KR100612616B1 - Signal-to-Noise Ratio Estimation Method Using Zero Crossing Point and Sound Source Direction Detection Method - Google Patents

Abstract

The present invention relates to a method for estimating a signal-to-noise ratio and detecting a sound source direction by using a variance of zero crossing time differences of voice signals received from two sensors.

A sound source direction detection method using a zero crossing point according to the present invention includes the steps of receiving a signal output from the same sound source using the two sensors and separating the received signal by channel; ZCPA coding to obtain a zero crossing point and a maximum value by ZCPA coding each channel signal separated in frequency for each channel; An ITD calculation step of calculating an ITD value by using the zero crossing point and the highest value for each channel obtained in the ZCPA coding step; A variance calculation step of dividing each channel signal into a plurality of windows including the same number of zero crossing points and obtaining a variance of ITD values for each channel and window; A signal-to-noise ratio calculation step of calculating a signal-to-noise ratio using the variance of the ITD value for each channel and window and the center frequency of the channel; A reliability enhancement step of extracting an ITD value whose signal-to-noise ratio is greater than a threshold; Obtaining a horizontal angle histogram of the ITD value using the signal-to-noise ratio as a weight, and determining a maximum value of the horizontal angle histogram as a spatial position of the sound source.

Source direction detection, zero crossing, ZCPA coding, time delay, signal to noise ratio estimation

Description

The signal-to-noise ratio estimation method and sound source localization method based on zero-crossings}             

1 is a block diagram illustrating a signal-to-noise ratio estimation method and a sound source direction detection method using a zero crossing point according to an embodiment of the present invention;

2 is a conceptual diagram of ZCPA coding for an i th channel signal;

3 is a view for explaining the movement of the zero crossing when the noise is mixed,

4 is a view summarizing the direction estimation results of the direction detection method by cross-correlation and the direction detection method according to the present invention when the sound source is located at 20 degrees and 40 degrees in various background noises.

5 is a horizontal angle histogram of a direction detection method based on cross correlation and a horizontal angle histogram of a direction detection method according to the present invention when a sound source is located at -10 degrees, 0 degrees, 10 degrees, and 40 degrees, respectively, in various background noises. It is a figure compared.

The present invention relates to a method for estimating a signal-to-noise ratio using a zero crossing point and a method for detecting a sound source direction. More specifically, the present invention relates to a signal-to-noise ratio using a dispersion value of zero-crossing time differences of voice signals generated by two sensors. It is about how to detect.

The direction detection technology of sound sources refers to the direction of sound sources by using time delays and intensity differences of voice signals generated from two or more sensors according to spatial positions of the sound sources. This technique is applied to the automatic speech recognition system using two ears, which is used to focus attention on a sound source in a specific direction in noise or to separate each sound source by using direction information when several sound sources are mixed. In addition, the direction detection technique of this sound source is applied to adaptive beamforming technique and directional filtering technique, and is used to remove active directional noise.

Conventional sound direction detection technology applies the filter bank signal connected to the two sensors to the time delay, finds the cross-correlation value for all time delays, and finds the time delay with the highest cross correlation value. Find. However, this conventional method is difficult to implement because many calculations are required to cross-correlate the speech signal for all time delays. In addition, a long time window is advantageous for finding the correct time delay of a sound source, but a short time window is advantageous for detecting the direction of a local sound source. This conventional method has a fixed time window for calculating a cross-correlation value. Because of this problem, it is difficult to determine the size of the window. Therefore, in general, the cross-correlation-based direction detection technique of the sound source is less accurate when the noise is mixed, when there are a plurality of sound sources, there is a problem that severe interference between the sound source.

The direction detection technique of a sound source modeling two ears is mostly described in L. Jeffress's paper [A place theory of sound localization, J. Comp. Physiology and Psychology, 41: 35-39, 1948, based on cross-correlation using time delay.

S. R. Stern and H. Colburn [Theory of binaural interaction based on auditory-nerve data: IV. A model of subjective lateral position, J. Acoust. Soc. Amer., 64 (1): 127-140, 1978] quantitatively obtain the interaural time difference (ITD) and the interaural intensity difference (IID), which are representative directions of representative sound sources, based on the cross-correlation presented by Jeffrey.

In addition, S. A. Shamma, N. Shen and P. Gopalaswamy have also published a paper in Stereausis: binaural processing without neural delays, J. Acoust. Soc. Amer., 86 (3): 989-1006, 1989] proposes a technique for obtaining direction information of a sound source using a time delay generated in an analog electronic circuit without artificially using a time delay.

