KR20060124443A - Sound source localization method using head related transfer function database - Google Patents

Abstract

A sound source localization method using a head related transfer function database is provided to accurately locate the sound source by simultaneously estimating horizontal and vertical angles of the sound source in a 3D space. Sound signals are acquired by two microphones while varying the location of a sound source in a 3D space(S1). A database of HRTF(Head Related Transfer Function) of the sound signals is formed in a predetermined range of an audible frequency range through a signal process scheme(S2). A sound signal generated by the sound source to be tracked is acquired by using two microphones(S3). The phase difference between the signals is calculated within the predetermined range(S4). The difference between the phase difference of the stored HRTF functions and that of the acquired sound signals is acquired, and the result is multiplied by the square of a coherence function. Then, the result is integrated over the predetermined range to obtain a phase scale(S5). The position of the sound source having a phase scale relatively smaller than the other phase scales is determined as a target sound source position(S6).

Description

Sound source location estimation method using head transfer function database {Sound source localization method using Head Related Transfer Function database}

1A is a conceptual diagram of a conventional sound source position estimation method, and FIG. 1B is a conceptual diagram of a conical conical surface.

2 is a conceptual diagram of a head transfer function.

3 is an experimental data showing the HRTF size of both ears according to the horizontal angle change when the altitude angle is 0 ° for a sound source at a distance of 1m.

4 is an experimental data showing the HRTF size of the left ear according to the change of the altitude angle when the horizontal angle is 30 ° for the sound source at a distance of 1m.

FIG. 5A is a phase scale calculated using HRTF data of a sound source according to a horizontal angle change at an elevation angle of 0 °, and FIG. 5B is a phase scale calculated in consideration of a free sound field.

6A is a phase scale calculated using HRTF data according to a change in elevation angle when the horizontal angle is 30 °, and FIG. 6B is a size scale. 6C is a phase scale calculated using HRTF data according to a change in elevation angle when the horizontal angle is 60 °, and FIG. 6D is a size scale.

7 is a flowchart illustrating a sound source position estimation method using a head transfer function database according to an embodiment of the present invention.

<Description of main parts of drawing>

10: sound waves,

20: microphone,

30: conical conical surface,

40: sound source,

50: platform,

54: first microphone,

55: second microphone,

100: flow chart.

The present invention provides a first step of acquiring a signal for constructing a database, a second step of constructing a database, a third step of acquiring a sound wave signal generated by a tracked sound source through a microphone, and a phase difference and / or magnitude ratio of these signals. And a fifth step of calculating a phase scale and / or a size scale, and a sixth step of estimating a track target sound source position. The present invention relates to a sound source position estimation method using a head transfer function database. .

사이의 관계는 다음과 같다.1A is a conceptual diagram of a conventional sound source position estimation method. As shown in FIG. 1A, the prior art in this field is a method of estimating the position of the sound source in three-dimensional space from the delay time of the sound wave 10 signal entering each microphone 20 assuming a free sound field. . At this time, the position angle and delay time of sound source

The relationship between

Where c is the speed of sound, d is the delay distance, 2a is the distance between the two microphones, and pi is the angle at which the sound source exists. Therefore, by estimating the delay time, the existence angle of the sound source can be known. In this case, the generalized cross-correlation method (GCC) is widely used to estimate the delay time between the output signals of the microphones 20. This is a method of estimating the delay time by obtaining a correlation of the sound wave 10 signals input to each microphone 20 to obtain a time delay having a maximum value.

This prior art is a method that does not take into account the shape of the platform on which the microphone 20 is mounted. However, the target platform such as a real robot has a specific shape, and the reflection, diffraction, scattering, etc. occur due to the shape of the platform until the sound wave 10 radiated from the sound source reaches each microphone 20. Therefore, these causes cause distortion of phase and magnitude. When assuming only a free sound field without considering the shape of a platform as in the prior art, accurate sound source position estimation may not occur.

 In addition, since the location of the sound source having the same delay time is numerous in the three-dimensional space, a cone of confusion 30 as shown in FIG. 1B is generated. In other words, when estimating the position of the sound source based only on the delay time difference, there is a problem in that the front-rear separation and the upper-low classification cannot be made. That is, there is a problem in that only two microphones 20 cannot simultaneously estimate the horizontal angle and the altitude angle at which the sound source exists.

The object of the present invention is to measure the head transfer function in advance and use it as a database, and then use it to intelligently estimate the horizontal and altitude angles of sound sources in a three-dimensional space. It is to provide a sound source position estimation method that can be utilized.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings such that those skilled in the art can easily implement the present invention.

