation time T _R, which coincides with the modulation spectrum of the modulated spectrum recovered signal of the sound source signal in the primary modulation frequency f _md. The modulation transfer function m (0) in the case of f m _{= 0} is because it is 1 for all T _R, the modulation spectrum of the source signal and the reverberation signal in the f m _{= 0} also always match. From these conditions, the following equation is derived.

Substituting this into equation (13) leads to the following equation:

Here, the modulation transfer function m (f _m) is a function of f _m, it can also viewed as apparently variable reverberation time parameter T _R. The blind estimation, because _{T R} is unknown, if indicated equation (4) _{m (f} m) of the _{m (f} m, _{T R)} and, the reverberation time _{T R} to be determined, can be represented by the following formula .

This equation enables blind estimation of reverberation time.

[Operation of reverberation time estimation device]
Next, the operation of the reverberation time estimation device will be described. The operator samples the sound with the microphone 2 of the reverberation time measuring apparatus 1 in the room where the reverberation time is measured. The analog acoustic signal output from the microphone 2 is converted into a digital acoustic signal by the A / D converter 3, and this digital acoustic signal (reverberation signal) is given to the digital signal processing circuit 4. The digital acoustic signal given to the digital signal processing circuit 4 is converted into a power envelope by the power envelope generation unit 41, and this power envelope signal is given to the low-pass filter 42 and the normalized modulation spectrum generation unit 44, respectively.

  The power envelope signal from which the high-frequency component has been removed by the low-pass filter 42 is given to the main modulation frequency acquisition unit 43. The main modulation frequency acquisition unit 43 acquires the main modulation frequency by obtaining a time shift amount with respect to the autocorrelation function. The main modulation frequency acquired here is a modulation frequency indicating a particularly high modulation spectrum in the entire modulation frequency. For example, the main modulation frequency matches the frequency when the power envelope is a sine wave.

  In the normalized modulation spectrum generation unit 44, Fourier transform is applied to the power envelope signal to obtain a modulation spectrum in the frequency domain. This modulation spectrum is normalized by the value of the modulation spectrum of the DC component.

  The main modulation frequency and modulation spectrum data output by the digital signal processing circuit 4 are supplied to the arithmetic circuit 5. The arithmetic circuit 5 reads MTF inverse filter data corresponding to a plurality of reverberation times from the memory 6 and applies the inverse filter of each reverberation time to the main modulation frequency and modulation spectrum received from the digital signal processing circuit 4. . Thereby, the modulation spectrum at the main modulation frequency of the sound source signal corresponding to each reverberation time is calculated.

  Next, the arithmetic circuit 5 selects one of the modulation spectra at the main modulation frequency of the sound source signal corresponding to each reverberation time obtained in this way, the magnitude closest to 0 dB, and the reverberation time is selected. Estimated as room reverberation time. Then, the arithmetic circuit 5 drives and controls the liquid crystal display unit 7 to display the estimated reverberation time.

[Evaluation experiment]
An artificial AM signal of Expression (6) was used as a sound source signal for evaluation. The power envelope was a sine wave with a modulation frequency of 5 Hz (modulation degree 1), and a signal obtained by multiplying this by a white noise carrier was used. The modulation spectrum for this power envelope has the same value for the modulation spectrum of 0 Hz and the modulation spectrum of 5 Hz. Next, the room reverberation impulse response defined by Equation (7) was used. Here, per one reverberation time T _R, we were prepared 100 kinds of white noise carrier. Further, reverberation time _{T R} utilized in this evaluation experiment, 0.1 seconds, 0.3 seconds, 0.5 seconds, and a five 1.0 seconds, and 2.0 seconds. Therefore, in this evaluation experiment, a total of 500 pulse responses were prepared, and a reverberation signal was created by convolving these with an artificial signal.

The reverberation time estimation method (hereinafter referred to as the present method) that is the same as the reverberation time estimation performed by the reverberation time estimation apparatus 1 according to the present embodiment for these 500 reverberation signals; The reverberation time was estimated by the reverberation time estimation method (hereinafter referred to as the conventional method) in the reverberation speech power envelope recovery method. FIG. 5 is a graph showing the results of the evaluation experiment. A solid line in the figure indicates an estimation result by the reverberation time estimation apparatus 1 according to the present embodiment, and a broken line in the figure indicates an estimation result by the conventional method. Moreover, the straight line (dotted line) in a figure has shown the ideal estimated value. The present method and the estimated values by the conventional method, shows the average value of the estimated values obtained from 100 different reverberation signals each reverberation time T _R. The error bars in the figure indicate the standard deviation at each reverberation time.

