KR101580868B1 - Apparatus for estimation of location of sound source in noise environment - Google Patents

Abstract

In the present specification, a plurality of microphones for receiving sounds including noise and sound and converting the sounds into electrical sound signals, a noise removing unit for filtering the sound signals to remove the noise, An apparatus and method for estimating a sound source position in a noisy environment including a sound detecting unit for detecting sound and a sound source position estimating unit for analyzing the detected sound and estimating the position of the sound source.

Description

[0001] APPARATUS FOR ESTIMATION OF LOCATION OF SOUND SOURCE IN NOISE ENVIRONMENT [0002]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and method for estimating the position of a sound source, and more particularly, to an apparatus and method for estimating a position of a sound source in a noisy environment.

Generally, the sound source direction detection technique has a high success rate in a noise-free environment. However, when noise is actually present, the sound and noise generated from the sound source are mixed with each other, so that the success rate of locating the sound source is low.

Patent No. 10-1269189

In order to solve the above-described problem, there is a need for an apparatus and method for removing a noise component and estimating the position of a sound source using only sound generated from the sound source.

An apparatus for estimating a sound source position in a noisy environment according to an embodiment of the present invention includes a plurality of microphones for receiving noise and a sound including a sound generated from a sound source and converting the sound into an electrical sound signal, And a sound source position estimator for estimating a position of the sound source by analyzing the detected sound. The sound source position estimating unit estimates the position of the sound source by analyzing the sound.

In an apparatus for estimating a sound source position in the noisy environment according to an embodiment of the present invention, the noise may include noise due to operation of the cleaner.

In an apparatus for estimating a sound source position in the noisy environment according to an embodiment, the noise canceller may filter the sound signal using a Karhunen-Loeve Transform (KLT) based filter.

In an apparatus for estimating a sound source position in the noisy environment according to an exemplary embodiment, the noise canceller may further include filtering the filtered sound signal using an SDW-MWF (Speech Distortion Weighted Multi-channel Wiener Filter) .

Also, in an apparatus for estimating a sound source position in the noisy environment according to an embodiment, the sound detection unit detects a sound interval in the sound signal through a VAD (Voice Activity Detection), and the sound source position estimation unit The direction of the sound source is estimated using the time difference of the received sound.

The apparatus for estimating a sound source position in the noisy environment according to an embodiment may further include a driving unit for moving the sound source position estimating apparatus in the direction of the estimated sound source.

A method of estimating a sound source position in a noisy environment according to an embodiment of the present invention includes receiving sound including noise and sound generated from a sound source through a microphone, Signal, filtering the sound signal by the processor to remove noise from the sound signal, detecting the sound in the filtered sound signal by the processor, and detecting, by the processor, the detected sound And estimating a position of the sound source.

Also, in a method of estimating a sound source position in a noisy environment according to an embodiment, the noise may include noise due to operation of the cleaner.

In the method of estimating a sound source position in a noisy environment according to an exemplary embodiment, filtering the sound signal includes filtering the sound signal using a Karhunen-Loeve Transform (KLT) -based filter .

In the method of estimating a sound source position in a noisy environment according to an exemplary embodiment, filtering the sound signal may include filtering the filtered sound signal using an SDW-MWF (Speech Distortion Weighted Multi-channel Wiener Filter) The method comprising the steps of:

Also, in the method of estimating a sound source position in a noisy environment according to an exemplary embodiment, the step of detecting the sound may include detecting a speech interval in the sound signal through a VAD (Voice Activity Detection), analyzing the detected sound The step of estimating the position of the sound source estimates the direction of the sound source using the time difference of the sound received by each microphone.

A computer-readable recording medium recording a computer program for performing a method for estimating a sound source position in a noisy environment according to an embodiment of the present invention includes: Receiving a sound including the sound; Converting the received sound into an electrical sound signal through a microphone; Filtering the sound signal by the processor to remove noise from the sound signal; Detecting, by the processor, the speech in the filtered sound signal; And analyzing the detected speech to estimate the location of the sound source by the processor. ≪ RTI ID = 0.0 > [0002] < / RTI >

According to an embodiment of the present invention, a noise canceling algorithm is added by a modification of a conventional voice source detection (VAD) unit and a preprocessor of a sound source direction detection while maintaining the procedure of a conventional sound source direction detection system The performance of the direction detection can be improved. As a result, the function of direction detection in a noisy environment can be easily applied to a home appliance.

