JP2005109914A - Method and device for reproducing high presence sound field, and method for preparing head transfer function database - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reproduce a high presence sound field by using a small-scale head transfer function database. <P>SOLUTION: A head transfer function calculated by measurement is simplified to a head transfer function obtained by taking the property of a hearing filter into consideration, and the simplified head transfer function is stored to prepare a database. The simplified head transfer function is selected from the database according to listener information and sound source position information, the simplified head transfer function is folded on a sound signal that the listener wants to receive, and the sound signal with the simplified head transfer function folded thereon is presented to the listener to reproduce a high presence sound field. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a high realistic sound field reproduction method using binaural technology, a method for creating a head related transfer function database used in this sound field reproduction method, and a high realistic sound field reproduction device, and in particular, using a database smaller than the conventional method, To provide a high realistic sound field reproduction method capable of giving a high sense of presence to a larger number of listeners, a method of creating a head related transfer function database used therefor, and a high realistic sound field reproduction device To do.

Conventionally, a stereo signal of a stereo sound source recorded in binaural (for example, a sound source recorded by convoluting a head-related transfer function equivalent to that of a listener using a dummy head) is used as a high-realistic sound field reproduction method using headphones. The listener's head-related transfer function and the sound signal from the sound source are measured in advance using a stereo sound source that is not binaurally recorded. There is a method of performing a convolution calculation with the stereo signal and giving a sense of out-of-head localization to the listener by the stereo signal subjected to the convolution calculation (Non-patent Document 1).
Corey I. Cheng, Gregory H. Wakefield, “Introduction to Head-Related Transfer Functions (HRTFs): Representations of HRTFs in Time, Frequency, and Space”, J. Audio Eng. Soc., Vol. 49, No. 4, pp.231-249,2001.

  The frequency characteristic of the head-related transfer function has a very complicated shape including large individual differences derived from differences in individual head shapes. In order to realize listening with high presence, conventionally, it has been necessary to measure a high-precision head-related transfer function for each listener in all directions and create a database of head-related transfer functions. Therefore, there has been a first problem that the database used in the conventional method becomes very large.

  As a method for solving this first problem, there is an attempt to simplify the head-related transfer function and compress the data amount. Methods for simplifying the head-related transfer function include a method using a bandpass filter and a polynomial approximation. However, the simplification of these head-related transfer functions does not consider features such as auditory binaural processing. That is, with the simplified method that does not take into account auditory features, a second problem arises that information compression by simplification adversely affects the sense of reality during sound field reproduction.

  An object of the present invention is to provide a high realistic sound field reproduction method using binaural technology by using a database having a scale smaller than that of the conventional method and to give a higher sense of presence to a larger number of listeners than the conventional method. is there.

  According to claim 1 of the present invention, a head-related transfer function database created by simplifying a head-related transfer function using a band-pass filter group having properties of an auditory filter and holding the simplified head-related transfer function Is a high realistic sound field reproduction method using, based on the listener information and the sound source position information obtained from the sound signal emitted from the sound source, select a simplified head related transfer function from the head related transfer function database, A highly realistic sound field reproduction method is proposed in which the selected simplified head-related transfer function is convoluted with an acoustic signal.

  According to a second aspect of the present invention, there is provided a high realistic sound field reproduction method according to the first aspect, wherein the signal obtained as a result of convolving the selected simplified head-related transfer function with the acoustic signal is electroacoustic converted. A sound field reproduction method is proposed.

  According to a third aspect of the present invention, in the high realistic sound field reproduction method according to the first aspect, the bandpass filters constituting the bandpass filter group gradually increase in bandwidth as the center frequency becomes higher. We propose a high-realistic sound field playback method.

  According to a fourth aspect of the present invention, in the high realistic sound field reproduction method according to the first aspect, an average value of gain is obtained for each frequency component included in the band of the bandpass filter, and the average value of the gain is used. Therefore, we propose a highly realistic sound field reproduction method that simplifies the head-related transfer function.