  Compared with the ideal straight line, the estimation result by the conventional method tends to be underestimated as the reverberation time becomes longer, and in particular, tends to saturate when the reverberation time exceeds 1 second. The cause of this phenomenon is that when recovering the power envelope generated from the reverberant signal, high-frequency components on the power envelope that could not be removed by the low-pass filter are emphasized by inverse filtering (differential processing), and their phases Depending on the state, there is an influence on the formation of valleys that define the accuracy of reverberation time estimation (which defines the degree of modulation). On the other hand, the estimation result of this method almost coincides with the ideal straight line, and it can be seen that the reverberation time is accurately estimated. In both methods, as the reverberation time increases, high frequency components on the power envelope are abnormally recovered by the inverse filter. In this case, the conventional method directly affects the high frequency because the reverberation time is estimated in the time domain. However, in this method, the reverberation time is estimated using the main frequency components in the modulation frequency domain. Not affected. This is why this method worked well.

(Embodiment 2)
The second embodiment is a reverberation time estimation device having a configuration particularly suitable for estimating reverberation time from human speech.

[Application of reverberation time estimation method to human speech signal]
6A to 6D are graphs showing a power envelope formed by using two sets of one period of a 10 Hz sine wave and a modulation spectrum thereof. The distance between the two sine wave sets of the power envelope shown in FIG. 6A is 0.1 seconds, similarly 0.2 seconds in FIG. 6B, 0.5 seconds in FIG. 6C, and 1.0 seconds in FIG. 6D. It is. Thus, as the flat interval between the two sets of sine waves of the power envelope is lengthened, the first peak exists at the modulation frequency f _mO Hz in the vicinity of 0 Hz on the modulation spectrum, and the power at that peak is It can be seen that the value E _x (f _mO ) (the portion indicated by ◯ in FIGS. 6A to _6D ) tends to approach the power value E _x (f _md ) at the main modulation frequency. Here, f _mO is the modulation frequency closest to 0 Hz except for 0 Hz. It can be _{seen that} such power at f _mO is not attenuated by reverberation with reference to FIG. Therefore, if the reverberation time is estimated using this f _mO as a “reference value” for power recovery of the main modulation frequency f _md , it is possible to accurately estimate the reverberation time for such a power envelope. It is considered to be.

FIG. 7 shows the distance (time interval) between two sets of sine waves as shown in FIGS. 6A to _6D, the power value E (f _mO ) at the reference frequency f _{mO, and} the power value E (at the main modulation frequency f _md ( It is the graph which showed the relationship with the difference with _fmd ). In the figure, the vertical axis represents E (f _mO ) −E (f _md ), and the horizontal axis represents the distance between two sets of sine waves. In FIG. 7, the frequency of the sine wave for one cycle used for the power envelope is set to 5, 10, and 20 Hz. From this figure, it can be seen that there is a case where E (f _mO ) = E (f _md ). Further, f _md appears as a modulation frequency corresponding to the time difference of the sine wave for one cycle. Therefore, when E (f _mO ) = E (f _md ) from this figure, f _md is suitable for comparison. It can be seen that the modulation frequency is low.

The reference value (power) on f _mO Hz is not completely attenuated by reverberation. The degree to which the reference value is affected by reverberation is an important issue. If the reference value is greatly attenuated due to reverberation, the estimated reverberation time value becomes shorter than the theoretical value.

8, MTF with a degree of modulation 1 in the reference value at a clean state _{(M (f m, T R} ): solid line) and, MTF with a degree of modulation 1 in the reference value attenuated by the reverberation (M ' _{(f m, T} R): the dashed line) is a graph showing a. As shown in the figure, when the reference value is attenuated, M (f _m , T _R ) and M ′ (f _m , T _R ) that coincide in the power E _y (f _md ) at the main modulation frequency f _md of the modulation spectrum Are different functions, the values of T _Rc and T _Rw are different when M (f _m , T _Rc ) = M ′ (f _m , T _Rw ).