In addition, J. Breebaart, S. Par and A. Kohlausch [Binaural processing model based on contralateral inhibition. I. Model structure, J. Acoust. Soc. Amer., 110 (2): 1074-1088, 2001, describes a side-tone phenomenon by applying two principles of suppression and excitation to the sound source signals of both ears, deviating from the cross-correlation based approach.

The beamforming technique is a technique of obtaining a signal from a desired direction by adding a weight by delaying and then weighting a plurality of antennas with different time values. Griffith (L. J. Griffiths) reported in A simple adaptive algorithm for real-time processing in antenna arrarys, Proc. IEEE, 57: 1696-1704, 1969], when knowing the cross-correlation value between the desired signal and the weight value in advance, obtains the weight value from the least mean square value of the error signal without the aid of the pilot signal. have.

As a directional filtering technique, T. Wittkop [Two-channel noise reduction algorithms motivated by models of binaural interaction, Ph.D. thesis Univ. Oldenburg, 2001], the frequency signals are separated by STFT (Short Time Fourier Transform) for two ears, cross-correlated to obtain direction information, and the signals are given different weights according to the direction and then I-STFT ( Inverse-STFT) to restore the sound source in a specific direction. Although this technique is used in hearing aids for the hearing impaired, it has a high effect, but it is not only difficult to detect the exact direction of the sound source, but also difficult to experimentally obtain parameter values to make a weight function according to the direction.

The present invention has been made to solve the above problems of the prior art, and estimates the signal-to-noise ratio using the zero-crossing and peak amplitudes (ZCPA) of the output signal of the filter bank, and the direction of the sound source The purpose is to provide a method of detection.

Signal-to-noise ratio estimation method using a zero crossing point according to the present invention for achieving the above object, two sensors corresponding to the two ears of a person, a filter bank for frequency-dividing the signals received from the two sensors for each channel, In the signal-to-noise ratio estimation method using a zero crossing point in a device having a processor for calculating the signal-to-noise ratio included in each channel signal separated from the filter bank,

Receiving signals output from the same sound source using the two sensors and separating the received signals by channels;

A ZCPA coding step of zero-crossing and peak amplitudes (ZCPA) coding each channel signal separated by frequency for each channel to obtain a zero crossing point and a maximum value;

An ITD calculation step of obtaining an interaural time difference (ITD) value by using the zero crossing point and the highest value for each channel obtained in the ZCPA coding step;

A variance calculation step of dividing each channel signal into a plurality of windows including the same number of zero crossing points and obtaining a variance of ITD values for each channel and window;

And a signal-to-noise ratio calculation step of calculating a signal-to-noise ratio using the dispersion of the ITD value for each channel and each window and the center frequency of the channel.

Sound source direction detection method using a zero crossing point according to the present invention for achieving the above object, two sensors corresponding to two ears of a person receiving a signal output from the same sound source, and the signal received from the two sensors In the sound source direction detection method using a zero crossing point in the device having a filter bank for separating the frequency for each channel, and a processor for detecting the direction of the sound source using each channel signal separated in the filter bank,

Receiving signals output from the same sound source using the two sensors and frequency-receiving the received signals for each channel;

A ZCPA coding step of zero-crossing and peak amplitudes (ZCPA) coding each channel signal separated by frequency for each channel to obtain a zero crossing point and a maximum value;

An ITD calculation step of obtaining an interaural time difference (ITD) value by using the zero crossing point and the highest value for each channel obtained in the ZCPA coding step;

A variance calculation step of dividing each channel signal into a plurality of windows including the same number of zero crossing points and obtaining a variance of ITD values for each channel and window;

A signal-to-noise ratio calculation step of calculating a signal-to-noise ratio using the variance of the ITD value for each channel and window and the center frequency of the channel;

A reliability enhancement step of extracting an ITD value whose signal-to-noise ratio is greater than a threshold;

And a direction estimating step of obtaining a horizontal angle histogram of the ITD value using the signal-to-noise ratio as a weight and determining the maximum value of the horizontal angle histogram as a spatial position of the sound source.

According to the present invention, a signal output from the same sound source is received by two sensors corresponding to the two ears of the human, and the frequency separation for each channel after calculating the signal-to-noise ratio included in each channel signal, the signal-to-noise ratio using a zero crossing point A computer readable recording medium having recorded thereon a program for executing the estimation method is provided.
Signal-to-noise ratio estimation method using the zero crossing point,
A ZCPA coding step of zero-crossing and peak amplitudes (ZCPA) coding each channel signal separated by frequency for each channel to obtain a zero crossing point and a maximum value;
An ITD calculation step of obtaining an interaural time difference (ITD) value by using the zero crossing point and the highest value for each channel obtained in the ZCPA coding step;
A variance calculation step of dividing each channel signal into a plurality of windows including the same number of zero crossing points and obtaining a variance of ITD values for each channel and window;
And a signal-to-noise ratio estimation method including a signal-to-noise ratio calculation step of calculating a signal-to-noise ratio using the dispersion of the ITD value and the center frequency of the channel for each channel and window.