  According to a fifth aspect of the present invention, there is provided a method for creating a head related transfer function database used in any one of the high realistic sound field reproduction methods according to the first to fourth aspects, wherein the bandwidth of the auditory filter is high on the low frequency side. It has the characteristics of a bandpass filter group that is narrower than the bandwidth on the band side and gradually wider than the bandwidth on the low band side on the high band side, and the frequency component included in the band of each bandpass filter of this bandpass filter group A head related transfer function database creation method is proposed in which an average value of gain is obtained and a simplified head related transfer function having the average value of gain as an element is stored to create a head related transfer function database.

  According to a sixth aspect of the present invention, a sound source, sound source position information selecting means for selecting sound source position information from an acoustic signal output from the sound source, listener information setting means for setting listener information, and for each listener and sound source position A head-related transfer function database storing a simplified head-related transfer function classified in consideration of the properties of the auditory filter, and sound source position information selected by the sound source position information selecting means and the listener information setting means A head-related transfer function selecting means for selecting a corresponding head-related transfer function from the head-related transfer function database according to the listener information, and a head-related transfer function selected by the head-related transfer function selecting means is convolved with an acoustic signal output from the sound source. Convolution operation means for performing arithmetic processing, and output means for outputting an acoustic signal in which the head-related transfer function is convoluted by the convolution operation means to the electroacoustic conversion means. To propose a high presence sensorineural field reproducing apparatus.

  The head-related transfer function database created by the method for creating a head-related transfer function database proposed in the present invention simplifies the head-related transfer function using the average value of the frequency components included in the band of each bandpass filter as an element. Since the characteristics of the head-related transfer function are smoothed, individual differences in the head-related transfer function can be reduced. As a result, one simplified head-related transfer function can be applied to a plurality of listeners, and the scale of the head-related transfer function database can be reduced. Further, by simplifying the frequency characteristic of the head-related transfer function while suppressing the influence on the loss of realism, it becomes possible to easily compress the information while suppressing the influence on the loss of realism.

  A head-related transfer function (also referred to as a simplified head-related transfer function or a simplified head-related transfer function) is recorded in the head-related transfer function database. A head-related transfer function, which is selected based on the sound source position information and the listener information and is recorded in the head-related transfer function database unit, taking into account the characteristics of the auditory filter, is convolved with the acoustic signal generated from the sound source by the convolution calculation means. The acoustic signal in which the head-related transfer function considering the properties of the auditory filter is convolved is transmitted as a highly realistic sound field to the listener via the headphones. The head-related transfer function database is created based on the head-related transfer functions measured for each listener and each sound source direction. The head-related transfer function created in the present invention has the same format as a conventional head-related transfer function database. However, as a database element, the head-related transfer function is converted through a filter group expressing the properties of the auditory filter. The head-related transfer function considering the characteristics of the filter is recorded. By the above conversion, individual differences in the head-related transfer function database are reduced, and a new listener who is not in the database can be given a high sense of presence by using data of other listeners. In addition, the frequency characteristic of the head related transfer function decreases with the frequency, and the frequency characteristics of the head related transfer function are reduced to a shape resembling a continuous and smooth polynomial function. The frequency characteristic can be expressed using approximation by a polynomial function or the like, and information compression of data constituting the head related transfer function can be performed efficiently and easily.

  In other words, the present invention uses the fact that in a binaural processing system for human hearing, acoustic signals are simplified without individual differences by a plurality of bandpass filters called auditory filters. For this reason, the present invention simplifies the head-related transfer function, which has been used in a conventional high-sense listening system and has an individual-specific and complicated shape, into an information-compressible one with little individual difference. I can do it.

  An embodiment of a highly realistic sound field reproducing apparatus according to the present invention will be described with reference to FIG. The sound source 1 is an arbitrary sound source. For example, a sound source of a type that reproduces an acoustic signal from a storage medium, or a sound source in the case of communication such as an audio conference can be considered. The sound signal output from the sound source 1 selects sound source position information attached to the sound signal by the sound source position information selection means 2, and the sound signal is sent to the convolution calculation means 3.

  Here, the sound source position information is represented by, for example, orthogonal coordinates (x, y, z) of a space with the center of the listener's head as the origin, and cylindrical coordinates (azimuth angle, elevation angle, distance). These can measure the position of the actual sound source or use information set in advance by the listener 8 or the like. In any case, it is essential to obtain the sound source position information before the convolution operation means 3 performs the operation.