In the sound source position estimation method using the head transfer function database according to the present invention, a first step of acquiring a signal for constructing a database, a second step of constructing a database by processing a signal acquired in the first step, and a target sound source are generated. A third step of acquiring a sound wave signal through a microphone, a fourth step of processing the signals acquired in the third step to calculate a phase difference and / or a magnitude ratio, a fifth step of calculating a phase scale and / or a magnitude scale, And a sixth step of estimating the track target sound source position.

2 is a conceptual diagram of a head transfer function. As shown in FIG. 2, sound waves generated from the sound source 40 reach the head center point of the platform 50, two measurement points, the first microphone call phone 54 and the second microphone 55. The definition of the head related transfer function is as follows.

,

,

Where P _S is the sound pressure at the sound source, P _ff is the sound pressure at the head center point, and P _S is the sound pressure at the measurement point. That is, the HRTF is when the sound wave emitted from the sound source 40 reaches the first and second microphones 54 and 55 of the platform 50 and there is no shape of the platform 50 when the platform 50 is present. The sound wave radiated from the sound source 40 reaches the head center position of the platform 50 and is a ratio of free-filed pressure. Therefore, the HRTF reflects the influence of reflection, diffraction, scattering, etc. according to the shape of the platform 50. The HRTF varies depending on the position where the sound source 40 exists, that is, the horizontal angle, the altitude angle, and the distance.

3 is experimental data showing the HRTF size of both ears according to the horizontal angle change at an altitude of 0 ° for a sound source at a distance of 1 m. 4 is experimental data showing the HRTF size of the left ear according to the elevation angle change at a horizontal angle of 30 ° for a sound source at a distance of 1m. That is, FIG. 3 and FIG. 4 show HRTF measurement results according to changes in the horizontal angle and the elevation angle of the sound source in the anechoic chamber with respect to the platform (dummy head) 50. The vertical axis is frequency and the horizontal axis is the size of HRTF.

As shown in Figures 3 and 4, HRTF has a property that depends on the horizontal angle and the elevation angle. In the present invention, after measuring these characteristics in advance and making a database, the position of the sound source is estimated by comparing the phase difference and magnitude ratio of the signal actually input to each microphone with the HRTF database. In the present invention, the phase scale and the size scale are used for the quantitative comparison between the measured result and the retained HRTF data.

,

,

은 오른쪽 마이크로폰과 왼쪽 마이크로폰 출력신호의 코히어런스 함수(coherence function),

는 오른쪽 머리전달함수와 왼쪽 머리전달함수의 위상차,

는 오른쪽 머리전달함수와 왼쪽 머리전달함수의 크기비,

는 추적 대상 음원에 대한 오른쪽 마이크로폰과 왼쪽 마이크로폰 출력신호의 위상차,

: 추적 대상 음원에 대한 오른쪽 마이크로폰과 왼쪽 마이크로폰 출력신호의 크기비이다. 적분 기호는 추정하고자 하는 음원의 주파수 대역만을 선택하여 더하는 것을 의미한다. 한예로, 사람의 음성을 추정하는 경우 음성주파수 대역인 300Hz~3.4kHz 대역의 차이만을 더하는 것이 된다. 이렇게 함으로써 원치 않는 주파수대역의 신호를 배제하고 음원의 위치를 추정할 수 있게 된다. 또한, 코히어런스 함수는 두 신호간의 일관성 혹은 데이터의 신뢰성을 나타내는 척도로써, 측정 신호의 위상차(크기비)와 HRTF 데이터의 위상차(크기비)간의 차이에 상관 함수를 곱해줌으로 인해 신뢰성 있는 정보만을 이용하게 되는 결과를 가져온다. here,

Is the coherence function of the right and left microphone output signals,

Is the phase difference between the right head transfer function and the left head transfer function,

Is the size ratio of the right head transfer function and the left head transfer function,

Is the phase difference between the right and left microphone output signals for the sound source to be tracked,

: The ratio of the right and left microphone output signals to the sound source to be tracked. The integral symbol means selecting and adding only a frequency band of a sound source to be estimated. For example, when estimating a human voice, only the difference between the 300 Hz and 3.4 kHz bands of the voice frequency band is added. This makes it possible to estimate the position of the sound source, excluding signals in unwanted frequency bands. In addition, the coherence function is a measure of the consistency or reliability of data between two signals. This results in using the bay.

If the HRTF data corresponding to the position of the sound source to be tracked is used, the difference between the phase difference (magnitude ratio) between the measured signals and the phase difference (magnitude ratio) using the HRTF data is reduced, so that the phase of the HRTF database by position of the sound source is retained We can estimate the position of the sound source (horizontal angle, elevation angle) by finding the HRTF data set that minimizes the phase scale and the size scale by obtaining the scale and the size scale respectively.

