JP5033156B2 - Sound image width estimation apparatus and sound image width estimation program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a sound image width estimating device capable of estimating sound image width with high precision based on the physical feature quantity obtained by analyzing digital acoustic signals constituted by two right and left channels. <P>SOLUTION: This sound image width estimating device 100 is provided with filter banks 7R, 7L for dividing a digital acoustic signal constituted of two right and left channels into a plurality of subband signals in frequency band having frequency band width of 1/6 octave or less per acoustic signal of right and left channels, a frequency band-based feature quantity calculating means 8<SB>f</SB>for calculating a frequency band-based feature quantity having at least one among cross-correlation degree between both ears, standard deviation in the direction of time axis of time difference between both ears or standard deviation in the direction of time axis of level difference between both ears as the feature quantity per subband signal, a means 9 for calculating a typical value of physical feature quantity from the frequency band-based feature quantity, and a sound image width-estimated value calcurating means 10 for calculating an estimated value of sound image width by applying the typical value of physical feature quantity to an estimating model equation. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a sound image width estimation device and a sound image width estimation program for estimating a sound image width, which is an auditory characteristic, based on physical feature values obtained by analyzing digital audio signals collected in two left and right channels.

The magnitude of the psychological effect of sound on humans can be quantified by subjective evaluation. Many attempts have been made to objectively evaluate the magnitude of the psychological effect obtained by this subjective evaluation based on physical features obtained by collecting and analyzing acoustic signals.
In this context, many studies have been conducted on the relationship between the sound image width and physical features, which is one of human auditory characteristics. One physical feature that is widely known and accepted in the field of acoustic analysis is IACC (interaural cross-correlation). In general, it is considered that the sound image width increases as IACC decreases, and many studies have been conducted on the analysis of IACC in various frequency bands (see, for example, Non-Patent Document 1).
In addition, a relationship between the sound image width and fluctuations of ITD (interaural time differences) and ILD (interaural level differences), which are physical features, has been reported (Non-patent Document 2). And Non-Patent Document 3).
Further, for example, in Patent Document 1, the apparent amplitude is determined based on _IACC , which is the maximum amplitude of IACF (interaural cross-correlation function) from an acoustic signal, and the width W _{IACC of the} maximum amplitude. A method for evaluating the width (ASW) of the sound source is described (see paragraph 0050).

JP 2003-57108 A

Masayuki Morimoto and Kazuhiro Iida, “Appropriate frequency bandwidth in measuring interaural cross-correlation as a physical measure of auditory source width”, Acourstical Science and Technology, Japan, Acoustical Society of Japan, 2005, Vol.26, No.2, p.179 -184 Russell Mason and Francis Rumsey, “A comparison of objective measurements for predicting selected subjective spatial attributes”, Audio Engineering Society 112th Convention Paper 5591, Germany, 2002 Jens Blauert and Werner Lindemann, “Auditory spaciousness: Some further psychoacoustic analyzes”, Journal of Acoustical Society of America, USA, 1986, Vol.80, No.2, p.533-542

  However, in the conventional objective evaluation method based on the physical feature value obtained by analyzing the acoustic signals collected from the left and right channels, the correlation between the physical feature value to be used and the subjective evaluation value is not necessarily high, and it can be generated from any sound source. It was not possible to accurately evaluate the sound image width.

  Accordingly, the present invention has been made in view of such problems, and an object of the present invention is to accurately estimate the sound image width based on physical feature values obtained by analyzing a digital acoustic signal composed of two channels on the left and right. It is to provide a sound image width estimation device.

  In order to achieve the above object, the sound image width estimation device according to claim 1 calculates a physical feature amount from a digital acoustic signal including two channels on the left and right sides, and calculates the calculated physical feature amount and physical feature amount and weight. A sound image width estimation device for estimating a sound image width by applying to a sound image width estimation model formula comprising coefficients, a frequency band dividing unit, a frequency-based feature amount calculating unit, a physical feature amount calculating unit, and an estimation And a value calculating means.

According to such a configuration, the sound image width estimation device uses a frequency band dividing unit to convert a digital audio signal having two channels on the left and right into a plurality of audio signals having a frequency bandwidth of 1/6 octave or less for each of the left and right channel acoustic signals. Divide into frequency band sub-band signals. Next, the sound image width estimation device uses the inter-aural cross-correlation degree and the inter-aural time difference time axis for each sub-band signal from the sub-band signal divided by the frequency band dividing unit by the frequency-band feature amount calculating unit. At least one of the standard deviation in the direction or the standard deviation in the time axis direction of the binaural level difference is calculated by the frequency band feature quantity representing the difference between the left and right channels of the subband signal. To do. Subsequently, in the sound image width estimation device, the physical feature amount calculating unit calculates the physical feature amount based on the frequency band-specific feature amount calculated by the frequency band-specific feature amount calculating unit. Then, the sound image width estimation device calculates an estimated value of the sound image width by applying the physical feature amount calculated by the physical feature amount calculation unit to the estimation model formula by the estimated value calculation unit.
Thus, the sound image width estimation device performs objective evaluation of the sound image width using the physical feature amount.

  The sound image width estimation device according to claim 2 is the sound image width estimation device according to claim 1, wherein the physical feature amount calculation means is an average, weighted average, maximum value, or median of feature amounts by frequency band. Any one of these is calculated as a physical feature amount.

According to such a configuration, the sound image width estimation device uses the physical feature amount calculating unit to calculate the average, weighted average, maximum value, or center of the feature amounts by frequency band for each subband signal calculated by the feature amount calculating unit by frequency band. Any one of the values is calculated as a physical feature amount. Then, the sound image width estimation device calculates an estimated value of the sound image width based on the physical feature amount calculated by the physical feature amount calculation unit by the sound image width estimation value calculation unit.
Thus, the sound image width estimation device calculates an estimated value of the sound image width based on the physical feature value obtained by collecting the feature values calculated for each frequency band into one value for each type of the feature value.

  The sound image width estimation apparatus according to claim 3 calculates a physical feature amount from a digital acoustic signal having two channels on the left and right, and uses the calculated physical feature amount as a sound image width estimation model including a physical feature amount and a weighting factor. A sound image width estimating apparatus that applies a formula to estimate a sound image width, and includes a frequency band dividing unit, a feature amount calculating unit for each frequency band, and an estimated value calculating unit.

According to such a configuration, the sound image width estimation device uses a frequency band dividing unit to convert a digital audio signal having two channels on the left and right into a plurality of audio signals having a frequency bandwidth of 1/6 octave or less for each of the left and right channel acoustic signals. Divide into frequency band sub-band signals. Next, the sound image width estimation device uses the inter-aural cross-correlation degree and the inter-aural time difference time axis for each sub-band signal from the sub-band signal divided by the frequency band dividing unit by the frequency-band feature amount calculating unit. At least one of the standard deviation in the direction or the standard deviation in the time axis direction of the binaural level difference is calculated by the frequency band feature quantity representing the difference between the left and right channels of the subband signal. To do. Then, the sound image width estimation apparatus applies the estimated value of the sound image width by applying to the estimation model equation, using the estimated value calculation means as an individual feature value for each frequency band calculated by the feature value calculation means for each frequency band as a physical feature value. calculate.
Thus, the sound image width estimation device performs objective evaluation of the sound image width using the physical feature amount calculated for each frequency band.

  The sound image width estimation apparatus according to claim 4 is the sound image width estimation apparatus according to claims 1 to 3, wherein the frequency band dividing unit divides the frequency band into subband signals having a frequency bandwidth of 1/12 octave or less. It was configured as follows.

According to such a configuration, the sound image width estimation device uses a frequency band dividing unit to convert a digital acoustic signal having two channels on the left and right into subbands having a frequency bandwidth of 1/12 octave or less for each of the left and right channel acoustic signals. Divide into signals. Subsequently, the sound image width estimation device calculates a feature value for each frequency band for each subband signal having a frequency bandwidth of 1/12 octave or less by the feature value calculation unit for each frequency band. In the sound image width estimation device, the physical feature quantity is calculated by the physical feature quantity calculation means based on the feature quantity by frequency band calculated by the feature quantity calculation means by frequency band. Alternatively, the sound image width estimation device uses the frequency band-specific feature values calculated by the frequency band-specific feature value calculation means as individual physical feature values. Then, the sound image width estimation device applies the physical feature amount calculated by the physical feature amount calculation unit or the feature amount by frequency band calculated by the feature amount calculation unit by frequency band to the estimation model equation by the estimated value calculation unit. An estimated value of the sound image width is calculated.
Thus, the sound image width estimation device calculates an estimated value of the sound image width based on the feature amount calculated for each frequency band finely divided into 1/12 octaves or less.

  The sound image width estimation device according to claim 5 is the sound image width estimation device according to any one of claims 1 to 4, and further includes a weight coefficient calculation unit.

According to such a configuration, the sound image width estimation apparatus uses the weighting coefficient calculation means to calculate the weighting coefficient in the estimation model formula as the interaural cross-correlation, the standard deviation in the time axis direction of the interaural time difference, or the interaural level difference. Physical feature based on at least one of the standard deviations in the time axis direction or interaural cross-correlation, standard deviation in the time axis direction of interaural time difference, or standard deviation in the time axis direction of interaural level difference Is calculated in advance by a regression analysis using at least one individual frequency band characteristic amount as an explanatory variable and the sound image width as an objective variable. Then, the sound image width estimation device is pre-calculated by the estimated value calculation means by the physical feature quantity calculated by the physical feature quantity calculation means or the feature quantity by frequency band calculated by the feature quantity calculation means by frequency band and the weight coefficient calculation means in advance. The estimated value of the sound image width is calculated by the estimation model formula using the weighting factor previously set.
Thereby, the sound image width estimation device calculates an estimated value of the sound image width according to the weighting coefficient determined by the regression analysis.

  The sound image width estimation program according to claim 6 calculates a physical feature amount from a digital acoustic signal composed of two channels on the left and right, and uses the calculated physical feature amount as a sound image width estimation model composed of a physical feature amount and a weighting factor. In order to estimate the sound image width by applying it to the equation, the computer is caused to function as a frequency band dividing unit, a characteristic amount calculating unit by frequency band, a physical feature amount calculating unit, and an estimated value calculating unit.