In addition, according to the present invention receives a signal output from the same sound source with two sensors corresponding to the two ears of the human, and the frequency separation for each channel and then using a channel signal to detect the direction of the sound source using a zero crossing point, Provided is a computer readable recording medium having recorded thereon a program for executing a sound source direction detecting method.
Sound source direction detection method using the zero crossing point,
A ZCPA coding step of zero-crossing and peak amplitudes (ZCPA) coding each channel signal separated by frequency for each channel to obtain a zero crossing point and a maximum value;
An ITD calculation step of obtaining an interaural time difference (ITD) value by using the zero crossing point and the highest value for each channel obtained in the ZCPA coding step;
A variance calculation step of dividing each channel signal into a plurality of windows including the same number of zero crossing points and obtaining a variance of ITD values for each channel and window;
A signal-to-noise ratio calculation step of calculating a signal-to-noise ratio using the variance of the ITD value for each channel and window and the center frequency of the channel;
A reliability enhancement step of extracting an ITD value whose signal-to-noise ratio is greater than a threshold;
Obtaining a horizontal angle histogram of the ITD value using the signal-to-noise ratio as a weight, and determining a maximum value of the horizontal angle histogram as a spatial position of the sound source.

Hereinafter, a method of estimating a signal-to-noise ratio using a zero crossing point and a sound source direction detection method will be described in detail with reference to the accompanying drawings.

The present invention estimates the signal-to-noise ratio included in the sound source by using a voice signal coding method using ZCPA (zero-crossing and peak amplitudes), and detects the direction of the sound source. The inventors of the present invention (D. S. Kim and S. Y. Lee and R. M. Kil) report in Audi proseccing of speech signals for robust speech recognition in real-world noidy environments, IEEE Trans. Speech and Audio Proc., 7 (1): 55-69, 1999] propose a speech signal coding method using ZCPA. The ZCPA voice signal coding method is a method created by the auditory system based on the method of coding the speech pattern of a sound source. Better recognition results in a noisy environment compared to the

1 is a block diagram illustrating a signal-to-noise ratio estimation method and a sound source direction detection method using a zero crossing point according to the present invention.

In order to detect the direction of the sound source θ _S , signals are received from two sensors (not shown) corresponding to two ears of a person, and the signals received from the sensors are channel-by-channel using two filter banks 111 and 112. Frequency separation. Each channel signal is ZCPA coded 121 and 122, and an interaural time difference (ITD) value and an interaural intensity difference (IID) value, which is representative direction information, is calculated from the ZCPA coding result. The signal-to-noise ratio (SNR) of the signal is calculated 141 using the calculated ITD value. An ITD value whose signal-to-noise ratio is greater than a threshold value is selected (151), weighted to the selected ITD value to generate a histogram (161), and then the maximum value is found from the histogram (171) to find the direction of the sound source.

Each step is explained in detail.

1. ZCPA Coding

First, a signal is received from two sensors. The received signal is a signal including interference noise and background noise in a sound source signal. The signal received from the sensor is frequency separated into each channel in the two filter banks 111 and 112.

Each frequency-separated channel signal is ZCPA coded, and the method of ZCPA coding is described in the paper published by the inventors of the present invention (DS Kim, SY Lee and RM Kil) as described above. in real-world noidy environments, IEEE Trans. Speech and Audio Proc., 7 (1): 55-69, 1999, which are described in detail herein.

에서 영교차점을 각각 t₁,t₂,...,t_N이라고 하면, ZCPA 코딩은 아래의 수학식 1과 같다.ZCPA coding of the speech signal represents the channel signal passing through the filter bank as the maximum value of the signal between the upstream zero crossing and the adjacent zero crossing. Expressing this formally, the i-th channel signal of the filterbank

If the zero crossing points at t ₁ , t ₂ , ..., t _N , respectively, ZCPA coding is given by Equation 1 below.

2 is a conceptual diagram of ZCPA coding for an i-th channel signal. That is, the maximum value of x _i (t) values in the interval from t _n-1 to t _n is output at the _n- th zero crossing point, and 0 is output in the remaining intervals.