FIG. 5A is a phase scale calculated using HRTF data of a sound source according to a horizontal angle change at an elevation angle of 0 °, and FIG. 5B is a phase scale calculated in consideration of a free sound field. The results of assuming a free sound field for comparison are also shown. The horizontal axis of the first figure of FIG. 5A is a phase scale, one axis of the vertical axis is an HRTF angle, and the other axis is an actual angle. The horizontal axis in the second figure is the HRTF angle and the vertical axis is the actual angle. B is also the same as above.

5A and 5B, it can be seen that the value of the phase scale is less when the head transfer function near the actual sound source position is used. That is, when the phase scale is calculated by changing the HRTF angle with respect to the actual sound source angle, the phase scale value is the smallest when the angle of the head transfer function and the actual sound source are the same. This shows that the position of the sound source can be estimated by obtaining the angle of the head transfer function that minimizes the phase scale. However, even at an angle symmetrical with the actual sound source position, the value of the phase scale is small, which means that the front-back confusion phenomenon occurs that the front and back distinctions are blurred by using only the phase difference. As with humans, the result is that the front and rear of the dummy head are almost symmetrical, and this phenomenon also occurs when a free-field is assumed. However, as shown in the above results, the estimation of the horizontal angle in consideration of the head transfer function improves the sound source position estimation performance rather than simply estimating the horizontal angle in the free sound field. This is especially true in the vicinity of the medicine, where the sound source is biased to one ear, because when the sound source is in this position, the sound waves radiated from the sound source hit the head of the dummy head and travel the longest to the opposite microphone.

6A is a phase scale calculated using HRTF data according to a change in elevation angle when the horizontal angle is 30 °, and FIG. 6B is a size scale. 6C is a phase scale calculated using HRTF data according to a change in elevation angle when the horizontal angle is 60 °, and FIG. 6D is a size scale. 6A, 6B, 6C, and 6D, the abscissa axis represents the HRTF angle, and the ordinate axis represents the actual angle.

As shown in FIG. 6A and FIG. 6C, when the phase scale is used, the front-and-back distinction occurs in the horizontal angle estimation, so that the phenomenon that cannot be distinguished from the upper and lower sides also occurs in the elevation angle estimation. This can be seen from the results. However, the reason why it is not obvious as in the case of the horizontal angle change is that the dummy head is not symmetrical with respect to the microphone, just as a person is not symmetrical with respect to both ears. Considering the outer ear, nose and head shape of the dummy head, the front and rear of the dummy head are almost symmetrical, but the upper and lower bodies are not symmetrical due to the torso, and the influence is contained in the size information of the head transfer function. have.

As shown in Figs. 6B and 6D, it is better to use the size scale for the elevation angle estimation. In the conventional free sound field, the delay time of each microphone cannot be used to estimate the horizontal and altitude angles as described above.

7 is a flowchart illustrating a sound source position estimation method using a head transfer function database according to an embodiment of the present invention. As shown in FIG. 7, the sound source position estimation method according to the present invention processes a signal acquired in a first step S1 and a first step S1 to acquire a signal for constructing a database. A second step S2, a third step S3 of acquiring a sound wave signal generated by the sound source to be tracked through the microphone, and a fourth step of processing the signals acquired in the third step to calculate a phase difference and / or a magnitude ratio (S4), a fifth step (S5) of calculating a phase scale and / or a size scale, and a sixth step (S6) of estimating a track target sound source position.

In the first step S1, the position (eg, horizontal angle, elevation angle) of the sound source 40 in a three-dimensional space (eg, an anechoic chamber) with respect to the platform (robot, dummy head, etc.) 50 to be applied. Acquisition of the sound wave signals through the two first and second microphones 54 and 55 while different.

In the second step S2, a database is constructed by obtaining a head transfer function of signals acquired in the first step S1 in a predetermined range of the audible frequency bands by signal processing.

In the third step S3, the sound wave signals generated by the sound source to be tracked are acquired through two microphones.

In the fourth step S4, phase difference, magnitude ratio, or phase difference and magnitude ratio between the signals acquired in the third step are calculated in a predetermined range of the audible frequency band by signal processing. In the fourth step S4, a coherence function, an auto-spectral density function, and a cross-spectral density function can be found, and the phase difference is used using these functions. Or it is preferable to obtain the size ratio.

In the fifth step S5, the square of the coherence function is the square of the difference between the phase differences of the head transfer function for each sound source position stored in the database in the second step S2 and the phase difference obtained in the fourth step S4. The phase scale is obtained by integrating the value multiplied by in a range of audible frequency bands. Further, in the fifth step, the square of the coherence function is the square of the difference between the size ratios of the head transfer function for each sound source position stored in the database in the second step S2 and the size ratio obtained in the fourth step S4. The magnitude scale can be obtained by integrating the value multiplied by in a range of audible frequencies. Alternatively, both phase and magnitude scales can be obtained.