According to such a configuration, the sound image width estimation program uses a frequency band dividing unit to convert a digital audio signal having two left and right channels into a plurality of audio signals having a frequency bandwidth of 1/6 octave or less for each of the left and right channel acoustic signals. Divide into frequency band sub-band signals. Next, the sound image width estimation program calculates the interaural cross-correlation degree and the time axis of the interaural time difference for each subband signal from the subband signal divided by the frequency band dividing unit by the frequency band feature amount calculating unit. At least one of the standard deviation in the direction or the standard deviation in the time axis direction of the binaural level difference is calculated by the frequency band feature quantity representing the difference between the left and right channels of the subband signal. To do. Subsequently, in the sound image width estimation program, the physical feature quantity is calculated by the physical feature quantity calculation unit based on the frequency band feature quantity calculated by the frequency band feature quantity calculation unit. Then, the sound image width estimation program calculates the estimated value of the sound image width by applying the physical feature amount calculated by the physical feature amount calculating unit to the estimation model formula by the estimated value calculating unit.
Thereby, the sound image width estimation program performs objective evaluation of the sound image width using the physical feature amount.

  The sound image width estimation program according to claim 7 calculates a physical feature amount from a digital acoustic signal having two channels on the left and right, and uses the calculated physical feature amount as a sound image width estimation model including a physical feature amount and a weighting factor. In order to estimate the sound image width by applying it to the equation, the computer is caused to function as a frequency band dividing unit, a characteristic amount calculating unit for each frequency band, and an estimated value calculating unit.

According to such a configuration, the sound image width estimation program uses a frequency band dividing unit to convert a digital audio signal having two left and right channels into a plurality of audio signals having a frequency bandwidth of 1/6 octave or less for each of the left and right channel acoustic signals. Divide into frequency band sub-band signals. Next, the sound image width estimation program calculates the interaural cross-correlation degree and the time axis of the interaural time difference for each subband signal from the subband signal divided by the frequency band dividing unit by the frequency band feature amount calculating unit. At least one of the standard deviation in the direction or the standard deviation in the time axis direction of the binaural level difference is calculated by the frequency band feature quantity representing the difference between the left and right channels of the subband signal. To do. Then, the sound image width estimation program applies the estimated value of the sound image width by applying to the estimation model formula, using the estimated value calculation means as the physical feature quantity of each individual frequency band feature quantity calculated by the frequency band feature quantity calculation means. calculate.
Accordingly, the sound image width estimation program performs objective evaluation of the sound image width using the physical feature amount calculated for each frequency band.

According to the invention described in claim 1 or claim 6, since the estimated value of the sound image width is calculated based on the feature amount calculated for each frequency band having a frequency bandwidth of 1/6 octave or less, it can be performed with stable accuracy. The sound image width can be estimated.
According to the second aspect of the present invention, in order to calculate the estimated value of the sound image width based on the physical feature value obtained by collecting the feature value calculated for each frequency band into one value for each type of the feature value, The estimated value of the sound image width can be calculated by simple calculation without increasing the number of weighting coefficients in the estimation model formula.
According to the invention described in claim 3 or claim 7, since the estimated value of the sound image width is calculated based on the feature amount calculated for each frequency band having a frequency bandwidth of 1/6 octave or less, the sound image width is accurately determined. Can be estimated.
According to the fourth aspect of the present invention, since the estimated value of the sound image width is calculated based on the feature amount calculated for each frequency band whose frequency bandwidth is equal to or less than 1/12 octave, the sound image width of the sound image width can be more stable. Estimation can be performed.
According to the invention described in claim 5, since the weighting coefficient in the estimation model formula is determined by regression analysis using subjective evaluation data and physical feature values corresponding to the subjective evaluation data, the sound image width can be estimated with high accuracy. It can be carried out.

It is a block diagram which shows the structure of the sound image width estimation apparatus of 1st Embodiment which concerns on this invention. It is a block diagram which shows the structure of the calculating means in the sound image width estimation apparatus of 1st Embodiment which concerns on this invention. It is a flowchart which shows the flow of a process of the sound image width estimation apparatus of 1st Embodiment which concerns on this invention. It is a flowchart which shows the flow of the weighting coefficient calculation process of the estimation model type | formula in the sound image width estimation apparatus of 1st Embodiment which concerns on this invention. It is a graph which shows the result of the Pearson correlation analysis between the physical feature-value used with the estimation model formula of the sound image width in this invention, and the subjective evaluation data of a sound image width, (1), (2), (3) and ( 4) shows the results when using violin G-line, A-line, D-line, and E-line open strings as the sound source. It is a block diagram which shows the structure of the sound image width estimation apparatus of 2nd Embodiment which concerns on this invention. It is a block diagram which shows the structure of the calculating means in the sound image width estimation apparatus of 2nd Embodiment which concerns on this invention. It is a flowchart which shows the flow of a process of the sound image width estimation apparatus of 2nd Embodiment which concerns on this invention. It is a flowchart which shows the flow of the weighting coefficient calculation process for physical feature-value representative value calculation in the sound image width estimation apparatus of 2nd Embodiment which concerns on this invention. It is a block diagram which shows the structure of the sound image width estimation apparatus of 3rd Embodiment which concerns on this invention. It is a flowchart which shows the flow of a process of the sound image width estimation apparatus of 3rd Embodiment which concerns on this invention. It is a schematic diagram which shows the structural example of the experimental apparatus for extract | collecting the subjective evaluation data used in order to determine the weighting coefficient of the estimation model formula of the sound image width in this invention. It is a graph which shows the example of the sound image width estimated value computed by the sound image width estimation apparatus which concerns on this invention.

Embodiments of the present invention will be described below with reference to the drawings as appropriate.
[First Embodiment]
First, the configuration of the sound image width estimation apparatus 100 according to the first embodiment of the present invention will be described with reference to FIG. As shown in FIG. 1, the sound image width estimation apparatus 100 includes a dummy head 1, microphones 2L and 2R, low-pass filters 3L and 3R, AD converters 4L and 4R, an arithmetic unit 5, a display unit 14, It is configured with. Further, the computing means 5 includes memories 6L and 6R, filter banks (frequency band dividing means) 7L and 7R, physical characteristic quantity calculating means for each frequency band (feature quantity calculating means for each frequency band) 8 _f , and physical characteristic quantities. Representative value calculating means (physical feature quantity calculating means) 9, sound image width estimated value calculating means (estimated value calculating means) 10, weight coefficient storage means 11, estimated value weight coefficient calculating means (weight coefficient calculating means) 12, And subjective evaluation data storage means 13.

  The dummy head 1 is a simulated head for collecting sound generated from the sound source SS to be tested by a binaural method. Microphones 2L and 2R are attached to the entrances of the left and right ears of the dummy head 1, respectively.

The microphones 2L and 2R are sound collection means for collecting sound generated from the sound source SS at the entrance of the left and right ears of the dummy head 1, respectively. Analog sound signals collected by the microphones 2L and 2R are input to the low-pass filters 3L and 3R, respectively.
In the first embodiment, the microphones 2L and 2R are arranged at the entrance portions of the left and right ears of the dummy head 1, but the microphones 2L and 2R are arranged at the eardrum portion of the dummy head 1 to collect sound. May be.
Further, instead of the dummy head 1, the microphones 2L and 2R may be arranged using a sphere simulating a human head, or the microphones 2L and 2R may be arranged in the form of a two-point microphone using a microphone stand. It may be arranged.

The low-pass filters 3L and 3R receive analog audio signals collected by the microphones 2L and 2R, respectively, and digitally convert high frequency components exceeding 1/2 of the sampling frequency fs from the input analog audio signals by the AD converters 4L and 4R. This is an anti-aliasing filter for preventing the occurrence of aliasing distortion by limiting the bandwidth before sampling (sampling). The low-pass filters 3L and 3R output analog audio signals with band restrictions to the AD converters 4L and 4R, respectively.
Since the upper limit of human audible frequency is 20 kHz, the sampling frequency fs needs to be 40 kHz or more, which is twice 20 kHz. For example, when the sampling frequency is fs = 48 kHz, the frequency components exceeding fs / 2 = 24 kHz may be band-limited by the low-pass filters 3L and 3R.

  The AD converters 4L and 4R input analog acoustic signals band-limited by the low-pass filters 3L and 3R, respectively, sample the input analog acoustic signals at a sampling frequency fs = 48 kHz, for example, and convert them into digital signals. The AD converters 4L and 4R respectively calculate the left channel acoustic signal sl (n) and the right channel acoustic signal sr (n) (where n represents the number of the sampled data) converted into digital signals, respectively. Output to the memories 6L and 6R of the means 5.

The computing means 5 receives the acoustic signals sl (n) and sr (n) collected by the binaural method and digitized by the AD converters 4L and 4R, and the inputted acoustic signals sl (n) and sr (n) Is an analysis means for calculating a sound image width estimated value (hat y) by analyzing the numerical value by numerical calculation. The computing means 5 can be realized using a general-purpose computer.
The computing means 5 outputs the calculated sound image width estimated value (hat y) to the display means 14.
Details of the computing means 5 will be described later.

The display unit 14 is a display device such as a liquid crystal display that displays the estimated sound image width (hat y) input from the calculation unit 5 so as to be visible.
The display means 14 appropriately displays the numerical value of the estimated sound image width (hat y) output from the computing means 5 at predetermined time intervals. The display means 14 may be displayed in a graph so that the temporal change of the estimated sound image width (hat y) can be easily grasped.

  The sound source SS is sound generation means for generating sound that induces a sound image width in humans. As the sound source SS to be tested, an arbitrary sound source such as a musical instrument or a speaker can be used, and the number of sound sources SS may be one or plural.

Next, the structure of each part of the calculating means 5 is demonstrated.
The memories 6L and 6R are storage means for storing the left channel acoustic signal sl (n) and the right channel acoustic signal sr (n) input from the AD converters 4L and 4R, respectively. The acoustic signals sl (n) and sr (n) stored in the memories 6L and 6R are read by the filter banks 7L and 7R as appropriate.