,

라고 가정한다. 여기서, n(=1,2,...,N)과 m(=1,2,...,M)은 각각 왼쪽과 오른쪽 귀의 i 번째 채널에서 발생하는 영교차점을 나타낸다.Each of the i-channel ZCPA coded signals

,

Assume that Where n (= 1,2, ..., N) and m (= 1,2, ..., M) represent the zero crossings occurring in the i-th channel of the left and right ears, respectively.

2. ITD value and IID value calculation

을 연산한다. 즉, 왼쪽 채널의 n번째 영교차점을 기준으로 오른쪽 채널의 가장 가까운 k번째 영교차점을 구하고, 두 영교차점의 차이를 ITD값으로서 구한다.Once the ZCPA coding values of the channels occurring at both ears are obtained, the zero crossing points of both channels are applied to Equation 2 below to determine the ITD value.

Calculate That is, the kth zero crossing point of the right channel is determined based on the nth zero crossing point of the left channel, and the difference between the two zero crossing points is obtained as the ITD value.

Here, the closest kth zero crossing point of the right channel based on the nth zero crossing point of the left channel is obtained as in Equation 3 below.

은 주파수-시간 영역에서 음성신호의 i 번째 주파수 채널과 n 번째 영교차점 시간에 해당하는 ITD값이다. 그리고, 이때의 IID값

은 최고치값들로부터 아래의 수학식 4와 같이 구한다.ITD value

Is the ITD value corresponding to the i th frequency channel and n th zero crossing time of the voice signal in the frequency-time domain. And IID value at this time

Is obtained from Equation 4 below.

In Equation 2 and Equation 3, the ITD value is obtained by obtaining the closest kth zero crossing point of the right channel based on the nth zero crossing point of the left channel. However, in the case of a high frequency signal, the time delay value between the left channel and the right channel is calculated. Because of the short length of one wavelength, if the ITD value is obtained only at the nearest zero crossing point, the exact time delay between the two channels cannot be obtained. Therefore, the ITD value of the number considering the maximum time delay value that can occur according to the spatial position of the sound source and the length of one wavelength of the center frequency of the channel is obtained. That is, when the maximum time delay value is a times the length of one wavelength of the center frequency of the channel, 2a of the zero crossing points of the adjacent right channel are found based on an arbitrary zero crossing point of the left channel, and 2a ITD values as the difference value. Calculate

3. Signal to Noise Ratio Calculation

으로부터 s개의 영교차점을 시간 윈도우로 사용하여 해당 시간 윈도우 에서의 ITD값의 평균과 분산을 계산한다. 본 발명에서는 인간의 음향 인지 특성을 고려하여 21개의 영교차점을 시간 윈도우로 결정한다. 이렇게 영교차점의 개수를 이용하여 시간 윈도우를 가변하면 저주파 채널에서는 시간 윈도우가 넓어지고, 고주파 채널에서는 시간 윈도우가 좁아지기 때문에, 인간의 음향 인지 특성에 부합된다.ITD value thus obtained

Using s zero crossings as a time window, we compute the mean and variance of the ITD values in that time window. In the present invention, 21 zero crossing points are determined as time windows in consideration of human acoustic recognition characteristics. If the time window is changed using the number of zero crossing points, the time window is widened in the low frequency channel and narrowed in the high frequency channel, thereby meeting the human acoustic recognition characteristics.

은 잡음에 의해 왜곡되기 때문에 이 ITD값

를 이용하면 음성신호와 잡음의 섞인 비율 즉, 신호대잡음비를 추정할 수 있게 된다. 일반적으로

값이 환경적인 요소나 측정오차 등의 잡음으로 인해 왜곡되는 현상은 아래의 수학식 5와 같이 표현된다.First, we look at the process of obtaining the mean and the variance from the ITD value. ITD value

Since this is distorted by noise, this ITD value

By using, it is possible to estimate the mixed ratio of speech signal and noise, that is, the signal-to-noise ratio. Generally

The phenomenon in which the value is distorted due to noise such as environmental factors or measurement errors is expressed as in Equation 5 below.

와

는 잡음이 없을 때의 영교차점이고,

와

은 잡음에 의한 영교차점의 왜곡 정도로서, 평균과 분산이 각각 0과 인 독립균등분포(identically and independently distributed)를 갖는다고 가정한다. 음원의 공간적인 위치에 따라 두 귀에서 발생하는 신호의 시간지연을 Δ라고 하면 좌측의 잡음이 없을 때의 영교차점

은 아래의 수학식 6과 같이 표현된다.here,

Wow

Is the zero crossing in the absence of noise,

Wow

Is the degree of distortion of the zero crossing due to noise, and mean and variance are 0 and Suppose we have an independently and independently distributed. Δ is the time delay of the signal from both ears according to the spatial position of the sound source.