  Reference numeral 4 denotes listener information setting means. The listener information for identifying the listener 9 is set in the listener information setting means 4. The listener information set in the listener information setting means 4 and the sound source position information selected by the sound source position information selection means 2 are input to the head-related transfer function selection means 6, and the head-related transfer function is selected by the head-related transfer function selection means 6. A head-related transfer function corresponding to the listener information and the sound source position information is selected from the group of head-related transfer functions stored in the database 5.

  That is, in the present invention, the head-related transfer function group in consideration of the auditory property is stored in the head-related transfer function database 5 categorized and stored according to the listener and the sound source position. The head-related transfer function selecting means 6 selects and extracts the head-related transfer function corresponding to the information and the sound source position information.

  The head-related transfer function selected by the head-related transfer function selecting means 6 is input to the convolution calculating means 3, and the convolution calculating means 3 performs an operation for convolving this head-related transfer function with the acoustic signal output from the sound source 1. This convolution operation means 3 can be realized by a program operating on dedicated hardware or a general-purpose computer.

  The acoustic signal in which the head-related transfer function is convoluted is output through the output means 7, drives the electroacoustic conversion means 8 such as headphones worn by the listener 9, is transmitted to the listener 9, and is transmitted to the listener 9. Gives a highly realistic sound field. The output unit 7 executes, for example, DA conversion processing or power amplification processing.

The processing procedure of the method for reproducing a highly realistic sound field according to the present invention will be described with reference to FIG.
In step SP1, information (listener information) for specifying the listener is set.
In step SP2, an acoustic signal is output from the sound source 1. The acoustic signal is divided into frames and output in consideration of convolution calculation processing and the like.
In step SP3, sound source position information attached to the acoustic signal output from the sound source 1 is selected.
In step SP4, a head-related transfer function corresponding to the listener and the sound source position is selected from the database 5.
In step SP5, an operation of convolving the head-related transfer function extracted from the database 5 with the acoustic signal is executed.
In step SP6, the output means 7 performs output processing.
In step SP7, it is determined whether or not the acoustic signal is finished. If the sound signal has not ended, the process returns to step SP2, the sound signal of the next frame is output, and steps SP3 to SP7 are repeated. When the end of the acoustic signal is detected in step SP7, the high realistic sound field reproduction processing program ends.

  Next, a method for creating the head related transfer function database 5 in consideration of the properties of the auditory filter proposed in claim 3 of the present invention will be described with reference to FIGS.

FIG. 3 shows the internal state of a conventional head related transfer function database. The conventional head-related transfer function database includes a head-related transfer function H _A1 for the direction 1 of the listener A, a head-related transfer function H _A2 for the direction 2, and a head-related transfer function H for the direction 3 for each of the listeners A, B, C. _A3 ..., A head related transfer function H _B1 with respect to direction 1 of listener B, a head related transfer function H _B2 with respect to direction ₂ , a head related transfer function H _{B3 with} respect to direction 3, and the like. Each of these head related transfer functions H _{A1 to} H _A3 ... And H _{B1 to} H _B3 . The measurement is generally performed by frequency analysis of each frequency component of the acoustic signal that is heard by the listeners A, B... By fast Fourier transform (FFT).

FIG. 4 conceptually shows a method of simplifying the head-related transfer function H _A1 obtained by measurement to the head-related transfer function H ′ _A1 in consideration of the properties of the auditory filter. A shown in FIG. 4 indicates a head-related transfer function H _A1 obtained by measurement. B represents a head-related transfer function H ′ _A1 in consideration of the properties of the auditory filter. The auditory filter has the properties of a plurality of bandpass filters as shown in FIG. 4C and is converted into a simplified head-related transfer function H ′ _A1 through this group of bandpass filters. As shown in FIG. 4D, the bandwidths of the plurality of bandpass filters exhibit characteristics that gradually become wider than the bandwidth on the low frequency side as it goes to the high frequency range.