1 is a block diagram illustrating an apparatus for estimating a sound source position in a noisy environment according to an embodiment of the present invention.
2 is a detailed configuration diagram of an apparatus 1000 for estimating a sound source position in a noisy environment according to an embodiment of the present invention.
3 is a diagram illustrating a PDOA scheme according to an embodiment of the present invention.
4 is a flowchart of a method of estimating a sound source position in a noisy environment according to an embodiment of the present invention.
5 to 7 are diagrams illustrating experimental results of an apparatus for estimating a sound source position in a noisy environment according to an embodiment of the present invention.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises ", or" having ", or the like, specify that there is a stated feature, number, step, operation, , Steps, operations, components, parts, or combinations thereof, as a matter of principle.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries should be construed as meaning consistent with meaning in the context of the relevant art and are not to be construed as ideal or overly formal in meaning unless expressly defined herein . Like reference numerals in the drawings denote like elements. In the following description, well-known functions or constructions are not described in detail to avoid unnecessarily obscuring the subject matter of the present invention. In addition, the size of each component in the drawings may be exaggerated for the sake of explanation and does not mean a size actually applied.

Embodiments described herein may be wholly hardware, partially hardware, partially software, or entirely software. A "unit," "module," "device," or "system" or the like in this specification refers to a computer-related entity such as a hardware, a combination of hardware and software, or software. A processor, an object, an executable, a thread of execution, a program, and / or a computer, for example, a computer, but is not limited to, a computer. For example, both an application running on a computer and a computer may correspond to a part, module, apparatus, or system of the present specification.

Embodiments have been described with reference to the flowcharts shown in the drawings. While the above method has been shown and described as a series of blocks for purposes of simplicity, it is to be understood that the invention is not limited to the order of the blocks, and that some blocks may be present in different orders and in different orders from that shown and described herein And various other branches, flow paths, and sequences of blocks that achieve the same or similar results may be implemented. Also, not all illustrated blocks may be required for implementation of the methods described herein. Furthermore, the method according to an embodiment of the present invention may be implemented in the form of a computer program for performing a series of processes, and the computer program may be recorded on a computer-readable recording medium.

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

1 is a block diagram illustrating an apparatus for estimating a sound source position in a noisy environment according to an embodiment of the present invention. Referring to FIG. 1, an apparatus 1000 for estimating a sound source position in a noisy environment receives a sound of a user 1 and other surrounding noise through a microphone, processes the sound, It can be estimated and moved toward the sound source.

In one example, the apparatus 1000 for estimating a source location in a noisy environment may be but is not limited to a robot cleaner, and the noise may include noise generated in the operation of the robot cleaner.

2 is a detailed configuration diagram of an apparatus 1000 for estimating a sound source position in a noisy environment according to an embodiment of the present invention. Referring to FIG. 2, an apparatus 1000 for estimating a sound source position in a noisy environment includes a plurality of microphones 100, a noise removing unit 200, a sound detecting unit 300, and a sound source position estimating unit 400. The apparatus 1000 for estimating a sound source position in a noisy environment may further include a driving unit 500 for moving the apparatus 1000 for estimating a sound source position in a noisy environment.

The above-described configuration is an exemplary configuration for explaining the features of the present invention. In addition to the above-described configuration, other components such as a data storage unit, a communication equipment, and a display may be further included in an apparatus for estimating a sound source position in a noisy environment.

The plurality of microphones 100 can receive sounds including noises and sounds generated from the sound sources, and convert the received sounds into electrical sound signals. The plurality of microphones 100 may provide the converted sound signal to the noise removing unit 200.

In one embodiment, the noise remover 200 may filter the sound signal to remove noise from the received sound. It is difficult to estimate the position of a sound source because a sound signal received in a noisy environment includes sound and noise generated from the sound source, so it is necessary to remove the noise portion.

For this, the noise removing unit 200 may filter the sound signal using a KLT (Karhunen-Loeve Transform) based filter.

A method using a KLT (Karhunen-Loeve Transform) based filter is a method of removing noise using a subspace of a covariance matrix.

[Equation 1]

y = x + d

를 추정한다.In Equation (1), x is a speech signal, d is a noise, and y is a sound signal input through a microphone. Through this algorithm, we obtain the filter gain H and multiply it by y

.