In the sixth step S6, the sound source position in the database having the phase scale obtained in the fifth step S5, the other phase scales among the size scales, the phase scale relatively smaller than the size scales, and the size scale is used as the target sound source position. Estimate.

The operation of the sound source position estimation method using the head transfer function database according to the present invention having the above configuration will be described.

First, the results of the experiments shown in FIGS. 2 and 3 used conventional signal amplifiers, microphones, and frequency analyzers for acquiring and processing sound wave signals, but alternative equipment and frequency analysis software capable of signal acquisition and processing may be used. have.

According to the present invention, the phase scale and the size scale obtained by using the head transfer function measured at the location of the track target sound source are the sound field (for example, infinite plane) and the sound source position tracking target of the head transfer function construction process. Assuming that this sound field (e.g., infinite plane) is exactly the same and there is no error in signal acquisition and processing, it is theoretically zero. The small phase and magnitude scales mean that the tracked sound source is near the sound source location when the head transfer function is measured. Even if the difference in sound field or small error in signal processing is taken into consideration, the function of these measures is sufficiently maintained.

Although the invention has been described primarily with reference to the above-described embodiments, many other possible modifications and variations can be made without departing from the spirit and scope of the invention. For example, using the absolute value of the difference, not the square value of the difference between the phase differences of the head transfer function for each sound source position stored in the database in the second step and the phase difference obtained in the fourth step while obtaining the phase scale. It will be a change that can be easily derived by those skilled in the art from the above-described embodiment.

When using the sound source position tracking method using the head transfer function according to the present invention, it is possible to estimate the horizontal angle and the altitude angle of the sound source in the three-dimensional space at the same time. Accordingly, a sound source position estimation method that can be utilized for the intelligent robot industry, security CCTV, teleconferencing camera, or the like is provided. In addition, the use of the phase scale and the size scale solves the above-mentioned problems of the prior art and enables more accurate sound source position estimation than the prior art.

The scope of the above-described invention is defined in the following claims, not bound by the description in the text of the specification, all modifications and variations belonging to the equivalent scope of the claims will fall within the scope of the invention.

Claims (6)

In the sound source position tracking method, A first step of acquiring a sound wave signal through two microphones while changing a position of a sound source in a three-dimensional space with respect to the platform 50 to be applied; A second step of constructing a database by obtaining a head transfer function of the sound wave signals in a predetermined range of an audible frequency band by signal processing; A third step of acquiring a sound wave signal generated by the sound source to be tracked through two microphones; A fourth step of calculating a phase difference between the signals acquired in the third step in a predetermined range of the audible frequency bands by signal processing; The phase scale is obtained by integrating the square value of the difference between the phase differences of the head transfer function for each sound source location stored in the database and the phase difference obtained in the fourth step by the square of the coherence function in a range of audio frequencies. Obtaining a fifth step; And A sixth step of estimating a sound source position in a database having a phase scale relatively smaller than other phase scales among the phase scales as a target sound source position; and tracking the sound source position using the head transfer function Way. The method of claim 1, The fourth step is to obtain a coherence function, auto-spectral density function, cross-spectral density function further characterized in that the sound source position tracking method using the head transfer function. In the sound source position tracking method, A first step of acquiring sound wave signals through two microphones while varying the position of the sound source 40 in a three-dimensional space with respect to the platform 50 to be applied; A second step of constructing a database by obtaining a head transfer function of the sound wave signals in a predetermined range of an audible frequency band by signal processing; A third step of acquiring a sound wave signal generated by the sound source to be tracked through two microphones; A fourth step of calculating a magnitude ratio between the signals acquired in the third step in a predetermined range of the audible frequency bands by signal processing; The magnitude scale by integrating a square value of the difference between the size ratios of the head transfer function for each sound source location and the size ratio obtained in the fourth step by the square of the coherence function in a range of audible frequency bands stored in the database. Obtaining a fifth step; And A sixth step of estimating a location of a sound source on a database having a size scale relatively smaller than other size scales among the size scales as a target sound source position; and tracking the sound source position using the head transfer function Way. The method of claim 3, wherein The fourth step is to obtain a coherence function, auto-spectral density function, cross-spectral density function further characterized in that the sound source position tracking method using the head transfer function. The method of claim 3, wherein The fourth step may further include calculating a phase difference between the signals acquired in the third step in a predetermined range of the audible frequency band by signal processing. In the fifth step, the square value of the difference between the phase differences of the head transfer function for each sound source position stored in the database in the second step and the phase difference obtained in the fourth step is multiplied by the square of the coherence function. Obtaining a phase scale by integrating in an audible frequency band of the sound source position tracking method using a head transfer function further comprising. The method of claim 5, The fourth step is to obtain a coherence function, auto-spectral density function, cross-spectral density function further characterized in that the sound source position tracking method using the head transfer function.

Images