Is expressed by Equation 6 below.

와

는 서로 독립이고 평균값이 0이라 가정하면, ITD값

의 평균과 분산은 각각 Δ와 2

가 된다.Distortion degree of zero crossing point by noise

Wow

Are independent of each other and the average value is 0, the ITD value

The mean and variance of are Δ and 2

Becomes

에 대해 살펴본다.Where the variance of the IDT values in the noisy channel

Take a look at

The channel considered first is assumed to be an ideal bandpass filter with a center frequency of ω _c and a bandwidth of 2b. When the bandwidth of the filter is quite small, a signal such as Equation 7 below occurs in the i-th channel of the left and right ears.

과

은 평균과 분산이 각각 0과 1인 백색 잡음이고, A와 B는 각각 신호와 잡음의 세기에 해당한다. 위 수학식 7에서의 두 신호를 각각 ZCPA 코 딩하여

와

를 얻고, 그 평균과 분산을 구하면 각각 Δ와 2

가 된다. 그리고, 위 식은 영교차점에서 아래 수학식 8의 조건을 만족하며, 도 3은 잡음에 따른 영교차점의 이동을 도시한다.At this time,

and

Is white noise with mean and variance of 0 and 1, respectively, and A and B correspond to signal and noise strengths, respectively. ZCPA coding of two signals in Equation 7 above

Wow

If we find the mean and the variance,

Becomes The above equation satisfies the condition of Equation 8 below at the zero crossing point, and FIG. 3 shows the movement of the zero crossing point according to the noise.

Here, the random variables x _n and y _n are defined as in Equation 9 below.

의 함수 형태로 볼 수 있기 때문에

이며,

이다.

인 경우에는 -π/2 ≤ x_n ≤ +π/2 구간에서 영교차점이 발생하지 않을 수도 있기 때문에, 고려 대상에서 제외한다.At this time, random variable

Can be seen as a function of

Is,

to be.

In this case, since the zero crossing may not occur in the interval -π / 2 ≤ x _n ≤ + π / 2, it is excluded from consideration.

From the nonlinear relationship between the random variables x _n and y _n , the probability distribution function of the random variable y _n is obtained as in Equation 10 below.

Here, if r _n follows a normal distribution, x _n also follows a normal distribution, and the probability distribution function is summarized as in Equation 11 below.

In this case, the variance of the random variable x _n is obtained as in Equation 12 below.

The average of _{n n} is obtained from the condition of the region of the random variable y _n as shown in Equation 13 below.

Here, if y = sinθ, the average of y _n in Equation 13 can be expressed as Equation 14 below.

The average function of y _n given above is an odd function, and the variance of y _n at this time is calculated as in Equation 15 below.

In the above formula, the fact that σ _x ≪ π / 2 is generally used. Now, using the assumption that v (t _n ) is white noise, Equation 16 is derived.

The signal-to-noise ratio (SNR) of the signal passing through the i-th channel is expressed by Equation 17 below.

Therefore, the variance of the random variable y _n satisfies Equation 18 below.

Therefore, the variance of the random variable x _n is expressed by Equation 19 below.

의 분산 사이의 수학식 20과 같은 관계를 이용하여, ITD값

의 분산을 구하면 아래의 수학식 21과 같다.Now the variance and ITD of the random variable x _n

The ITD value is obtained by using the relationship

The variance of is calculated by Equation 21 below.

과 신호대잡음비(SNR)의 관계식은 아래의 수학식 22와 같다.Finally ITD value

And the signal-to-noise ratio (SNR) is expressed by Equation 22 below.

의 평균과 분산을 구하는데, 이 국지적 ITD값

의 평균과 분산은 아래의 수학식 23과 같다.In the present invention, using the equations derived from the above-described process, the local ITD value from the s zero crossing time window

Find the mean and the variance of

The mean and the variance of are as shown in Equation 23 below.

In this case, i (= 1, ..., P) corresponds to the channel number of the filter bank, j (= 1, ..., N-s + 1) corresponds to the number of zero crossing point. The signal-to-noise ratio SNR _i (j) at the i-th channel and the j-th zero crossing point is expressed by Equation 24 below.

Where ω _C is the center frequency of the channel.

The ITD value and IID value located in the center of each time window are set as the ITD value and IID value of the corresponding time window. This may be expressed as an equation (25) and (26). It is assumed that the number s of zero crossings included in each time window is odd.

4. Weighted histogram of ITD and IID values

In order to select highly reliable data among the ITD and IID values obtained by Equation 24, the ITD and IID values whose SNR estimates are larger than the threshold τ are taken. The signal-to-noise ratio (SNR) estimated value is used as a weight to generate a histogram of the ITD value or the ITD value and the IID value.