The value of f _mO depends on the FFT length. FIG. 9A is a graph showing the relationship between the reference frequency f _mO and the MTF value for a plurality of reverberation times, and FIG. 9B shows the _{case where} the power E _y (f _mO ) of the reference frequency f _mO is used as a reference value. It is a graph which shows the relationship between the error of the reverberation time of and the modulation frequency. 9A is reverberation time _{T R} is for the case of each 0.1,0.3,0.5,1.0,2.0 _seconds, f mO = 0.025,0.05,0.1Hz The above MTF values are shown. When the sampling frequency f _s is 20 kHz, this figure shows the attenuation of the reference value when the FFT length is 30, 20, and 10 seconds in order to increase the frequency resolution. FIG. 9B shows the relationship between the reverberation time error and the modulation frequency for each case of the reference frequency f _mO = 0.025, _{0.05, and} 0.1 seconds. From this figure, it can be seen that the shorter the FFT length, the longer the reverberation time, and the lower the modulation frequency, the greater the error. However, when f _mO = 0.025 and the reference value is set above _0.5 Hz, the error in the reverberation time is at most even in the case of T _R = 2.0 seconds where the error becomes the largest in the low modulation frequency region. ^10-3 digits. That is, if the FFT length is about 20 seconds, the estimated value of the reverberation time is considered to have almost no influence of errors even if the power at the modulation frequency component closest to 0 Hz is excluded except 0 Hz.

  On the other hand, high frequency energy concentrates on the rise and fall of human voice, and therefore, the power envelope of the band-divided voice often has a shape with a long flat section between peaks as shown in FIG. 6D. Seen. FIG. 10A and FIG. 10B show a power envelope graph and a modulation spectrum graph of human speech when the band is divided. FIG. 10A shows a case of a sound source signal to which reverberation is not added, and FIG. 10B shows a case of a reverberation signal to which reverberation is added. In the modulation spectrum of the sound source signal, the power of the modulation frequency very close to 0 Hz and the power of the main component are equal. Therefore, it is considered that the reverberation time can be estimated by performing band division on the audio signal.

[Configuration of reverberation time estimation device]
Based on the principle as described above, in this embodiment, an acoustic signal (audio signal) is divided into bands, and an estimated value of the reverberation time is obtained based on the power envelope of each divided band. A channel suitable for processing is selected on the basis of the power envelope, and the average value of the estimated reverberation time of the selected channel is adopted as the final estimated value of the reverberation time. In addition, the reverberation time estimation apparatus according to the present embodiment is configured to generate a modulation spectrum normalized by the reference value using a modulation spectrum value at a modulation frequency closest to 0 Hz as a reference value except 0 Hz. Yes.

  FIG. 11 is a block diagram showing a configuration of a reverberation time estimation apparatus according to Embodiment 2 of the present invention. As shown in FIG. 11, reverberation time estimation apparatus 201 according to the present embodiment includes functional blocks of power envelope generation unit 41, low-pass filter 42, main modulation frequency acquisition unit 43, and normalized modulation spectrum generation unit 48. In addition, a digital signal processing circuit 204 having functional blocks of a band dividing unit 45 and a channel selecting unit 46 is provided.

  The band dividing unit 45 includes a constant band pass filter bank (not shown) having a bandwidth of 100 Hz, and is configured to divide the digital acoustic signal output from the A / D converter 3 into 100 channels. ing. Each channel signal band-divided by the band dividing unit 45 is supplied to the power envelope generating unit 41, and a power envelope signal of each channel signal is generated.

The power envelope signal output from the power envelope generation unit 41 is supplied to the low pass filter 42 and the normalized modulation spectrum generation unit 48 and also to the channel selection unit 46. The normalized modulation spectrum generation unit 48 of the present embodiment _uses the power value at the reference frequency f _mO that is the modulation frequency closest to 0 Hz except 0 Hz as a reference value, and the modulated spectrum signal normalized by this reference value It is configured to output.

  The channel selection unit 46 includes a high level part detection unit 46a, a first selection unit 46b, a second selection unit 46c, and a third selection unit 46d. A channel used for reverberation time estimation is selected.