The filter banks (frequency band dividing means) 7L and 7R are each composed of a band-pass filter group having a characteristic of passing a plurality of frequency bands f different from each other. Here, f indicates a number for identifying a frequency band, and f = 1, 2,... F is an integer of 2 or more.
The filter banks 7L and 7R read the left channel acoustic signal sl (n) and the right channel acoustic signal sr (n) stored in the memories 6L and 6R, respectively, and read the acoustic signals sl (n) and sr (n The frequency band components sl (n, f) and sr (n, f) of a plurality of frequency bands f are obtained as a set of outputs of each bandpass filter. That is, the filter banks 7L and 7R are frequency band dividing means for dividing the acoustic signals sl (n) and sr (n) into a plurality of frequency band components sl (n, f) and sr (n, f). The filter banks 7L and 7R convert the frequency band components sl (n, f) and sr (n, f) of the acoustic signals sl (n) and sr (n) into frequency band physicals corresponding to the respective frequency bands f. It outputs to the feature-value calculation means _8f .

The filter banks 7L and 7R can be configured with a group of equal ratio band filters such as a 1/6 octave band filter, for example. Preferably, a narrow band 1 / n octave band filter (where n is an integer of 1 or more) having a frequency bandwidth of 1/6 octave or less, more preferably 1/12 octave or less can be used.
In addition, each filter which comprises a filter group can be comprised with a FIR (finite impulse response) filter.

The frequency characteristic-specific physical feature quantity calculation means (frequency-band characteristic quantity calculation means) 8 _f (f = 1, 2,..., F) are respectively left and right corresponding to the frequency band f of the acoustic signal from the filter banks 7L and 7R. The frequency band components sl (n, f) and sr (n, f) are input, and the input left and right frequency band components sl (n, f) and sr (n, f) are analyzed, and each frequency band f is analyzed. The physical feature quantity representative value calculating means 9 calculates the physical feature quantity by frequency band (feature quantity by frequency band) x _a (f), x _t (f), x _l (f) as three types of physical feature quantities. Output to.

The physical feature quantity representative value calculation means (physical feature quantity calculation means) 9 includes F sets of physical features by frequency band from F frequency band physical feature quantity calculation means 8 _f (f = 1, 2,..., F). Quantities x _a (f), x _t (f), and x _l (f) are input, and the input physical characteristic amounts x _a (f), x _t (f), and x _l (f) by frequency band are physical characteristics. For each type of quantity, the physical feature quantity representative values X _a , X _t , and X _l are calculated and output to the sound image width estimated value calculating means 10 or the estimated value weight coefficient calculating means 12.
When calculating the weighting factors C _a , C _t, and C _l for each physical feature amount representative value X _a , X _t , X _l in the estimation model formula for estimating the sound image width, the physical feature amount representative value calculation is performed. The means 9 outputs the physical feature quantity representative values X _a , X _t , and X _l to the estimated value weight coefficient calculating means 12. When estimating the sound image width using the estimation model formula and the weighting coefficients C _a , C _t, and C _l , the physical feature quantity representative value calculating unit 9 uses the physical feature quantity representative values X _a , X _t , X ₁ is output to the sound image width estimated value calculating means 10.

Here, with reference to FIG. 2 (refer to FIG. 1 as appropriate), the detailed configuration of the frequency-specific physical feature quantity calculating means 8 _f and the physical feature quantity representative value calculating means 9 will be described.
As shown in FIG. 2, the physical characteristic amount calculating means 8 _f for each frequency band includes windowing means 20L _f and 20R _f , CCC (interaural cross-correlation coefficient) calculating means 21 _f , , Level calculating means 22L _f and 22R _f , IACC calculating means 23 _f , ITD calculating means 24 _f , ILD calculating means 25 _f , ILD standard deviation calculating means 26 _f , IACC average calculating means 27 _f , ITD Standard deviation calculating means 28 _f .
The physical feature quantity representative value calculating means 9 includes an ILD standard deviation representative value calculating means 30, an IACC average representative value calculating means 31, and an ITD standard deviation representative value calculating means 32.

The windowing means 20L _f and 20R _f receive the frequency band components sl (n, f) and sr (n, f) of the corresponding frequency band f from the filter banks 7L and 7R, respectively, and the input frequency band components sl ( n, f) and sr (n, f) are multiplied by a time window w (n) to sequentially extract a signal having a predetermined time length.
The windowing means 20L _f and 20R _f output the cut signal sequences yl _k (n, f) and yr _k (n, f) to the level calculation means 22L _f and 22R _f , respectively, and the signals of the left and right channels. The columns yl _k (n, f) and yr _k (n, f) are output to the CCC calculating means 21 _f .

Here, assuming that the number of data of signals cut out from the frequency band components sl (n, f) and sr (n, f) by the windowing means 20L _f and 20R _f is N (N is an integer of 1 or more), the time window w (n) can be expressed by equation (1).

In addition, the time length cut out by the time window w (n) can be set to, for example, 10 (ms) to 100 (ms).
Here, assuming that the time length is t (ms) and the sampling frequency in the AD converters 4L and 4R is fs (Hz), the number N of data of the extracted signals is N = t10 ⁻³ fs.

Further, the windowing means 20L _f and 20R _f have a movement width d (d is d) with respect to the frequency band components sl (n, f) and sr (n, f), respectively, in the time domain by the time window w (n). A signal sequence is cut out by shifting while shifting by an integer of 1 or more. The signal sequences yl _k (n, f) and yr _k (n, f) cut out k-th from the frequency band components sl (n, f) and sr (n, f) of the left channel and the right channel are respectively expressed by the formulas ( 2-1) and formula (2-2).

Here, when the time difference between both ears is τ (ms) and the sampling frequency is fs (Hz), the movement width d can be d ≧ τ10 ⁻³ fs. That is, the position to be cut out by the time window w (n) can be shifted by the number of data corresponding to the interaural time difference τ or more. As a result, the movement analysis with the time width corresponding to the movement width d as the time resolution can be performed by each analysis means such as the CCC calculation means 21 _f and the level calculation means 22L _f and 22R _{f in the} subsequent stage.

The level calculation means 22L _f and 22R _f input the signal sequences yl _k (n, f) and yr _k (n, f) from the windowing means 20L _f and 20R _{f of the} corresponding frequency band f, respectively, and input k The acoustic energy levels (hereinafter referred to as levels) slE _k (f) and srE _k (f) in the second signal sequence yl _k (n, f) and yr _k (n, f) And the equation (3-2), and outputs to the ILD calculating means 25 _f of the corresponding frequency band f.

The CCC (interaural cross-correlation coefficient) calculation means 21 _f respectively outputs signal sequences yl _k (n, f) and yr _k from the windowing means 20L _f and 20R _{f of the} corresponding frequency band f. (N, f) is input, and the interaural cross-correlation coefficient CCC _k (τ, f) in the input k-th signal sequence yl _k (n, f) and yr _k (n, f) is expressed by the formula ( 4) and output to the IACC calculation means 23 _f and ITD calculation means 24 _f of the corresponding frequency band f.

IACC (absolute maximum value of the interaural cross-correlation coefficient; interaural cross correlation) calculating unit 23 _f, the corresponding frequency band f of the CCC calculating unit 21 between both ears from _f correlation coefficient CCC _k (tau, f) is input, and the interaural cross-correlation degree IACC _k (f), which is the maximum amplitude in the input binaural cross-correlation coefficient CCC _k (τ, f), is calculated by the equation (5-1). , And output to the IACC average calculating means 27 _f of the corresponding frequency band f.
Note that the IACC calculation unit 23 _f may calculate the interaural cross-correlation degree IACC _k (f) by the equation (5-2) instead of the equation (5-1).

The inter-ural time difference (ITD) calculating unit 24 _f receives the interaural cross-correlation coefficient CCC _k (τ, f) from the CCC calculating unit 21 _f of the corresponding frequency band f, and the equation (6) -1), the time difference τ giving the maximum amplitude in the input interaural cross-correlation coefficient CCC _k (τ, f) is calculated, and the calculated time difference τ is used as the interaural time difference ITD _k (f). It outputs to the ITD standard deviation calculation means 28 _f of the frequency band f.
The ITD calculation unit 24 _f may calculate the interaural time difference ITD _k (f) by the equation (6-2) instead of the equation (6-1).

An ILD (interaural level difference) calculation means 25 _f receives the levels slE _k (f) and srE _k (f) from the level calculation means 22L _f and 22R _f of the corresponding frequency band f, From the input levels slE _k (f) and srE _k (f), the interaural level difference ILD _k (f) is calculated by the equation (7), and the calculated interaural level difference ILD _k (f) Output to the ILD standard deviation calculating means 26 _f of the corresponding frequency band f.

IACC average calculating unit 27 _f, the corresponding type the interaural cross-correlation IACC _k (f) from IACC calculating means 23 _f of the frequency band f, all the sections cut out by windowing means 20L _f and 20R _f k (k = 1,2, ..., T) by entering the interaural cross-correlation IACC _k (f) in, by equation (8), the interaural cross-correlation IACC _k in the time axis direction (f) The average is calculated, and the calculated average is output to the IACC average representative value calculation means 31 as x _a (f), which is one of the physical features for each frequency band.

The physical feature amount x _a (f) for each frequency band is a simple average of the interaural cross-correlation degree IACC _k (f) calculated for each movement width d, but is not limited thereto. A weighted average may be used, or a maximum value or a median value may be used.

ITD standard deviation calculating means 28 _f, the corresponding interaural time difference from ITD calculation means 24 _f of the frequency band f enter the _ITD k (f), all the sections k cut out by windowing means 20L _f and 20R _f When the interaural time difference ITD _k (f) at (k = 1, 2,..., T) is input, the standard deviation of the interaural time difference ITD _k (f) in the time axis direction is calculated by Equation (9). The calculated standard deviation is output to the ITD standard deviation representative value calculation means 32 as x _t (f), which is one of the physical features for each frequency band.