5. Direction detection

The maximum value of the histogram of the ITD and IID values is determined by the ITD and IID values generated according to the spatial position of the actual sound source.

[Experiment result]

In order to demonstrate the effect of the present invention, the following experimental environment is constructed. First, the sound source uses a head-related transfer-function database to generate a sound signal reaching both ears at a specific coordinate position in space. The wah-angle filter forms a gammatone filter bank having six channels between 0.2 kHz and 1.2 kHz. Since the ITD value corresponding to the azimuth angle of the sound source is different for each channel, the function relationship between the horizontal angle of the channel and the ITD value is examined for the HRTF database value using the tone pulse having the center frequency of the channel.

In this experimental environment, the direction detection method according to the present invention and the direction detection method according to the present invention are compared.

In the cross-direction detection method, the size of the time window for calculating cross-correlation for each channel is fixed to 20 msec, and the ITD value is calculated every 10 msec. The final direction for the sound source is then determined by the peak values in the horizontal angle histogram made from the calculated ITD values. Similarly, in the direction detection method according to the present invention, after calculating the ITD value from the zero crossing point for each channel, the time window size for SNR estimation is used as 21 zero crossing points. This time window size is associated with human cognitive abilities, using only short time information of 5-20 msec on high frequency channels, while using longer time information of 50-140 msec on low frequency channels. That is, the direction detection method by cross correlation uses the same sized time window, whereas the direction detection method according to the present invention varies the size of the time window according to the frequency of human acoustic recognition.

First, in order to evaluate the accuracy of the conventional direction detection method based on cross-correlation and the direction detection method according to the present invention in the background noise, the following experimental environment is constructed. The sound source of the Ti20 database is placed at a horizontal angle of 20 degrees or 40 degrees using HRTF. As background noise, white noise, speech noise of a NOISEX database, and car noise are used. In this experiment, we experiment 90 times with different background noise to measure the accuracy of the two-way detection method, and then measure the mean and standard deviation of the results.

4 is a table summarizing the experimental results, where (a) and (b) are white noise, (c) and (d) are crowd noise, and (e) and (f) are vehicle crossover environments. Experimental results for the direction detection method and the direction detection method according to the present invention. As can be seen from this result, the direction detection method according to the present invention is more accurate than the direction detection method by cross-correlation, and even when the noise is mixed, the standard deviation is low, and thus it can be seen that the direction detection method is more robust against noise.

In the following experiment, four sound sources (two female and two female speakers) were used in the TI-DIGIT database to compare the performance of the conventional cross-correlation method and the direction detection method according to the present invention when several sound sources exist at the same time. Select two people and experiment with various background noises. All four sound sources should have the same intensity, and are located at horizontal angles of -10 degrees, 0 degrees, 10 degrees, and 40 degrees, respectively.

5 is an experimental result, a comparison of the horizontal angle histogram of the conventional direction detection method and the direction detection method according to the present invention, Figures 5a and 5b is white noise, Figures 5c and 5d are crowd noise, Figure 5e and 5F shows the result of using automobile noise as background noise. All background noise is set to 5dB intensity. This is a weighted histogram considering only the ITD value. When the weighted histogram is generated in consideration of both the ITD value and the IID value, a three-dimensional histogram is obtained, and thus the sound source direction can be detected more accurately. As a result of the histogram, the direction detection method according to the present invention clearly distinguishes four sound sources, but the conventional direction detection method by cross-correlation can distinguish only one or two distinct sound sources due to a lot of noise generated between the sound source and the sound source. This phenomenon is a result of cross-correlation direction ITD calculation using a fixed time window. When several sound sources are mixed, mutual interference between sound sources occurs, making it difficult to accurately detect the direction of sound sources. On the other hand, the direction detection method according to the present invention can detect the direction of the sound source accurately because the mutual interference between the sound source is low because the size of the time window is variable.

Although the technical spirit of the present invention has been described above with reference to the accompanying drawings, it is intended to exemplarily describe the best embodiment of the present invention, but not to limit the present invention. In addition, it is obvious that any person skilled in the art may make various modifications and imitations without departing from the scope of the technical idea of the present invention.