  12A to 12C are schematic diagrams illustrating a channel selection process used for reverberation time estimation by the channel selection unit 46. As shown in FIG. 12A, first, the high level portion detector 46a causes the rising time point b (n) and the falling time point e () of the high level portion of the power envelope signal based on a predetermined reference level (indicated by a broken line in the figure). n) are detected. Next, in the section from the rising time point b (n) to the falling time point e (n), the time point at which the power is maximum is detected and set as the peak time point p (n). The detected rising time point b (n), peak time point p (n), and falling time point e (n) are taken as one set. The high level portion detection unit 46a is configured to detect a set of all high level portions included in the power envelope signal in this way.

  As shown in FIG. 12B, the first selection unit 46b determines whether or not a minute peak exists between the rising point b (n) and the falling point e (n-1) of the previous set. Is configured to determine. When a minute peak exists, the set is excluded from the target used for reverberation time estimation.

  As illustrated in FIG. 12B, the second selection unit 46c is configured to determine whether or not a large valley is included in the set when such a minute peak does not exist. The second selection unit 46c includes a low-pass filter (not shown), and the valley is detected by the low-pass filter. A set having such a valley is excluded from the target used for reverberation time estimation.

  As described above, the third selection unit 46d, as shown in FIG. 12C, when there are two consecutive sets in one power envelope in which no minute peak exists and no valley exists. In addition, the power difference and time difference between the two peaks are obtained, it is determined whether or not these are greater than or equal to a predetermined reference value, and if both the power difference and time difference are greater than or equal to the reference value, Select the channel to be used for reverberation time estimation. The selection result obtained by the channel selection unit 46 in this way is given to the arithmetic circuit 205.

  The arithmetic circuit 205 receives the output data from the main modulation frequency acquisition unit 43 and the normalized modulation spectrum generation unit 48 and calculates the estimated value of the reverberation time in each channel in the same manner as in the first embodiment. The arithmetic circuit 205 receives the output data from the channel selection unit 46, calculates the average value of the reverberation time estimation results of the target channel used for reverberation time estimation, and uses this average value as the reverberation time estimation value. The estimated reverberation time is displayed on the liquid crystal display unit 7 by the arithmetic circuit 205 driving and controlling the liquid crystal display unit 7.

  In addition, since the other structure of the reverberation time estimation apparatus 201 which concerns on this Embodiment is the same as that of the reverberation time estimation apparatus 1 which concerns on Embodiment 1, it attaches | subjects the same code | symbol about the same component. The description is omitted.

[Evaluation experiment]
Evaluation test of reverberation time estimation method (hereinafter referred to as “this method”) that is the same as the reverberation time estimation performed by the reverberation time estimation apparatus according to the present embodiment using eight sentences uttered by a female speaker as an acoustic signal for evaluation. Carried out. The utterance content of the evaluation acoustic signal is as follows.
(1) “If you wish to register to participate in the 1st International Interpretation Conference, please specify the address, name and presentation / listening on the designated application form and apply to the International Conference Secretariat. . "
(2) “Yes. This is the first Interpreting Telephone International Conference Secretariat.”
(3) “Hello. I would like to apply for an international interpreting conference. What procedures should I take?”
(4) “To participate in an international interpreting telephone conference, you must register using the designated application form.”
(5) "How much does it cost to just attend rather than present at the meeting?"
(6) “If you wish to make a presentation, the participation fee including preliminary collection fee and registration fee is 40,000 yen.”
(7) “In the case of attendance only, it is possible to accept it on the day, and the cost including the proceedings collection cost 35,000 yen.”
(8) “How do I get an application form for registration?”
Reverberation time _{T R} utilized was 0.1,0.3,0.5,1.0, and was five 2.0 seconds. 50 types of impulse responses were prepared for each reverberation time of Equation (7). The total number of stimuli is 2000 (8 × 5 conditions × 50 carriers). In this experiment, for each of the present method and the conventional method, an average value of estimated reverberation times obtained for all utterances was obtained, and the average value was evaluated.

FIG. 13 is a graph showing the results of the evaluation experiment. The drawing method is the same as in FIG. In this method, the reverberation time _{T R} is 0.1, 0.3, 0.5, and when 1.0 seconds, taken generally give estimation result close to the ideal value. Although reverberation time T _R is also the time of 2.0 seconds, which is slightly overestimated, it can be seen that good results.