ILD standard deviation calculation means 26 _f, the corresponding ILD calculation means enter the interaural level difference _ILD k (f) from 25 _f of the frequency band f, 20L _f and all the sections k (k cut out by 20R _f = 1, 2,..., T), the interaural level difference ILD _k (f) is input, and the standard deviation of the interaural level difference ILD _k (f) in the time axis direction is calculated by the equation (10). The calculated standard deviation is output to the ILD standard deviation representative value calculating means 30 as x _l (f), which is one of the physical features for each frequency band.

The IACC average representative value calculating means 31, the ITD standard deviation representative value calculating means 32, and the ILD standard deviation representative value calculating means 30 are respectively an IACC average calculating means 27 _f , an ITD standard deviation calculating means 28 _f, and an ILD standard deviation calculating means 26 _f. Frequency-specific physical feature values x _a (f), x _t (f) and x _l (f) calculated for each frequency band f from the input frequency feature physical parameters x _a (f) , X _t (f) and x _l (f), which are representative values of physical feature values X _a , X _t and X _l are calculated. The IACC average representative value calculating means 31, the ITD standard deviation representative value calculating means 32, and the ILD standard deviation representative value calculating means 30 respectively calculate the calculated physical feature quantity representative values X _a , X _t, and X _l as sound image width estimated value calculating means. 10 or the estimated value weighting coefficient calculation means 12.

As described above, when calculating the weighting factors C _a , C _t, and C _l for each physical feature amount representative value X _a , X _t, and X _l in the estimation model formula for estimating the sound image width, the physical feature amount The IACC average representative value calculating means 31, the ITD standard deviation representative value calculating means 32, and the ILD standard deviation representative value calculating means 30, which are constituent elements of the representative value calculating means 9, are respectively the physical feature quantity representative values X _a , X _t, and X ₁ is output to the estimated value weighting coefficient calculating means 12. Further, when the sound image width is estimated using the estimation model formula and the weight coefficients C _a , C _t, and C _l , the IACC average representative value calculating unit 31, the ITD standard deviation representative value calculating unit 32, and the ILD standard deviation representative value The calculating means 30 outputs the physical feature quantity representative values X _a , X _t and X _l to the sound image width estimated value calculating means 10, respectively.

Here, the IACC average representative value calculating means 31, the ITD standard deviation representative value calculating means 32, and the ILD standard deviation representative value calculating means 30 in the first embodiment are respectively physical characteristic amounts x _a (f) and x _{t for} each frequency band. As representative values of (f) and x _l (f), the physical feature value x for each frequency band calculated for each frequency band f by Expression (11-1), Expression (11-2), and Expression (11-3). The average of _a (f), x _t (f), and x _l (f) is calculated to be the physical feature quantity representative values X _a , X _t, and X _l .

  In this way, as the physical feature quantity used in the estimation model formula, the estimation model formula is obtained by using the representative value obtained by consolidating the physical feature quantity by frequency band calculated for each frequency band into one value for each type of physical feature quantity. The number of weighting coefficients can be reduced, and the calculation of the estimated sound image width (hat y) and the collection of subjective evaluation data for determining the weighting coefficient can be simplified.

Returning to FIG. 1 (see FIG. 2 as appropriate), the description of the configuration of the sound image width estimation apparatus 100 will be continued.
The sound image width estimated value calculating means (estimated value calculating means) 10 includes physical feature quantity representative values X _a from the IACC average representative value calculating means 31, the ITD standard deviation representative value calculating means 32, and the ILD standard deviation representative value calculating means 30, respectively. , X _t and X _l , and the weight coefficients C _a , C _t, and C _l calculated and stored in advance by the estimated value weight coefficient calculation means 12 are read from the weight coefficient storage means 11 and the equation ( The estimated value (hat y) of the sound image width is calculated by the estimated model formula shown in 12), and the calculated estimated value (hat y) is output to the display means 14.

As shown in Expression (12), the estimated sound image width (hat y) in the first embodiment is an absolute value of a three-dimensional vector having three physical feature quantity representative values X _a , X _t, and X _l as elements. Can be calculated as

The weighting factor storage unit 11 is a storage unit that stores the weighting factors C _a , C _t, and C _l of the estimation model formula shown in Formula (12) calculated by the estimated value weighting factor calculation unit 12. The weight coefficients C _a , C _t, and C _l stored in the weight coefficient storage unit 11 are read out by the sound image width estimated value calculation unit 10 when the sound image width is estimated, and the sound image width estimated value (hat y) is calculated. Used for calculation.

The estimated value weight coefficient calculating means 12 reads subjective evaluation data y _i stored in the subjective evaluation data storage means 13 in advance, and also includes an IACC average representative value calculating means 31, an ITD standard deviation representative value calculating means 32, and an ILD standard. The physical feature quantity representative values X _ai , X _ti, and X _li respectively corresponding to the subjective evaluation data y _i are input from the deviation representative value calculation means 30, and the input subjective evaluation data y _i and physical feature quantity representative value X _ai are input. , X _ti and X _li are used to calculate the weighting factors C _a , C _t and C _l of the estimated model equation shown in Equation (12) by the least square method which is a regression analysis method. Then, the calculated weighting factors C _a , C _t and C _l are stored in the weighting factor storage means 11. Note that i is a number for identifying individual subjective evaluation data.

Here, the physical characteristic amount representative value X _ai corresponding to subjective assessment data y _i, and the X _ti and X _li, the same sound field conditions and subject when give the subjective assessment data y _i, the dummy head 1 The acoustic signals are collected using the attached microphones 2L and 2R, and finally the IACC average representative value calculating means 31, the ITD standard deviation representative value calculating means 32, and the ILD standard deviation representative value are calculated using the respective analysis means described above. These are the physical feature quantity representative values X _a , X _t and X ₁ output from the means 30.

Next, _a method for calculating the weighting factors C _a , C _t, and C _l by the estimated value weighting factor calculation unit 12 in the first embodiment will be described.
In the first embodiment, in the estimation model formula shown in Formula (12) in which three physical feature quantity representative values X _a , X _t and X _l are explanatory variables and the sound image width y is an objective variable, a regression analysis method is used. The weighting coefficients C _a , C _t and C _l are calculated by the least square method. That is, a set of the subjective evaluation data y _i of the sound image width and the predicted value (hat y _i ) calculated by the estimation model formula is prepared in advance, and the weight coefficients C _a , C _t, and C _l are calculated by the least square method. calculate.

As shown in the equation (13), the weighting factors C _a , C _t and C that minimize the sum of squares J of the difference between the subjective evaluation data y _i and the predicted value (hat y _i ) calculated by the estimation model equation. _l is calculated. In Equation (13), S is the number of subjective evaluation data.

Here, for simplification of calculation, if the objective variable is the square of the sound image width y _i for convenience, the weight coefficients C _a , C _t, and C _l that minimize the sum of squares J shown in Expression (14) are set. Will be calculated.

Substituting equation (12) into the estimated value (hat y _i ) of equation (14), the sum of squares J can be expressed as equation (15).

Here, the condition for minimizing the square sum J is that the partial differentiation of the square sum J shown in Expression (16) by the weighting factors C _a , C _t, and C _l is zero.

  As a result, the simultaneous equations shown in Expression (17) are obtained.

Here, the weighting factor _C a shown in equation (17), in simultaneous equations for the variables _{C t} and _{C l,} the coefficients for weighting coefficient _C a, _{C t} and _{C l} as in equation (18) _{a 11} ˜a ₃₃ and b ₁ ˜b ₃ .

With _a 11 _{~a 33} and _b 1 ~b ₃ defined in formula (18), equation (17) can be expressed by the equation (19-1). The equation (19-1) can be transformed as the equation (19-2).

Here, a _{11 to} a ₃₃ and b _{1 to} b ₃ were collected at the position of the subject when the subjective evaluation data y _i and the subjective evaluation data y _i were obtained, as shown in the equation (17). It can be calculated using the physical feature value representative values X _ai , X _ti and X _li obtained by analyzing the acoustic signal.

The estimated value weighting factor calculation means 12 can calculate the weighting factors C _a , C _t and C _l by the procedure described above.

The subjective evaluation data storage means 13 is a storage means for storing the subjective evaluation data y _i of the sound image width for calculating the weighting coefficients C _a , C _t and C _l . The subjective evaluation data y _i stored in the subjective evaluation data storage means 13 is read by the estimated value weight coefficient calculating means 12 and used for calculating the weight coefficients C _a , C _t and C _l .

  The configuration of the sound image width estimation apparatus 100 has been described above, but the present invention is not limited to this. For example, the calculation means 5 of the sound image width estimation apparatus 100 can be realized by causing a general computer to execute a program and operating a calculation device or a storage device in the computer. This program (sound image width estimation program) can be distributed via a communication line, or can be distributed by writing on a recording medium such as a CD-ROM.

Next, the operation of the sound image width estimation apparatus 100 will be described with reference to FIG. 3 (see FIGS. 1 and 2 as appropriate).
As shown in FIG. 3, the sound image width estimation apparatus 100 first calculates weight coefficients C _a , C _t, and C _l in the estimation model formula shown in Formula (12) by the estimated value weight coefficient calculation means 12. And stored in the weight coefficient storage means 11 (step S10). When the weight coefficients C _a , C _t and C _l are already stored in the weight coefficient storage means 11, the weight coefficient calculation processing step of this estimation model formula can be omitted. Details of the weighting factor calculation processing step of the estimated model formula will be described later.

  Next, the sound image width estimation apparatus 100 uses the microphones 2L and 2R attached to the dummy head 1 to collect the sound generated from the sound source SS to be tested by the binaural method, and uses the collected analog sound signal as the low-pass filter 3L. And 3R through the AD converters 4L and 4R, the acoustic signals sl (n) and sr (n) converted into digital signals are stored in the memories 6L and 6R (step S11).

The sound image width estimation apparatus 100 reads out the acoustic signals sl (n) and sr (n) stored in the memories 6L and 6R in step S11 by the filter banks 7L and 7R, and the frequency band components sl (n , F) and sr (n, f) and output to the windowing means 20L _f and 20R _f of the corresponding frequency band-specific physical feature quantity calculating means 8 _f of the frequency band f (step S12).
Here, the sound image width estimation apparatus 100 performs calculations using 1/6 octave band filters as the filter banks 7L and 7R, respectively.