만큼 계산량이 적다. 예를 들면, N=2.5×10⁴, T=1.6×10^-3[sec], fs=1.0×10⁵[Hz], fs=1.0×10⁵ [Hz], s=21인 경우, 종래기술의 방향 탐지방법은 4×10⁶의 연산이 필요하지만, 본 발명에 따른 방향 탐지방법은 4×10³의 연산이 필요하다. 즉, 본 발명은 종래에 비해 계산량이 1/1000로 줄어든다.The sound source direction detection method using the zero crossing point according to the present invention has an advantage that the amount of calculation is small compared to the sound source direction detection method by the conventional cross-correlation. When the maximum time delay window size for ITD calculation is T [sec] and the sampling frequency of the voice signal is f _s [Hz] in N voice data of the channel, the conventional sound source direction detection method using cross correlation is O ( NTf _s ) has a computational complexity, but the sound source direction detection method using the zero crossing point according to the present invention has a complexity of O (Nsf _i / f _s ). Where s is the number of zero crossings for signal-to-noise ratio calculation, and f _i [Hz] is the center frequency of the channel. Comparing the two values, the direction detection method according to the present invention compared to the prior art

As little computation. For example, in the case of N = 2.5 × 10 ⁴ , T = 1.6 × 10 ⁻³ [sec], fs = 1.0 × 10 ⁵ [Hz], fs = 1.0 × 10 ⁵ [Hz], s = 21, the prior art The direction detection method of requires a calculation of 4 × 10 ⁶ , but the direction detection method according to the present invention requires a calculation of 4 × 10 ³ . That is, the present invention reduces the calculation amount to 1/1000 as compared with the conventional.

In addition, the sound source direction detection method according to the present invention is more accurate direction detection because there is less interference between sound sources in a general environment in which a plurality of sound sources exist at the same time or a variety of noise than the conventional cross-direction direction detection method It is possible to separate the sound source accurately.

Claims (14)

Two sensors corresponding to two ears of a person, a filter bank for frequency-dividing the signals received from the two sensors for each channel, and a processor for calculating a signal-to-noise ratio included in each channel signal separated in frequency from the filter bank. In the signal-to-noise ratio estimation method using the zero crossing point in one device, Receiving the signal output from the same sound source by the filter bank using the two sensors and separating the received signal by channel for each frequency; A ZCPA coding step of the processor obtaining zero-crossing points and peak values by zero-crossing and peak amplitudes (ZCPA) coding each channel signal separated in frequency for each channel; An ITD calculation step of the processor obtaining an interaural time difference (ITD) value by using the zero crossing point and the highest value for each channel obtained in the ZCPA coding step; A dispersion calculation step of the processor dividing each channel signal into a plurality of windows including the same number of zero crossing points and obtaining a variance of the ITD values for each channel and each window; And a signal-to-noise ratio calculation step in which the processor calculates a signal-to-noise ratio using the dispersion of the ITD value for each channel and each window and the center frequency of the channel. The method of claim 1, wherein the ITD calculation step, A method for estimating a signal-to-noise ratio using a zero crossing point, comprising: obtaining a zero crossing point of a right channel closest to an arbitrary zero crossing point of a left channel, and setting a difference value between the two zero crossing points as an ITD value. The method of claim 1, wherein the ITD calculation step, If the maximum time delay value of the left channel and the right channel is less than k times the length of one wavelength of the center frequency of the channel, 2k zero crossings of the right channel adjacent to any zero crossing of the left channel are obtained, and the zero of the left channel is obtained. A signal-to-noise ratio estimation method using a zero crossing point, wherein a difference value between an intersection point and zero crossing points of 2k right channels is set as an ITD value. The method of claim 1, wherein each window of the dispersion calculation step, A method for estimating a signal-to-noise ratio using a zero crossing point, wherein a high frequency channel is divided into a time window having a smaller size than a low frequency channel including a number of zero crossing points according to human acoustic recognition characteristics. The method of claim 1, wherein the signal-to-noise ratio calculation step, The signal-to-noise ratio at the j-th zero crossing point window of the i-th channel is calculated by applying the center frequency (ω _C ) of the channel and the variance (S _i ² (j)) of the ITD value to the following equation. Signal-to-Noise Ratio Estimation Method Using Intersection Points. [Equation]