(Embodiment 3)
Embodiment 3 is a reverberation time estimation device realized by a computer executing a computer program used for estimating reverberation time. Note that reverberation time estimation apparatus 301 according to the present embodiment implements substantially the same processing as software by reverberation time estimation apparatus 201 according to Embodiment 2.

[Configuration of reverberation time estimation device]
FIG. 14 is a block diagram showing a configuration of a reverberation time estimation apparatus according to Embodiment 3 of the present invention. As illustrated in FIG. 14, the computer 301 a includes a main body 311, an image display unit 312, and an input unit 313. The main body 311 includes a CPU 311a, a ROM 311b, a RAM 311c, a hard disk 311d, a reading device 311e, an input / output interface 311f, and an image output interface 311g. The image output interface 311g is connected by a bus 311i.

  The CPU 311a can execute a computer program loaded in the RAM 311c. Then, when the CPU 311 a executes a reverberation time estimation program 314 a described later, the computer 301 a functions as the reverberation time estimation device 301.

  The ROM 311b is configured by a mask ROM, PROM, EPROM, EEPROM, or the like, in which computer programs executed by the CPU 311a, data used for the same, and the like are recorded.

  The RAM 311c is configured by SRAM, DRAM, or the like. The RAM 311c is used for reading the reverberation time estimation program 314a recorded in the hard disk 311d. Further, when the CPU 311a executes a computer program, it is used as a work area of the CPU 311a.

  The hard disk 311d is installed with various computer programs to be executed by the CPU 311a, such as an operating system and application programs, and data used for executing the computer programs. A reverberation time estimation program 314a described later is also installed in the hard disk 311d.

  The hard disk 311d stores MTF inverse characteristics having reverberation time as a parameter as function data that can be processed by the CPU 311a. The function data of the inverse filter (inverse characteristic) stored in the hard disk 311d is prepared for a plurality of reverberation times given in advance as estimation candidates. The details of the inverse filter function data are the same as in the first embodiment.

  The reading device 311e is configured by a flexible disk drive, a CD-ROM drive, a DVD-ROM drive, or the like, and can read a computer program or data recorded on a portable recording medium 314. The portable recording medium 314 stores a reverberation time estimation program 314a for causing the computer to function as a reverberation time estimation device, and the computer 301a reads the reverberation time estimation program 314a from the portable recording medium 314. The reverberation time estimation program 314a can be installed in the hard disk 311d.

  The reverberation time estimation program 314a is provided not only by the portable recording medium 314 but also from an external device that is communicably connected to the computer 301a via an electric communication line (whether wired or wireless). It can also be provided through a communication line. For example, the reverberation time estimation program 314a may be stored in a hard disk of a server computer on the Internet, and the computer 301a may access the server computer to download the computer program and install it on the hard disk 311d. Is possible.

The hard disk 311d is installed with a multitask operating system such as Windows (registered trademark) manufactured and sold by Microsoft Corporation.
In the following description, the reverberation time estimation program 314a according to the first embodiment is assumed to operate on the operating system.

  The input / output interface 311f is, for example, a serial interface such as USB, IEEE 1394, or RS-232C, a parallel interface such as SCSI, IDE, or IEEE 1284, an analog interface including a D / A converter, an A / D converter, or the like. It is configured. An input unit 313 having a keyboard and a mouse is connected to the input / output interface 311f, and the user can input data to the computer 301a by using the input unit 313. In addition, the input unit 313 is provided with a microphone, and is configured to convert sound into an electric signal, A / D convert this by an input / output interface 311f, and acquire a digital signal. .

  The image output interface 311g is connected to an image display unit 312 configured by an LCD, a CRT, or the like, and outputs a video signal corresponding to image data given from the CPU 311a to the image display unit 312. The image display unit 312 displays an image (screen) according to the input video signal.

[Operation of reverberation time estimation device]
Next, the operation of the reverberation time estimation apparatus 301 will be described. FIG. 15 is a flowchart showing a flow of operations of reverberation time estimation apparatus 301 according to the present embodiment. First, an operator samples sound (sound) with a microphone of the input unit 313 in a room where reverberation time is measured. The analog sound signal output from the microphone is converted into digital sound data such as PCM by an A / D converter provided in the input / output interface 311f, and this sound data is given to the CPU 311a.