The sound image width estimation apparatus 100 uses the windowing means 20L _f and 20R _f to input frequency band components sl (n, f) and sr (n, f) of the corresponding frequency band f input from the filter banks 7L and 7R in step S12. ) Is multiplied by a time window w (n), and acoustic signals yl _k (n, f) and yr _k (n, f) of a predetermined time length at positions shifted by a predetermined movement width d are sequentially applied. cut.
The sound image width estimation apparatus 100 sequentially outputs the left-channel acoustic signal yl _k (n, f) cut out by the windowing means 20L _f to the corresponding frequency band f level calculation means 22L _f and CCC calculation means 21 _f. At the same time, the sound signal yr _k (n, f) of the right channel cut out by the windowing means 20R _f is sequentially output to the level calculating means 22R _f and the CCC calculating means 21 _f of the corresponding frequency band f (step S13).

The sound image width estimation apparatus 100 uses the level calculation means 22L _f and 22R _f to input acoustic signals yl _k (n, f) and yr of predetermined time lengths sequentially input from the windowing means 20L _f and 20R _{f in} step S13, respectively. _The levels slE _k (f) and sr _k E (f) are calculated from _k (n, f), respectively, and sequentially output to the ILD calculation means 25 _f corresponding to each frequency band f (step S14).
In parallel, the sound image width estimation apparatus 100 uses the acoustic signals yl _k (n, f) and yr _k (n, f) sequentially input from the windowing means 20L _f and 20R _f in step S13 by the CCC calculating means. The interaural cross-correlation coefficient CCC _k (f) is calculated and sequentially output to the IACC calculation means 23 _f and the ITD calculation means 24 _f corresponding to each frequency band f (step S14).

The sound image width estimation apparatus 100 calculates the interaural cross-correlation degree IACC _k (f) from the interaural cross-correlation coefficient CCC _k (f) input from the CCC calculation unit 21 _{f in} step S14 by the IACC calculation unit 23 _f . calculated and sequentially outputs the IACC average calculating unit 27 _f (step S15).
In parallel, the sound image width estimation apparatus 100 uses the ITD calculator 24 _{f to} calculate the interaural time difference ITD _k (f) from the interaural cross-correlation coefficient CCC _k (f) input from the CCC calculator 21 _{f in} step S14. ) is calculated sequentially outputs the ITD standard deviation calculating means 28 _f (step S15).
In parallel, the sound image width estimation apparatus 100 further performs inter-aural interplay between the levels slE _k (f) and sr _k E (f) input from the level calculation units 22L _f and 22R _{f in} step S14 by the ILD calculation unit 25 _f . The level difference ILD _k (f) is calculated and sequentially output to the ILD standard deviation calculating means 26 _f (step S15).

The sound image width estimation apparatus 100 calculates the average in the time axis direction by calculating the average in the time axis direction from the interaural cross-correlation degree IACC _k (f) input from the IACC calculation unit 23 _{f in} step S15 by the IACC average calculation unit 27 _f . The average is sequentially output to the IACC average representative value calculating means 31 as the physical characteristic amount x _a (f) for each frequency band (step S16).
In parallel, the sound image width estimation apparatus 100 calculates the standard deviation in the time axis direction from the interaural time difference ITD _k (f) input from the ITD calculation unit 24 _{f in} step S15 by the ITD standard deviation calculation unit 28 _f . The calculated standard deviation is sequentially output to the ITD standard deviation representative value calculating means 32 as a physical characteristic amount x _t (f) for each frequency band (step S16).
In parallel, the sound image width estimation apparatus 100 further calculates the standard deviation in the time axis direction from the interaural level difference ILD _k (f) input from the ILD calculation unit 25 _{f in} step S15 by the ILD standard deviation calculation unit 26 _f . The calculated standard deviation is sequentially output to the ILD standard deviation representative value calculating means 30 as the physical characteristic amount x _l (f) for each frequency band (step S16).

The sound image width estimation apparatus 100 calculates the average of the physical characteristic amount x _a (f) for each frequency band input from the IACC average calculation unit 27 _f in step S16 by the IACC average representative value calculation unit 31, and calculates the calculated average. and it outputs the sound image width estimate calculation unit 10 as a physical feature quantity representative value _{X a} (step S17).
In parallel, the sound image width estimation apparatus 100 calculates the average of the physical characteristics x _t (f) for each frequency band input from the ITD standard deviation calculation unit 28 _f in step S16 by the ITD standard deviation representative value calculation unit 32. , and it outputs the average calculated as a physical feature quantity representative value X _t sound image width estimation value calculating means 10 (step S17).
In parallel, the sound image width estimation apparatus 100 further calculates the average of the physical features x _l (f) for each frequency band input from the ILD standard deviation calculation unit 26 _f in step S16 by the ILD standard deviation representative value calculation unit 30. , and outputs the sound image width estimate calculation unit 10 the average calculated as a physical feature quantity representative value X _l (step S17).

The sound image width estimation apparatus 100 receives the physical feature amount input from the IACC average representative value calculation means 31, the ITD standard deviation representative value calculation means 32, and the ILD standard deviation representative value calculation means 30 in step S17 by the sound image width estimation value calculation means 10. From the representative values X _a , X _t and X _l and the weight coefficients C _a , C _t and C _l stored in the weight coefficient storage means 11 by the estimated value weight coefficient calculation means 12 in step S10, the equation (12 ) To calculate the estimated sound image width (hat y) and output it to the display means 14 (step S18).

The sound image width estimation apparatus 100 displays the sound image width estimated value (hat y) input from the sound image width estimated value calculating means 10 in step S18 so as to be visible on the display means 14 (step S19).
With the above processing, the sound image width estimation apparatus 100 can estimate the sound image width.

Next, the operation of the sound image width estimation apparatus 100 in the weighting factor calculation processing step (step S10) of the estimation model formula shown in FIG. 3 will be described with reference to FIG. 4 (see FIGS. 1 and 2 as appropriate).
As shown in FIG. 4, the sound image width estimation apparatus 100 first inputs subjective evaluation data y _i obtained by subjective evaluation performed in advance by an input unit (not shown) and stores it in the subjective evaluation data storage unit 13 ( Step S30).

Next, the sound image width estimation apparatus 100 collects the acoustic signal corresponding to the subjective evaluation data y _i input in step S30 by the binaural method using the microphones 2L and 2R, and uses the collected analog acoustic signal as the low-pass filters 3L and 3R. Are stored in the memories 6L and 6R as acoustic signals sl (n) and sr (n) converted into digital signals by the AD converters 4L and 4R (step S31).

The processing from step S32 to step S37 is the same as the processing from step S12 to step S17 in the processing shown in FIG.
Note that the sound image width estimation apparatus 100 repeats the processing of steps S31 to S37 corresponding to the number S of subjective evaluation data y _i input in step S30, and causes the estimated value weight coefficient calculation means 12 to receive S sets of subjective subjects. Data consisting of evaluation data y _i and physical feature quantity representative values X _ai , X _ti and X _li is stored.

The sound image width estimation apparatus 100 uses the estimated feature weight coefficient calculation unit 12 to input physical feature values from the IACC average representative value calculation unit 31, the ITD standard deviation representative value calculation unit 32, and the ILD standard deviation representative value calculation unit 30 in step S37. S sets of data including representative values X _ai , X _ti, and X _li and subjective evaluation data y _i input from input means (not shown) and stored in the subjective evaluation data storage means 13 in step S30 are used. Then, the weight coefficients C _a , C _t and C _l are calculated by the least square method (step S38), and the calculated weight coefficients C _a , C _t and C _l are stored in the weight coefficient storage means 11 (step S39). .
Thus, the sound image width estimation apparatus 100 ends the weighting factor calculation process of the estimation model formula.

  Next, referring to FIG. 5 (refer to FIG. 1 and FIG. 2 as appropriate), between the physical feature amount used in the sound image width estimation model expression in the present invention shown in Expression (12) and the subjective evaluation data of the sound image width. The result of Pearson correlation analysis will be described.

(1) to (4) in FIG. 5 are the results of Pearson correlation analysis using continuous sounds of violin G-line, A-line, D-line, and E-line as sound sources, respectively. 5 (1) to (4), the horizontal axis indicates the frequency bandwidth divided by the filter banks 7L and 7R. From the left in each figure, there is no frequency band division (1 band), 1/1 octave. The case where the band is set to 1/96 octave band is shown. The vertical axis represents the Pearson correlation coefficient. The data indicated by “◆”, “□”, and “▲” are the subjective evaluation data of the sound image width, the physical feature value representative value X _a that is the average for the frequency band of the binaural cross-correlation, respectively, binaural the correlation coefficient between the average of a physical characteristic amount representative value X _t and a physical feature quantity representative value X _l is the average of the frequency band of the standard deviation of the interaural level difference for the frequency band of the standard deviation between the time difference Show.

Both results show that the correlation becomes higher as the frequency bandwidth is narrowed. In particular, the correlation is high at 1/6 octave band or lower, and in the case of narrow band below 1/12 octave band, It can be seen that the value of the relation number is saturated.
From this analysis result, it is possible to predict the sound image width with stable accuracy by setting the frequency bandwidth divided by the filter banks 7L and 7R to preferably 1/6 octave band or less, more preferably 1/12 octave band or less. I understand.

[Second Embodiment]
Next, with reference to FIG.6 and FIG.7, 100 A of sound image width estimation apparatuses of 2nd Embodiment in this invention are demonstrated.
As shown in FIG. 6, the sound image width estimation apparatus 100 </ b> A according to the second embodiment includes a calculation means 5 </ b> A instead of the calculation means 5 in the same manner as the sound image width estimation apparatus 100 according to the first embodiment shown in FIG. 1. Is different. Specifically, the sound image width estimation apparatus 100A according to the second embodiment is different from the sound image width estimation apparatus 100 according to the first embodiment shown in FIG. 1 in the physical feature quantity representative value calculation unit 9 and the subjective evaluation data storage unit 13. Instead, the physical feature quantity representative value calculation means 9A and the subjective evaluation data storage means 13A are provided, and the representative value weight coefficient storage means 15 and the representative value weight coefficient calculation means 16 are further provided.