  Here, it has become clear that the human auditory binaural processing system processes acoustic information separately for each frequency by a band-pass filter group, and this band-pass filter group is generally called an auditory filter. As a whole, the auditory filter is a band-pass filter group having a property that the width is small in the low band and the width is large in the high band. The acoustic signal to which the filter corresponding to the auditory filter is applied does not lose information by the auditory filter in the process of the listener's auditory processing system. On the contrary, even if the signal before passing through the auditory filter is given to the listener 9 as in the existing system, it is processed by the auditory filter in the auditory processing system and can be heard in the same manner as the signal that has passed through the auditory filter in advance. . Therefore, it is sufficient to have a database based on the characteristics of the head-related transfer function after being simplified through the auditory filter. The nature of this auditory filter has already been determined by experiment.

The band width erb of each bandpass filter constituting the auditory filter is the center frequency f _c (Hz) of each bandpass filter. Erb = 24.7 × (0.00437 × f _c +1) (1)
Is approximated by Center frequency f _c of the band-pass filters may be arbitrary, bandwidth erb is determined when determining the center frequency.

を計算して（１）式が求められる。 (1) As the process of deriving the equation, first the frequency f _c with a head related transfer function H obtained in FFT (f),
f _min = f _c − (1/2) * (24.7 (0.00437 f _c +1))
f _max = f _c + (1/2) * (24.7 (0.00437 f _c +1))
Seeking

Is calculated to obtain equation (1).

For example, FIG. 3 shows the result of calculation using the frequency used in each FFT as the center frequency when 512-point FFT is performed at a sampling frequency of 48 kHz. It is expressed as the center frequency f _{cn of the} nth bandpass filter from the lower frequency. At this time, the band width erb _n of this band pass filter is 24.7 × (0.00437 × f _cn +1), and the gain g _n of this band pass filter is g _n = Σ _{k = i} ^j h _k / (j -I + 1) (2)
It becomes. Here, h _k represents the gain at the k-th FFT point. i and j are f _cn −erb _n / 2 < f _ci and f _cj < f _cn + erb _n / 2 (3)
In other words, the expression (2) sets the average of the gain h _k as the gain gn of the _n-th bandpass filter for the range of the width erb _n with the center frequency f _cn as the center. Represents. An example of calculation around 10 kHz is shown as an example. When n is 110, the center frequency _fc110 is 10.3 kHz, and the bandwidth erb ₁₁₀ is 1138 Hz according to the equation (1), and this range includes eleven FFT frequencies from 107 to 117 according to the equation (3). The average value 44.4 dB of the 107th to 117th FFT values h _{107 to} h ₁₁₇ is the gain g _{110 of the 110th} bandpass filter. Through the above procedure, the head-related transfer function (FIG. 4B) considering the properties of the auditory filter applies the filter group (FIG. 4C) expressing the properties of the auditory filter to the head-related transfer function (FIG. 4A) obtained by measurement. Created as

FIG. 5 shows a flowchart for explaining the processing procedure of the head related transfer function database creation method according to the present invention.
In step SP1, the head-related transfer function is measured.
In step SP2, the center frequency f _cn of an arbitrary band pass filter of the band pass filter group expressing the characteristics of the auditory filter is set.
The bandwidth erb _n of the bandpass filter specified by the center frequency f _cn set in step SP3 is obtained.
In step SP4, a frequency component (FFT frequency component) included in the band 6 of the band pass filter is selected from the head related transfer function database obtained by measurement, and the average value of the gain is obtained.
In step SP5, the average value of the gain obtained in step SP4 is set as the gain of the bandpass filter, and the center frequency f _cn obtained together with this gain value is stored in the database 5 as an element of the head-related transfer function considering the characteristics of the auditory filter. Store.
In step SP6, it is determined whether or not the gains of all bandpass filters have been determined. If not, the process returns to step SP2, and steps SP2 to SP5 are repeated. When the gains of all bandpass filters are determined, the database creation program ends.