  First, the CPU 311a divides the acoustic data into a plurality of bands (step S1). At this time, the acoustic data is divided into channels for each 100 Hz bandwidth. Next, the CPU 311a generates a power envelope from the acoustic data of each channel, and stores each power envelope in the RAM 311c (step S2). The CPU 311a selects one channel (step S3), reads the power envelope of this channel from the RAM 311c, and removes high frequency components with a predetermined cutoff frequency (step S4). Then, the CPU 311a calculates an autocorrelation function for the power envelope from which the high frequency component has been removed, and calculates a main modulation frequency (step S5).

  Next, the CPU 311a reads again the power envelope of the channel from the RAM 311c, executes a Fourier transform process (step S6), uses the power at the modulation frequency closest to 0 Hz except for 0 Hz as a reference value, and is obtained in step S6. A normalized modulation spectrum obtained by normalizing the modulation spectrum with the reference value is calculated (step S7). Next, the CPU 311a estimates the reverberation time based on the main modulation frequency calculated in step S5 and the normalized modulation spectrum calculated in step S7 (step S8). The reverberation time is estimated as follows. First, the CPU 311a reads out the inverse filter of each reverberation time from the hard disk 311d, applies the inverse filter of each reverberation time to the main modulation frequency calculated in step S5 and the modulation spectrum calculated in step S7, and applies it to each reverberation time. The modulation spectrum at the main modulation frequency of the corresponding sound source signal is calculated. Next, the CPU 311a selects one of the modulation spectra at the main modulation frequency of the sound source signal corresponding to each reverberation time thus obtained, the magnitude closest to 0 dB (1 in terms of modulation degree). The reverberation time is estimated as the room reverberation time. The CPU 311a stores the estimated value of the reverberation time thus obtained in the RAM 311c in association with the channel that has been processed at that time.

  Next, the CPU 311a reads the power envelope of the channel again from the RAM 311c, and determines whether or not the channel is suitable for use in estimation of reverberation time based on the power envelope (step S9). This processing includes (1) detection of a high level portion (set of rising time, peak time, and falling time) of a power envelope, and (2) determination of whether or not a minute peak exists between adjacent high level portions. (3) Determination of whether or not a valley exists in a high level part, and when both a minute peak and a valley do not exist, (4) Power difference and time difference between adjacent high level parts are equal to or greater than a predetermined reference value It is comprised by each process of determination whether it is. Note that the details of these processes are the same as those described in the second embodiment, and a description thereof will be omitted.

  Next, the CPU 311a determines whether or not the processing in steps S4 to S9 has been performed on all channels (step S10), and there is a channel in which the processing in steps S4 to S9 has not been completed in step S10. If so (NO in step S10), one of the unprocessed channels is selected (step S11), and the process returns to step S4. When the processing of steps S3 to S8 has been completed for all channels in step S9 (YES in step S9), CPU 311a is determined to be suitable for use in estimation of reverberation time in step S9. An average value of the estimated reverberation time corresponding to the channel is calculated (step S12), the image output interface 311g is driven, and this average value is displayed on the image display unit 312 as an estimated value of the reverberation time for the sound signal. (Step S13), and the process ends.

(Other embodiments)
In the first embodiment, the normalized modulation spectrum generation unit 44 generates the modulation spectrum signal in which the size of the modulation spectrum is normalized by the modulation spectrum of the DC component (the modulation spectrum when the modulation frequency is 0 Hz). Although the case where it is configured has been described, the present invention is not limited to this. For example, even when the modulation spectrum is normalized by the power value at the modulation frequency closest to 0 Hz, excluding 0 Hz, Good.

  In the second and third embodiments, the normalized modulation spectrum generation unit generates a modulation spectrum normalized by the reference value using the modulation spectrum value at the modulation frequency closest to 0 Hz except 0 Hz as a reference value. However, the present invention is not limited to this, and it may be configured to generate a modulation spectrum signal in which the size of the modulation spectrum is normalized by the modulation spectrum of the DC component. Good.