The physical feature quantity representative value calculation means 9 in the first embodiment uses the physical feature quantities x _a (f), x _t (f), and x ₁ as frequency feature representative values X _a , X _t, and X _l , respectively. The average of _l (f) is calculated, but the physical feature quantity representative value calculation unit 9A in the second embodiment uses the physical feature quantity x _{a for} each frequency band as the physical feature quantity representative values X _a , X _t, and X _l , respectively. A weighted average of (f), x _t (f) and x _l (f) is calculated.

As shown in FIG. 7, the physical feature quantity representative value calculation means 9A of the calculation means 5A in the second embodiment is the same as the physical feature quantity representative value calculation means 9 of the calculation means 5 in the first embodiment shown in FIG. In place of the ILD standard deviation representative value calculating means 30, the IACC average representative value calculating means 31 and the ITD standard deviation representative value calculating means 32, the ILD standard deviation representative value calculating means 30A, the IACC average representative value calculating means 31A and the ITD standard deviation The difference is that the representative value calculating means 32A is provided.
The same components as those in the first embodiment shown in FIGS. 1 and 2 are denoted by the same reference numerals, and description thereof will be omitted as appropriate.

The ILD standard deviation representative value calculation means 30A in the second embodiment receives the physical characteristic amount x _l (f) for each frequency band from the ILD standard deviation calculation means 26 _f and the weight coefficient c _l ( It reads f), as a physical feature quantity representative values _{X l,} calculates the weighted average by formula (20-3). ILD standard deviation representative value calculating unit 30A outputs the calculated physical characteristic amount representative value X _l sound image width estimation value calculating means 10.
The IACC average representative value calculation means 31A in the second embodiment receives the physical characteristic amount x _a (f) for each frequency band from the IACC average calculation means 27 _f and the weight coefficient c _a (f) from the weight coefficient storage means 15. And _a weighted average is calculated as the physical feature value representative value Xa by the equation (20-1). The IACC average representative value calculating unit 31A outputs the calculated physical feature amount representative value _Xa to the sound image width estimated value calculating unit 10.
The ITD standard deviation representative value calculation means 32A in the second embodiment receives the frequency band-specific physical feature value x _t (f) from the ITD standard deviation calculation means 28 _f and the weight coefficient storage means 15 from the weight coefficient c _t ( It reads f), as a physical feature quantity representative value _{X t,} and calculates the weighted average by formula (20-2). The ITD standard deviation representative value calculating unit 32A outputs the calculated physical feature amount representative value _Xt to the sound image width estimated value calculating unit 10.

Similar to the subjective evaluation data storage unit 13 in the first embodiment shown in FIG. 2, the subjective evaluation data storage unit 13A is a weight for calculating the estimated sound image width (hat y) by the representative value weight coefficient calculation unit 12. Subjective evaluation data y _i used for calculating the coefficients C _a , C _t and C _l is stored. In addition, the subjective evaluation data storage unit 13A has weight coefficients c _a (f) and c _t (f) for calculating the physical feature quantity representative values X _a , X _t and X _l by the representative value weight coefficient calculation unit 16. And subjective evaluation data y _i used for calculating c _l (f). Subjective evaluation data y _i used in calculating the weighting factors C _a , C _t and C _l and subjectivity used in calculating the weighting factors c _a (f), c _t (f) and c _l (f) The evaluation data y _i may share the same data, or may be different data.
These subjective evaluation data y _i are input by an input unit (not shown) and stored in the subjective evaluation data storage unit 13A.

The weighting factor calculation means 16 reads the subjective evaluation data y _i from the subjective evaluation data storage means 13A, and the subjective evaluation data from the IACC average calculation means 27 _f , the ITD standard deviation calculation means 28 _f, and the ILD standard deviation calculation means 26 _f , respectively. The three types of physical feature amounts x _ai (f), x _ti (f) and x _li (f) corresponding to the data y _i are inputted, and the subjective evaluation data y inputted for each type of physical feature amount Using a plurality of sets of data consisting of _i and frequency band physical feature quantities x _ai (f), x _ti (f), and x _li (f), equations (21-1), (21-2) and The weighting coefficients c _a (f), c _t (f), and c _l (f) of the estimation model expression of the sound image width shown in Expression (21-3) are calculated by the least square method that is a regression analysis technique. Then, the weighting factor calculation unit 16 stores the calculated weighting factors c _a (f), c _t (f), and c _l (f) in the weighting factor storage unit 15. Note that i is a number for identifying individual subjective evaluation data.

Here, the subjective evaluation data _{y i} in the corresponding frequency band specific physical feature quantity _x ai _(f), and _x ti (f) and _x li (f) includes the subject of when to obtain the subjective evaluation data _{y i} Under the same sound field conditions, acoustic signals are collected using the microphones 2L and 2R attached to the dummy head 1, and finally the IACC average calculating means 27 _f and the ITD standard deviation calculating means 28 are used using the above-described analyzing means. _f and ILD standard deviation calculation means 26 frequency bands output from the _f-specific physical characteristic amount _x ai _(f), is that the _x ti (f) and _x li (f).
The calculation method of the weighting factors c _a (f), c _t (f), and c _l (f) is the same as the calculation method of the weighting factors C _a , C _t, and C _l in the first embodiment described above. Therefore, explanation is omitted.

In the second embodiment, the weighted average is used as the representative value of the physical feature values x _a (f), x _t (f), and x _l (f) for each frequency band. As representative values of the feature quantities x _a (f), x _t (f), and x _l (f), as shown in the formulas (22-1) to (22-3), the physical feature quantities by frequency band x The weighted maximum value for each physical feature quantity type of _a (f), x _t (f), and x _l (f), and the frequency as shown in Expressions (23-1) to (23-3) band specific physical feature quantity _x a _(f), may be used the median of each type of physical feature values of _x t (f) and _x l (f).

However, the formula (23-1) to formula in (23-3), median _{(a 1,} a 2, ..., _{a F)} is () elements _{a 1,} a 2 in, ..., in _{a F} This function calculates the median.
In addition, the weight coefficients c _a (f), c _t (f), and c _l (f) in these expressions (22-1) to (22-3) and (23-1) to (23-3) ) Can be determined by the same method as the weighted average calculation method described above.

The weighting coefficient storage means 15 is a weighting coefficient of the estimated model expression of the sound image width shown in the expressions (21-1), (21-2) and (21-3) calculated by the representative value weighting coefficient calculating means 16. Calculation of c _a (f), c _t (f), and c _l (f), that is, physical feature representative values X _a , X _t, and X _l shown in equations (20-1) to (20-3) Storage means for storing the weighting factors c _a (f), c _t (f) and c _l (f) of the _equation . The weighting coefficients c _a (f), c _t (f), and c _l (f) stored in the weighting coefficient storage unit 15 are the IACC average representative value calculating unit 31A, the ITD standard deviation representative value calculating unit 32A, and the ILD standard deviation representative. It is read by the value calculating means 30A and used to calculate the physical feature quantity representative values X _a , X _t and X _l , respectively.

The estimated value weighting coefficient calculating means 12 in the second embodiment is a method similar to the estimated value weighting coefficient calculating means 12 in the first embodiment, and the weighting factors C _a and C _{t of} the estimated model formula shown in Expression (12) And C _l are calculated, and the calculated weight coefficients C _a , C _t and C _l are stored in the weight coefficient storage means 11.

At this time, the IACC average representative value calculating means 31A, the ITD standard deviation representative value calculating means 32A, and the ILD standard deviation representative value calculating means 30A are respectively weight coefficients c _a (calculated in advance by the representative value weight coefficient calculating means 16). f), c _t (f) and c _l (f) are read from the weight coefficient storage means 15, and the read weight coefficients c _a (f), c _t (f) and c _l (f) are respectively calculated as IACC averages. From the physical features x _a (f), x _t (f), and x _l (f) for each frequency band inputted from the means 27 _f , the ITD standard deviation calculating means 28 _f and the ILD standard deviation calculating means 26 _f , The feature quantity representative values X _a , X _t and X _l are calculated, and the calculated physical feature quantity representative values X _a , X _t and X _l are output to the estimated value weight coefficient calculation means 12.

Next, the operation of the sound image width estimation apparatus 100A of the second embodiment will be described with reference to FIG. 8 (see FIGS. 6 and 7 as appropriate).
As illustrated in FIG. 8, the sound image width estimation apparatus 100 </ _b > A first uses the representative value weighting coefficient calculation unit 16 to represent the physical feature quantity representative values X _a and X shown in Expressions (20-1) to (20-3). Weight coefficients c _a (f), c _t (f), and c _l (f) in the calculation formulas for _t and X _l are calculated and stored in the weight coefficient storage unit 15 (step S50). If the weighting factors c _a (f), c _t (f), and c _l (f) are already stored in the weighting factor storage means 15, this weighting factor calculation processing step for calculating the physical feature quantity representative value is omitted. can do. Details of the weighting factor calculation processing step for calculating the physical feature quantity representative value will be described later.

Subsequently, the sound image width estimation apparatus 100A calculates weighting factors C _a , C _t and C _l in the estimation model equation shown in the equation (12) by the estimated value weighting factor calculation unit 12, and the weighting factor storage unit 11 (Step S51). When the weight coefficients C _a , C _t and C _l are already stored in the weight coefficient storage means 11, the weight coefficient calculation processing step of this estimation model formula can be omitted. Note that step S51 is the same as step S10 in the process of the sound image width estimation apparatus 100 of the first embodiment shown in FIG.

  The processes in steps S52 to S57 are the same as steps S11 to S16 in the process of the sound image width estimation apparatus 100 of the first embodiment shown in FIG.