Using two sensors corresponding to two ears of a person receiving a signal output from the same sound source, a filter bank for separating the signals received from the two sensors for each channel, and each channel signal separated in the frequency from the filter bank In the sound source direction detection method using a zero crossing point in the device having a processor for detecting the direction of the sound source, Receiving, by the filter bank, a signal output from the same sound source using the two sensors, and separating the received signal for each channel by frequency; A ZCPA coding step of the processor obtaining zero-crossing points and peak values by zero-crossing and peak amplitudes (ZCPA) coding each channel signal separated in frequency for each channel; An ITD calculation step of the processor obtaining an interaural time difference (ITD) value by using the zero crossing point and the highest value for each channel obtained in the ZCPA coding step; A dispersion calculation step of the processor dividing each channel signal into a plurality of windows including the same number of zero crossing points and obtaining a variance of the ITD values for each channel and each window; A signal-to-noise ratio calculation step of calculating, by the processor, a signal-to-noise ratio using the dispersion of the ITD value for each channel and window and the center frequency of the channel; A reliability enhancement step of the processor extracting an ITD value whose signal-to-noise ratio is greater than a threshold; And a direction estimating step, wherein the processor obtains a horizontal angle histogram of an ITD value having the signal-to-noise ratio as a weight, and determines a maximum value of the horizontal angle histogram as a spatial position of a sound source. The method of claim 6, wherein the ITD calculation step, A method of detecting a sound source direction using a zero crossing point, comprising: obtaining a zero crossing point of a right channel closest to an arbitrary zero crossing point of a left channel, and setting a difference value between the two zero crossing points as an ITD value. The method of claim 6, wherein the ITD calculation step, If the maximum time delay value of the left channel and the right channel is less than k times the length of one wavelength of the center frequency of the channel, 2k zero crossings of the right channel adjacent to any zero crossing of the left channel are obtained, and the zero of the left channel is obtained. A sound source direction detection method using a zero crossing point, characterized in that the difference between the intersection point and the zero crossing point of 2k right channels is set as the ITD value. The method of claim 6, wherein each window of the dispersion calculation step, A method of detecting a sound source direction using a zero crossing point, wherein a high frequency channel is divided into a time window having a smaller size than a low frequency channel, including a number of zero crossing points according to human acoustic recognition characteristics. The method of claim 6, wherein the signal-to-noise ratio calculation step, The signal-to-noise ratio at the j-th zero crossing point window of the i-th channel is calculated by applying the center frequency (ω _C ) of the channel and the variance (S _i ² (j)) of the ITD value to the following equation. Sound source direction detection method using intersection points. [Equation]

The method of claim 6, And applying an IID (interaural intensity difference) value by applying the zero crossing point and the maximum value information for each channel obtained in the ZCPA coding step to the following equations. [Equation]

here,

Is the IID value,

Is the ZCPA coding result signal of the i-th channel received from the sensor corresponding to the left ear,

Is the ZCPA coding result signal of the i-th channel received from the sensor corresponding to the right ear. 12. The method of claim 11, wherein the reliability improvement step extracts an ITD value and an IID value whose signal-to-noise ratio is greater than a threshold value, and the direction estimation step obtains a horizontal angle histogram of an ITD value and an IID value that weights the signal-to-noise ratio, The sound source direction detection method using a zero crossing point, characterized in that for determining the highest value of the horizontal angle histogram as a spatial position of the sound source. In the system that receives the signal output from the same sound source with two sensors corresponding to the two ears of the human, separates the frequency for each channel and calculates the signal-to-noise ratio included in each channel signal, A ZCPA coding step of zero-crossing and peak amplitudes (ZCPA) coding each channel signal separated by frequency for each channel to obtain a zero crossing point and a maximum value; An ITD calculation step of obtaining an interaural time difference (ITD) value by using the zero crossing point and the highest value for each channel obtained in the ZCPA coding step; A variance calculation step of dividing each channel signal into a plurality of windows including the same number of zero crossing points and obtaining a variance of ITD values for each channel and window; A computer program for executing a signal-to-noise ratio estimation method using a zero crossing point including a signal-to-noise ratio calculation step of calculating the signal-to-noise ratio using the variance of the ITD value and the center frequency of the channel for each channel and window. Recordable media. In a system for receiving a signal output from the same sound source to the two sensors corresponding to the two ears of the human being, frequency separation for each channel and then detecting the direction of the sound source using each channel signal, A ZCPA coding step of zero-crossing and peak amplitudes (ZCPA) coding each channel signal separated by frequency for each channel to obtain a zero crossing point and a maximum value; An ITD calculation step of obtaining an interaural time difference (ITD) value by using the zero crossing point and the highest value for each channel obtained in the ZCPA coding step; A variance calculation step of dividing each channel signal into a plurality of windows including the same number of zero crossing points and obtaining a variance of ITD values for each channel and window; A signal-to-noise ratio calculation step of calculating a signal-to-noise ratio using the variance of the ITD value for each channel and window and the center frequency of the channel; A reliability enhancement step of extracting an ITD value whose signal-to-noise ratio is greater than a threshold; A computer recording a program for executing a sound source direction detection method using a zero crossing point comprising a direction estimating step of obtaining a horizontal angle histogram of an ITD value weighted by the signal-to-noise ratio, and determining a maximum value of the horizontal angle histogram as a spatial position of a sound source. Recordable media that can be read by

Images