  In the third embodiment, the acoustic data is band-divided into a plurality of channels, a reverberation time estimation value is calculated for each channel, and a channel suitable for use in reverberation time estimation is selected and selected. The configuration for obtaining the estimated value of the reverberation time for this sound data by averaging the estimated values of the reverberation time of the selected channels has been described. However, the present invention is not limited to this, and band division and channel selection are not performed. A configuration may be adopted in which one power envelope is calculated for the acoustic data, and an estimated value of the reverberation time is calculated based on the power envelope.

  In Embodiments 2 and 3 described above, the main modulation frequency is obtained and the normalized modulation spectrum is generated for the power envelopes of all the channels, whereby the reverberation time estimation values corresponding to the respective channels are individually obtained. And selecting a channel suitable for use in estimating the reverberation time from all channels, and calculating an average of the estimated reverberation time obtained from the power envelope of the channel suitable for use in estimating the reverberation time. However, the present invention is not limited to this, and a channel suitable for estimation of reverberation time is selected from the power envelopes of all channels, and an estimated value of reverberation time is obtained from the power envelope of the selected channel. The average value may be calculated. In this case, since the main modulation frequency is acquired and the normalized modulation spectrum is generated only for the power envelope of the selected channel, the processing can be performed efficiently.

  In the first to third embodiments, the MTF inverse filter having the reverberation time as a parameter is the function data that can be processed by the arithmetic circuit 5 or the CPU 311a as the memory 6 or the predetermined number of reverberation times. The modulation spectrum at the main modulation frequency of the sound source signal corresponding to each reverberation time is calculated by applying an inverse filter of each reverberation time to the acquired main modulation frequency and modulation spectrum stored in the RAM 311c. Of the modulation spectra at the main modulation frequency of the sound source signal corresponding to each reverberation time obtained in this way, one whose magnitude is closest to 0 dB is selected, and the reverberation time is estimated as the room reverberation time. The configuration. However, it is not limited to the above configuration. The above inverse filter is not stored as function data but as a lookup table in the memory 6 or the RAM 311c, and the arithmetic circuit 5 or the CPU 311a refers to the lookup table of the inverse filter and corresponds to the sound source signal corresponding to each reverberation time. The modulation spectrum at the main modulation frequency may be obtained. In each of the above embodiments, the reverberation time estimation device is one device, but the present invention is not limited to such a mode. For example, each unit included in the reverberation time estimation device is provided in a separate device, and the devices perform data communication via a communication network or the like, so that the processing in each of the above embodiments is realized. It may be a simple configuration.

  A reverberation time estimation apparatus and a reverberation time estimation method according to the present invention include a reverberation time estimation apparatus and a method for blindly estimating a reverberation time without using an original sound signal based on a frequency-sequence modulation spectrum obtained from a time-series acoustic signal. Useful as.

The block diagram which shows the structure of the reverberation time estimation apparatus which concerns on Embodiment 1 of this invention. The graph which shows MTF (modulation degree). The graph which shows the power envelope of the acoustic signal to which reverberation is not added. The graph which shows the modulation spectrum of the acoustic signal to which reverberation is not added. The graph which shows the power envelope of the acoustic signal at the time of adding reverberation to the acoustic shown in FIG. 3A. The graph which shows the modulation spectrum of the acoustic signal shown to FIG. 4A. 6 is a graph showing a result of an evaluation experiment of the same reverberation time estimation method as the reverberation time estimation performed by the reverberation time estimation apparatus according to Embodiment 1; A power envelope made by using two sets of one 10 Hz sine wave and a graph showing its modulation spectrum (time difference between sets 0.1 seconds). A power envelope made by using two sets of one 10 Hz sine wave and a graph showing its modulation spectrum (time difference between sets: 0.2 seconds). A power envelope created by using two sets of one 10 Hz sine wave and a graph showing its modulation spectrum (time difference between sets 0.5 seconds). A power envelope created by using two sets of one 10 Hz sine wave cycle and a graph showing its modulation spectrum (time difference between sets: 1.0 second). The graph which showed the relationship between the time interval between two sets of sine waves shown to FIG. 6A-D, and the difference of the power value in a reference frequency and a main modulation frequency. The graph which shows MTF which made the modulation degree in the reference value in a clean state 1, and MTF which made the modulation degree in the reference value attenuate | damped by reverberation 1. The graph which showed the relationship between a reference frequency and the value of MTF about several reverberation time. The graph which shows the relationship between the error of reverberation time, and a modulation frequency when the power of a reference frequency is made into a reference value. The power envelope graph and the modulation spectrum graph of human voice (without reverberation added) when the band is divided. The graph of the power envelope and modulation spectrum of the human voice (with reverberation added) when the band is divided. The block diagram which shows the structure of the reverberation time estimation apparatus which concerns on Embodiment 2 of this invention. The schematic diagram for demonstrating the selection process of the channel used for the reverberation time estimation by a channel selection part. The schematic diagram for demonstrating the selection process of the channel used for the reverberation time estimation by a channel selection part. The schematic diagram for demonstrating the selection process of the channel used for the reverberation time estimation by a channel selection part. The graph which shows the result of the evaluation experiment of the reverberation time estimation method same as the reverberation time estimation which the reverberation time estimation apparatus which concerns on Embodiment 2 implements. The block diagram which shows the structure of the reverberation time estimation apparatus which concerns on Embodiment 3 of this invention. 10 is a flowchart showing a flow of operations of the reverberation time estimation apparatus according to the third embodiment.