Sound image width estimating apparatus 100A, by IACC average representative value calculating unit 31A, a frequency band specific physical feature quantity _x a (f) input from the IACC average calculating unit 27 _f at step S57, the the weight coefficient storage unit 15 in step S50 From the stored weight coefficient c _a (f), _a weighted average is calculated by the equation (20-1), and the calculated weighted average is used as a physical feature amount representative value X _{a to} calculate _a sound image width estimated value. 10 (step S58).
Sound image width estimator 100A, in parallel, by ITD standard deviation representative value calculating unit 32A, ITD and standard deviation calculating means each frequency band inputted from the 28 _f physical feature amount _x t (f) in step S57, the in step S50 A weighted average is calculated from the weighting coefficient c _t (f) stored in the weighting coefficient storage means 15 by the equation (20-2), and the calculated weighted average is used as the physical feature quantity representative value X _t. It outputs to the sound image width estimated value calculation means 10 (step S58).
Sound image width estimator 100A further in parallel, by ILD standard deviation representative value calculating unit 30A, an ILD standard deviation calculation means 26 frequency bands specific physical feature quantity input from the _{f x} l _(f) in step S50, step S50 Then, the weighted average is calculated from the weight coefficient c _l (f) stored in the weight coefficient storage means 15 by the equation (20-3), and the calculated weighted average is used as the physical feature quantity representative value X _l. Is output to the sound image width estimated value calculating means 10 (step S58).

The sound image width estimation apparatus 100A receives the physical feature amount input from the IACC average representative value calculation means 31A, the ITD standard deviation representative value calculation means 32A, and the ILD standard deviation representative value calculation means 30A in step S58 by the sound image width estimation value calculation means 10. From the representative values X _a , X _t and X _l and the weight coefficients C _a , C _t and C _l stored in the weight coefficient storage means 11 by the estimated value weight coefficient calculation means 12 in step S51, the equation (12 ) To calculate the estimated sound image width (hat y) and output it to the display means 14 (step S59).

The sound image width estimation apparatus 100A displays the sound image width estimated value (hat y) input from the sound image width estimated value calculating means 10 in step S59 so as to be visible on the display means 14 (step S60).
With the above processing, the sound image width estimation apparatus 100A can estimate the sound image width.

Next, referring to FIG. 9 (refer to FIG. 6 and FIG. 7 as appropriate), the operation of the sound image width estimation apparatus 100A in the physical coefficient representative value calculation weight coefficient calculation processing step (step S50) shown in FIG. explain.
As shown in FIG. 9, the sound image width estimation apparatus 100A first inputs subjective evaluation data y _i obtained by subjective evaluation performed in advance by input means (not shown) and stores it in the subjective evaluation data storage means 13A ( Step S70).

Next, the sound image width estimation apparatus 100A collects the acoustic signal corresponding to the subjective evaluation data y _i input in step S70 by the binaural method using the microphones 2L and 2R, and uses the collected analog acoustic signal as the low-pass filters 3L and 3R. Are stored in the memories 6L and 6R as acoustic signals sl (n) and sr (n) converted into digital signals by the AD converters 4L and 4R (step S71).

The processing from step S72 to step S76 is the same as the processing from step S53 to step S57 in the processing shown in FIG.
Note that the sound image width estimation apparatus 100A repeats the processing of steps S71 to S76 corresponding to the number S of subjective evaluation data y _i input in step S70, and causes the representative value weight coefficient calculation means 16 to receive S sets of subjectives. Data consisting of evaluation data y _i and physical characteristics by frequency band x _a (f), x _t (f) and x _l (f) is stored.

The sound image width estimation apparatus 100A uses the representative value weighting coefficient calculation unit 16 for each frequency band input from the IACC average representative value calculation unit 31A, the ITD standard deviation representative value calculation unit 32A, and the ILD standard deviation representative value calculation unit 30A in step S76. The physical feature values x _ai (f), x _ti (f), and x _li (f) and the subjective evaluation data y input from the input unit (not shown) and stored in the subjective evaluation data storage unit 13A in step S70. Using the S sets of data consisting of _i , weight coefficients c _a (f), c _t (f) and c _l (f) are calculated by the method of least squares (step S78), and the calculated weight coefficient c _a (F), c _t (f) and c _l (f) are stored in the weight coefficient storage means 15 (step S79).
Thus, the sound image width estimation apparatus 100A ends the weighting coefficient calculation process for calculating the physical feature quantity representative value.

[Third Embodiment]
Next, a sound image width estimation apparatus 100B according to a third embodiment of the present invention will be described with reference to FIG.
As shown in FIG. 10, the sound image width estimation apparatus 100 </ b> B according to the third embodiment is different from the sound image width estimation apparatus 100 according to the first embodiment shown in FIG. Is different. Specifically, the sound image width estimation apparatus 100B according to the third embodiment is different from the sound image width estimation apparatus 100 according to the first embodiment shown in FIG. In place of the estimated width value calculation means 10, the weight coefficient storage means 11 and the subjective evaluation data storage means 13, a sound image width estimation value calculation means 10B, a weight coefficient storage means 11B and a subjective evaluation data storage means 13B are provided. Is different.

The sound image width estimated value calculating means 10 in the first embodiment is a sound image width estimated value (in accordance with an estimation model formula using physical feature quantity representative values X _a , X _t and X _l calculated by the physical feature quantity representative value calculating means 9. In contrast to calculating the hat y), the sound image width estimated value calculating means 10B in the third embodiment is the physical feature quantity x _a (f) for each frequency band calculated by the physical feature quantity calculating means 8 _f for each frequency band. ), X _t (f) and x _l (f) are treated as physical feature quantities, and an estimation model using these frequency band physical feature quantities x _a (f), x _t (f) and x _l (f) The estimated sound image width (hat y) is calculated by the equation.
The same constituent elements as those in the first embodiment shown in FIG. 1 are denoted by the same reference numerals, and description thereof will be omitted as appropriate.

Sound width estimation value calculation means in the third embodiment 10B includes, IACC average calculating unit ₂₇ f of the frequency band specific physical feature calculating unit _{8 f,} ITD standard deviation calculating means 28 _f and ILD standard deviation calculation unit 26 _f (FIG. 2 ), The physical feature amount by frequency band x _a (f), which is the average of the interaural cross-correlation for each frequency band in the time axis direction, and the standard deviation in the time axis of the interaural time difference by frequency band inputs the frequency band specific physical feature amount x _{t (f)} and the standard deviation of the time axis direction of the level difference between each frequency band of the binaural frequency band specific physical feature quantity x _{l (f)} is the weighting factor The weight coefficients c _a (f), c _t (f), and c _l (f) are read from the storage unit 11B, and the estimated sound image width (hat y) is calculated by the estimation model formula shown in the formula (24). The sound image width estimated value calculation means 10B outputs the calculated sound image width estimated value (hat y) to the display means 14.

Note that the estimation model formula is not limited to the formula (24). For example, as shown in the formula (25), the physical feature amounts x _a (f), x _t (f), and x _l (f) for each frequency band. Other estimation model formulas represented by) may be used.

The weighting factor storage unit 11B stores the weighting factors c _a (f), c _t (f), and c _l (f) of the estimation model formula shown in the formula (24) calculated by the estimated value weighting factor calculation unit 12B. It is a storage means. The weight coefficients c _a (f), c _t (f), and c _l (f) stored in the weight coefficient storage unit 11B are read out by the sound image width estimated value calculation unit 10B when the sound image width is estimated. This is used to calculate the estimated sound image width (hat y).

The estimated value weighting coefficient calculation means 12B reads the subjective evaluation data y _i from the subjective evaluation data storage means 13B, and at the same time, the IACC average calculation means 27 _f and the ITD standard deviation calculation means 28 _{f of the} physical characteristic amount calculation means 8 _f by frequency band. And ILD standard deviation calculation means 26 _f (see FIG. 2), three types of frequency feature physical characteristics x _ai (f), x _ti (f) and x _li (f) corresponding to the subjective evaluation data y _i , respectively. And a plurality of sets of data consisting of the input subjective evaluation data y _i and frequency band physical feature quantities x _ai (f), x _ti (f), and x _li (f), The weight coefficients c _a (f), c _t (f), and c _l (f) of the estimation model formula of the sound image width shown in FIG. 6 are calculated by the least square method that is a regression analysis method. Then, the weighting factor calculation unit 12B stores the calculated weighting factors c _a (f), c _t (f), and c _l (f) in the weighting factor storage unit 11B. Note that i is a number for identifying individual subjective evaluation data.

Here, the subjective evaluation data _{y i} in the corresponding frequency band specific physical feature quantity _x ai _(f), and _x ti (f) and _x li (f) includes the subject of when to obtain the subjective evaluation data _{y i} Under the same sound field conditions, acoustic signals are collected using the microphones 2L and 2R attached to the dummy head 1, and finally the IACC average calculating means 27 _f and the ITD standard deviation calculating means 28 are used using the above-described analyzing means. _f and ILD standard deviation calculating means 26 _f (refer to FIG. 2) are physical features x _ai (f), x _ti (f) and x _li (f) for each frequency band output from the frequency band.
The calculation method of the weighting factors c _a (f), c _t (f), and c _l (f) is the same as the calculation method of the weighting factors C _a , C _t, and C _l in the first embodiment described above. Therefore, explanation is omitted.

The estimated value weight coefficient calculating means 12B in the third embodiment is the same method as the estimated value weight coefficient calculating means 12 in the first embodiment, and the weight coefficient c _a (f) of the estimated model equation shown in Expression (24). , C _t (f) and c _l (f) are calculated, and the calculated weight coefficients c _a (f), c _t (f) and c _l (f) are stored in the weight coefficient storage unit 11B.

The subjective evaluation data storage unit 13B calculates weight coefficients c _a (f), c _t (f), and c _l (f) for calculating the sound image width estimated value (hat y) by the estimated value weight coefficient calculating unit 12B. Subjective evaluation data y _i used in the process is stored. The subjective evaluation data y _i is input by an input unit (not shown) and stored in the subjective evaluation data storage unit 13B.