&Quot; (2) "

The actual error can be expressed by the equation (2). Also, as the error is minimized, the objective is to minimize r _x and r _d since it approaches the actual desired value. For this, if r _x is minimized and r _d is fixed to a constant value in Equation (2), an optimal H value can be obtained by using a Kuhn-Tuker condition and a Lagrangian multiplier have. Equation (2) can be modified as shown in Equation (3) below for simple calculation.

&Quot; (3) "

In the above equation, R _x is the covariance matrix of the speech signal, and R _d is the covariance matrix of the noise signal. And V is the eigenvector of Σ = R _d ^-1 R _x . The G value is a diagonal matrix and can be expressed by Equation (4) below.

&Quot; (4) "

In Equation (4), K denotes the frame size of the entire signal, and M denotes the number of positive eigenvalues of 裡. Finally, the gain of the KLT-based filter is obtained to eliminate the noise.

In one embodiment, the noise remover 200 may further filter the filtered sound signal using a Speech Distortion Weighted Multi-channel Wiener Filter (SDW-MWF).

That is, the noise removing unit 200 can remove the noise-canceled signal through the KLT-based filter in the single channel (each microphone) through the SDW-MWF using the multi-channel characteristic . The main theory of the Wiener filter is to obtain the filter gain H satisfying the minimum mean square error (MMSE).

The result obtained from the above KLT-based filter can be transformed into the frequency domain through the STFT and expressed as Equation (5) below.

&Quot; (5) "

Equation (5) denotes an input signal, a voice signal, and a noise signal at a k-th frequency bin of the i-th microphone. Then, each of the microphone signals in the same manner as in Equation (5) can be grouped together and expressed as a single vector. Then, the noise can be selectively removed by using the expression expressed in this way.

&Quot; (6) "

First, the noise eliminator 200 obtains a filter gain that minimizes MMSE in the same manner as in Equation (6). The filter gain H can be obtained by performing a partial differentiation for H.

The obtained filter gain is expressed by Equation (7) below.

&Quot; (7) "

In Equation (7), e denotes a vector for selecting a microphone to be filtered, and Ryy and Rdd denote an autocorrelation matrix of an input signal (speech) and a noise signal. Equation (7) represents the final filter gain of the MWF. However, in general, the noise cancellation algorithm has a trade-off relationship between the performance of the filter after filtering and the distortion of the final signal. A way to directly control this relationship is to adjust the noise weight of the equation by solving Equation 7 with SDW-MWF. This filter gain can be expressed as Equation (8) below.

&Quot; (8) "

Equation (8) is a formula for the final filter gain. The final speech signal can be estimated by multiplying Equation (8) by the result of removing the noise through the KLT-based filter.

In one embodiment, the speech detector 300 may detect the speech in the filtered speech signal. Specifically, the voice detector 300 can detect a voice interval in the voice signal through voice activity detection (VAD). VAD is a method of detecting a person's voice in a given section. Although the VAD has various methods, the voice detector 300 according to an embodiment of the present invention can use a combination of a VAD method using a harmonic characteristic and a VAD method using a band-pass filter But are not limited thereto.

A harmonic of a voice is a characteristic that is caused by the resonance of the vocal cords and means a characteristic having energy at an integer multiple of the fundamental frequency of the voice. That is, the fundamental frequency should be obtained in order to use the harmonic characteristic, and the fundamental frequency means a peak point of the first frequency bin when the noise canceled signal is represented in the frequency domain.

&Quot; (9) "

N is the number of harmonic frequencies. And F (n.l) denotes a magnitude value at the n-th harmonic frequency of the l-th frame. Hence, Equation (9) means an average value of harmonic components in a given lth frame.

In order to actually detect a voice, the voice detection unit 300 normalizes the values obtained through Equation (9) in each frame using a result obtained in an interval where there is no voice in the initial part of the voice signal to obtain a value . In addition, a value equal to or greater than the threshold obtained in each frame period can be recognized as speech.

 The voice detection unit 300 may use a band-pass filter. The bandpass filter is a filtering method that fetches only data of a certain specified range of frequencies. Here, for the VAD, only the frequency band of the voice band is filtered through the bandpass filter to obtain the energy. Then, when the energy value thus obtained exceeds a predetermined threshold value, it is recognized as a voice.

The voice detection unit 300 adds the values obtained from the above-mentioned two methods and sets them as one value. Then, when this value exceeds a predetermined threshold value, the VAD can be performed in a manner of recognizing it as a voice.