Claims (9)

A power envelope generating means for generating a time-series power envelope corresponding to the acoustic signal based on the time-series acoustic signal to which reverberation is added;
A modulation spectrum generating means for generating a modulation spectrum of a frequency sequence based on the power envelope generated by the power envelope generating means;
A reverberation time estimation device comprising: a reverberation time estimation unit that estimates a reverberation time corresponding to a transfer function related to a reverberation characteristic of a system in which the acoustic signal is observed based on the modulation spectrum generated by the modulation spectrum generation unit. A main modulation frequency specifying means for specifying a main modulation frequency indicating a modulation spectrum larger than a surrounding frequency region in the modulation spectrum of the frequency sequence;
When the reverberation time estimation means applies an inverse transfer function of the transfer function to the modulation spectrum of the frequency sequence, the modulation spectrum at the main modulation frequency after application indicates an original sound with no reverberation added. The reverberation time corresponding to the transfer function is estimated so as to approximately match the modulation spectrum at the main modulation frequency of the modulation spectrum of the frequency sequence corresponding to the original sound signal of the sequence. Reverberation time estimation device.   3. The main modulation frequency specifying unit is configured to obtain an autocorrelation function for the power envelope and to specify an inverse of a time shift amount at which the autocorrelation function exhibits a peak as the main modulation frequency. Reverberation time estimation device. A low pass filter applied to the power envelope generated by the power envelope generating means;
The reverberation time estimation apparatus according to claim 2 or 3, wherein the main modulation frequency specifying unit is configured to specify the main modulation frequency based on a power envelope output from the low-pass filter. Band dividing means for dividing the acoustic signal into a plurality of channels;
The reverberation time estimation apparatus according to any one of claims 1 to 4, further comprising: a channel determination unit that determines a channel to be used for reverberation time estimation from each channel divided by the band division unit. The power envelope generating means is configured to generate a power envelope for each channel band-divided by the band dividing means,
The power envelope generated by the power envelope generating means further comprises a high level part detecting means for detecting a high level part exceeding a predetermined reference value,
6. The reverberation time estimation device according to claim 5, wherein the channel determination unit is configured to determine a channel used for reverberation time estimation based on the high level portion detected by the high level portion detection unit.   The channel determining means determines whether or not a minute peak exists between the two high level parts detected by the high level part detecting means. If there is a minute peak, the channel determining means The reverberation time estimation device according to claim 6, wherein the reverberation time estimation device is configured to be excluded from a channel used for estimation.   The channel determination means determines whether or not a valley exists in the high level part detected by the high level part detection means, and if the valley exists, excludes the channel from the channel used for estimation. The reverberation time estimation apparatus according to claim 6 or 7, wherein the reverberation time estimation apparatus is configured to. Generating a time-series power envelope corresponding to the acoustic signal based on the time-series acoustic signal to which reverberation is added; and
Generating a modulation spectrum of a frequency sequence based on the generated power envelope;
Reverberation time estimation method comprising: estimating reverberation time corresponding to a transfer function related to reverberation characteristics of a system in which the acoustic signal is observed based on the generated modulation spectrum.

Images