Next, referring to FIG. 11 (refer to FIG. 10 as appropriate), the operation of the sound image width estimation apparatus 100B of the third embodiment will be described.
As shown in FIG. 11, the sound image width estimation apparatus 100B first uses the estimated value weighting coefficient calculation unit 12B to calculate the weighting coefficient c _a of the estimated model formula of the sound image width estimated value (hat y) shown in Expression (24). f), c _t (f) and c _l (f) are calculated and stored in the weight coefficient storage means 11B (step S90). When the weighting factors c _a (f), c _t (f), and c _l (f) are already stored in the weighting factor storage unit 11B, the weighting factor calculation processing step of this estimation model formula can be omitted. . The weighting factor calculation processing step of the estimation model formula is different from the weighting factor calculation processing step of the estimation model formula in the first embodiment shown in FIG. 3 as the physical feature quantity representative values X _a , X _t and X Instead of _l , the physical characteristics by frequency band x _a (f), x _t (f) and x _l (f) are used instead of _l , and weight coefficients c _a (f) are substituted for the weight coefficients C _a , C _t and C _l. , C _t (f) and c _l (f) are the same except that they are calculated, and thus detailed description thereof is omitted.

  The processing in steps S91 to S96 is the same as that in steps S11 to S16 in the processing of the sound image width estimation apparatus 100 of the first embodiment shown in FIG.

The sound image width estimation device 100B uses the sound image width estimation value calculation means 10B to perform the IACC average calculation means 27 _f , the ITD standard deviation calculation means 28 _f and the ILD standard deviation calculation means of the frequency band physical feature value calculation means 8 _f in step S96. 26 _f (refer to FIG. 2), the physical features x _a (f), x _t (f) and x _l (f) for each frequency band respectively input from 26 _f (see FIG. 2) and stored in the weight coefficient storage means 11B in step S90. A sound image width estimated value (hat y) is calculated from the weighting coefficients c _a (f), c _t (f), and c _l (f) by the equation (24), and the calculated sound image width estimated value (hat y) is calculated. It outputs to the display means 14 (step S97).

The sound image width estimation device 100B displays the sound image width estimation value (hat y) input from the sound image width estimation value calculation unit 10B in step S97 so as to be visible on the display unit 14 (step S98).
With the above processing, the sound image width estimation apparatus 100B can estimate the sound image width.

Next, examples of the present invention will be described.
In the sound image width estimation apparatus 100 shown in FIG. 1 and FIG. 2, a continuous sound by each open string of a violin is recorded and used as the sound source SS. Subjective evaluation is performed by using the experimental apparatus 110 shown in FIG. 12 and appropriately recording speakers SS _{1 to} SS ₃ on the semicircle SC on the front side of the subject SUB centered on the subject SUB. It was performed by playing the continuous sound of the violin. At this time, the subject SUB can perceive various sound image widths by adjusting the arrangement number of the speakers SS _{1 to} SS _{3 as} the sound source and the arrangement angle θ _SS of the speakers SS ₁ to SS ₃ around the subject SUB. Controlled.
The subjective evaluation value of the sound image width was converted into a horizontal angle θ with the center of the head of the subject SUB as the viewpoint.

  Next, the dummy head 1 was placed at the same position as when the subject SUB evaluated under the same sound field conditions as in the subjective evaluation described above, and acoustic signals were collected using the binaural method using the microphones 2L and 2R. An estimated value (hat y) of the sound image width was calculated for the collected acoustic signal using the sound image width estimation apparatus 100 shown in FIGS. 1 and 2. At this time, the frequency band was divided using filter banks 7L and 7R constituted by 1/24 octave filters having a lower limit frequency of 150 Hz and an upper limit frequency of 12 kHz.

In this example, subjective evaluation was performed on 20 types of sound stimuli. The result is shown in FIG. In FIG. 13, the horizontal axis represents the estimated value of the sound image width (hat y), and the vertical axis represents the subjective evaluation value of the sound image width, and the results are shown. As shown in FIG. 13, according to the present invention, it is possible to estimate the sound image width better than the conventional technique.
In the present embodiment, the linear function of the physical feature amount is used as the estimation model formula of the sound image width, but is not limited to this, and a quadratic function, a power function, an exponential function, etc. of the physical feature amount are used. It can also be used.

DESCRIPTION OF SYMBOLS 1 Dummy head 2L, 2R Microphone 3L, 3R Low pass filter 4L, 4R AD converter 5, 5A, 5B Calculation means 6L, 6R Memory 7L, 7R Filter bank (frequency band division means)
8 _f frequency band specific physical feature calculating unit (frequency band feature quantity calculating means)
9, 9A Physical feature quantity representative value calculation means (physical feature quantity calculation means)
10, 10B Sound image width estimated value calculating means (estimated value calculating means)
11, 11B Weight coefficient storage means 12, 12B Estimated value weight coefficient calculation means (weight coefficient calculation means)
13, 13A, 13B Subjective evaluation data storage means 14 Display means 15 Weight coefficient storage means 16 Representative value weight coefficient calculation means 20L _f , 20R _f Windowing means 21 _f CCC calculation means 22L _f , 22R _f level calculation means 23 _f IACC calculation Means 24 _f ITD calculating means 25 _f ILD calculating means 26 _f ILD standard deviation calculating means 27 _f IACC average calculating means 28 _f ITD standard deviation calculating means 30, 30A ILD standard deviation representative value calculating means 31, 31A IACC average representative value calculating means 32, 32A ITD standard deviation representative value calculation means 100, 100A, 100B Sound image width estimation device SS, SS _{1 to} SS ₃ sound source SUB subject

Claims (7)

A physical feature amount is calculated from a digital acoustic signal having two channels on the left and right, and the calculated sound feature width is applied to a model image estimation model of the sound image width including the physical feature amount and a weight coefficient. A sound image width estimating device for
Frequency band dividing means for dividing the digital audio signal consisting of two left and right channels into a plurality of frequency band subband signals having a frequency bandwidth of 1/6 octave or less for each of the left and right channel audio signals;
From the subband signals divided by the frequency band dividing means, the interaural cross-correlation, the standard deviation in the time axis direction of the binaural time difference, or the standard in the time axis direction of the binaural level difference for each subband signal. A frequency band feature quantity calculating means for calculating a frequency band feature quantity which is a feature quantity by frequency band representing a difference between the left and right channels of the subband signal as at least one of the deviations;
Physical feature quantity calculating means for calculating the physical feature quantity based on the frequency band feature quantity calculated by the frequency band feature quantity calculating means;
An estimated value calculating means for calculating the estimated value of the sound image width by applying the physical feature value calculated by the physical feature value calculating means to the estimated model equation;
A sound image width estimation apparatus comprising:   The physical feature quantity calculating means calculates any one of the average, weighted average, maximum value, or median of the feature quantity by frequency band for each subband signal calculated by the feature quantity calculation means by frequency band, The sound image width estimation device according to claim 1, wherein the sound image width estimation device calculates the physical feature amount. A physical feature amount is calculated from a digital acoustic signal having two channels on the left and right, and the calculated sound feature width is applied to a model image estimation model of the sound image width including the physical feature amount and a weight coefficient. A sound image width estimating device for
Frequency band dividing means for dividing the digital audio signal consisting of two left and right channels into a plurality of frequency band subband signals having a frequency bandwidth of 1/6 octave or less for each of the left and right channel audio signals;
From the subband signals divided by the frequency band dividing means, the interaural cross-correlation, the standard deviation in the time axis direction of the binaural time difference, or the standard in the time axis direction of the binaural level difference for each subband signal. A frequency band feature quantity calculating means for calculating a frequency band feature quantity which is a feature quantity by frequency band representing a difference between the left and right channels of the subband signal as at least one of the deviations;
Estimated value calculating means for calculating the estimated value of the sound image width by applying the individual characteristic values for each frequency band calculated by the frequency band characteristic amount calculating means to the estimation model formula as the physical feature quantities;
A sound image width estimation apparatus comprising:   4. The sound image width estimation apparatus according to claim 1, wherein the frequency band dividing unit divides the frequency band into subband signals having a frequency bandwidth of 1/12 octave or less.   Furthermore, the weighting factor calculating means for calculating the weighting factor in the estimation model formula is provided, and the weighting factor calculating means uses the physical feature amount as an explanatory variable and the weighting factor by regression analysis using the sound image width as a target variable. The sound image width estimation device according to any one of claims 1 to 4, wherein the sound image width estimation device is calculated. A physical feature amount is calculated from a digital acoustic signal having two channels on the left and right, and the calculated sound feature width is applied to a model image estimation model of the sound image width including the physical feature amount and a weight coefficient. Computer to
Frequency band dividing means for dividing the digital audio signal consisting of two channels on the left and right into subband signals of a plurality of predetermined frequency bands for each of the left and right channel audio signals;
From the subband signals divided by the frequency band dividing means, the interaural cross-correlation, the standard deviation in the time axis direction of the binaural time difference, or the standard in the time axis direction of the binaural level difference for each subband signal. Frequency band feature quantity calculating means for calculating a frequency band feature quantity which is a feature quantity by frequency band representing a difference between the left and right channels of the subband signal as at least one of the deviations;
Physical feature quantity calculating means for calculating the physical feature quantity based on the frequency band feature quantity calculated by the frequency band feature quantity calculating means;
An estimated value calculating means for calculating the estimated value of the sound image width by applying the physical feature value calculated by the physical feature value calculating means to the estimated model equation;
A sound image width estimation program that functions as a computer program. A physical feature amount is calculated from a digital acoustic signal having two channels on the left and right, and the calculated sound feature width is applied to a model image estimation model of the sound image width including the physical feature amount and a weight coefficient. Computer to
Frequency band dividing means for dividing the digital audio signal consisting of two channels on the left and right into subband signals of a plurality of predetermined frequency bands for each of the left and right channel audio signals;
From the subband signals divided by the frequency band dividing means, the interaural cross-correlation, the standard deviation in the time axis direction of the binaural time difference, or the standard in the time axis direction of the binaural level difference for each subband signal. Frequency band feature quantity calculating means for calculating a frequency band feature quantity which is a feature quantity by frequency band representing a difference between the left and right channels of the subband signal as at least one of the deviations;
Estimated value calculating means for calculating the estimated value of the sound image width by applying each frequency band-specific feature value calculated by the frequency band-specific feature value calculating means to the estimated model equation as the physical feature value;
A sound image width estimation program that functions as a computer program.

Images