6 and 7 show the difference in characteristics between the head-related transfer function created in the present invention and the conventional head-related transfer function.
FIG. 6 shows an example of the frequency characteristic of the head-related transfer function in consideration of the properties of the auditory filter, and FIG. 7 shows the frequency characteristic of the head-related transfer function obtained by the measurement from which it was based. In both figures, the horizontal axis represents frequency, and the vertical axis represents relative level gain (unit: dB). The portions enclosed by circles A and B are the corresponding data. As is clear from FIG. 6, the frequency characteristics of the head related transfer function considering the properties of the auditory filter used in the present invention are reduced to some extent by the decrease in the portion that changes rapidly depending on the frequency, and the presence or absence of large recesses and the depth. You can see that As the filter group expressing the properties of the auditory filter in FIG. 6, the expression by a band pass filter group based on the equivalent rectangular bandwidth is used.

  The high realistic sound field reproduction method according to the present invention is suitable for sound reproduction using headphones, and can be applied to, for example, a video conference apparatus to give a highly realistic sound to a listener.

The block diagram for demonstrating one Example of the high realistic sound field reproduction apparatus by this invention. The flowchart for demonstrating the procedure of the highly realistic sound field reproduction method by this invention. The figure for demonstrating the mode inside the conventional head-related transfer function database. The figure for demonstrating the head-related transfer function database preparation method by this invention. The flowchart for demonstrating the procedure of the head-related transfer function database preparation method by this invention. The graph for demonstrating the frequency characteristic of the head related transfer function created with the head related transfer function database creation method by this invention. The graph for demonstrating the frequency characteristic of the conventional head-related transfer function.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Sound source 2 Sound source position information selection means 3 Convolution calculation means 4 Listener information setting means 5 Head-related transfer function database 6 Head-related transfer function selection means 7 Output means 8 Electroacoustic conversion means 9 Audience

Claims (6)

Simplify the head-related transfer function using a group of bandpass filters that have the properties of auditory filters,
A high realistic sound field reproduction method using a head related transfer function database created by holding the simplified head related transfer function,
Based on the listener information and the sound source position information obtained from the sound signal emitted from the sound source, a simplified head related transfer function is selected from the head related transfer function database, and the selected simplified head related transfer function is A method for reproducing a highly realistic sound field, wherein the sound signal is folded into an acoustic signal.   2. The high realistic sound field reproduction method according to claim 1, wherein a signal obtained as a result of convolving the selected simplified head-related transfer function with the acoustic signal is subjected to electroacoustic conversion. Sound field playback method.   2. The high realistic sound field reproduction method according to claim 1, wherein the band pass filters constituting the band pass filter group gradually increase in bandwidth as the center frequency becomes higher. Sound field playback method.   2. The high realistic sound field reproduction method according to claim 1, wherein an average value of gain is obtained for each frequency component included in a band of the bandpass filter, and the average value of the gain is used to obtain a head related transfer function. A highly realistic sound field reproduction method characterized by simplifying the sound.   5. A method of creating a head related transfer function database used in any one of the high realistic sound field reproduction methods according to claim 1, wherein a bandwidth of the auditory filter is narrower on a low frequency side than on a high frequency side. The characteristics of the bandpass filter group that gradually becomes wider than the bandwidth on the low frequency side on the high frequency side, obtain the average value of the gain of the frequency component included in the band of each bandpass filter of this bandpass filter group, A head-related transfer function database creation method characterized by storing a simplified head-related transfer function using the average value of gain as an element to create a head-related transfer function database. Sound source,
Sound source position information selection means for selecting sound source position information from the acoustic signal output by the sound source;
Listener information setting means for setting listener information;
A head-related transfer function database storing a head-related transfer function classified according to the listener and for each sound source position and simplified in consideration of the properties of the auditory filter;
A head-related transfer function selecting means for selecting a corresponding head-related transfer function from the head-related transfer function database according to the sound source position information selected by the sound source position information selecting means and the listener information set by the listener information setting means;
A convolution operation means for performing a convolution operation processing on the acoustic signal output from the sound source with the head-related transfer function selected by the head-related transfer function selection means;
An output means for outputting an acoustic signal in which the head-related transfer function is convoluted by the convolution operation means to the electroacoustic conversion means;
A highly realistic sound field reproduction apparatus characterized by comprising:

Images