In one embodiment, the sound source location estimating unit 400 may estimate the location of the sound source by analyzing the detected sound. In one example, the sound source position estimation unit 400 may estimate the direction of the sound source using the time difference of the sound received by each microphone.

The sound source position estimating unit 400 may estimate the position of a sound source using a Phase Difference of Arrival (PDOA) method.

Time delay of arrival (TDOA) is a method of detecting a direction using a correlation of time differences of sound sources entering a plurality of microphones in the time domain. That is, the TDOA method is a method of detecting the actual angle by comparing the TDOA of the input signal of each microphone with the virtual TDOA map between each microphone created beforehand and the actual environment. However, this TDOA method can solve this problem by detecting the direction by the PDOA method performed in the frequency domain because spatial aliasing occurs at a high frequency.

3 is a diagram illustrating a PDOA scheme according to an embodiment of the present invention.

Referring to FIG. 3, in order to perform the PDOA algorithm, a virtual map to be compared with an actual phase difference should be made.

&Quot; (10) "

In the above equation, SM _i denotes the distance from the sound signal to the i-th microphone, and Vs denotes the sound velocity (340 m / s). Therefore, Equation (10) is an expression indicating a difference in time required for the sound signal located at the angle? To reach the i-th microphone and the j-th microphone. If the above equation is expressed in the frequency domain, it can be expressed as Equation (11) below.

&Quot; (11) "

Where f represents a frequency bin. Using the above equation (11), a set of two mates from four micros can be expressed as shown in the following equation (12).

&Quot; (12) "

The above equation (12) represents a virtual PDOA map composed of a total of six. Then, the actual phase difference for comparison with the virtual map is obtained.

&Quot; (13) "

Where k is the number of frames in the conversion process into the frequency domain. Therefore, the above expression shows the phase difference between the i-th microphone input from a certain angle and the signal input through the j-th microphone.

&Quot; (14) "

Equation 14 is a cost function that compares the difference between the virtual PDOA map and the actual PDOA value at every frequency bin and microphone pair. This cost function is a value using cosine, meaning that a larger value means less error.

&Quot; (15) "

Therefore, the sound source position estimating unit 400 can use the PDOA method of estimating the most satisfactory angle satisfying the above values at all angles and finally expressing the angle by the estimated angle.

The apparatus 1000 for estimating a sound source position in a noisy environment in one embodiment may further include a driving unit 500 for moving the sound source position estimating apparatus in the direction of the estimated sound source. Accordingly, the apparatus 1000 can move in the direction of the sound source. To this end, the driving unit 500 may include a wheel and a motor unit.

4 is a flowchart of a method of estimating a sound source position in a noisy environment according to an embodiment of the present invention. Referring to FIG. 4, a method for estimating a sound source position in a noisy environment includes receiving a sound including noise and a sound generated from a sound source through a microphone (S10), and receiving the sound through an microphone (S30) of filtering the sound signal to remove noise from the sound signal by the processor (S30), detecting (S40) the sound in the filtered sound signal by the processor, and And analyzing the detected voice by the processor to estimate the position of the sound source (S50).

In one example, the noise described above may include noise generated by the operation of the vacuum cleaner.

The step S30 of filtering the sound signal may include filtering the sound signal using a Karhunen-Loeve Transform (KLT) -based filter. In another embodiment, the step S30 may include filtering a speech distortion weighted multi- The method may further include filtering the filtered sound signal using a channel-Wiener filter.

In addition, the step of detecting the voice (S40) may detect a voice interval in the voice signal through VAD (Voice Activity Detection). The step of estimating the position of the sound source by analyzing the detected sound may estimate the direction of the sound source using the time difference of the sound received by each microphone.

5 to 7 are diagrams illustrating experimental results of an apparatus for estimating a sound source position in a noisy environment according to an embodiment of the present invention.

FIG. 5 is a result of experimentally recording a sound source signal in a situation where the robot cleaner is actually driven, in order to confirm the performance of the sound source direction detection technique in a noisy environment.

The experiment was carried out with a total of 12 times from 0 degree to 330 degree at intervals of 30 degrees in the form of five fires in a certain angle. Also, considering that it is applied in actual household, experiments were performed at various SNRs such as 4dB, 1dB, -3dB, -5dB, and -7dB.

In FIG. 5, (a) and (b) are input signals which are not processed in the time domain and the spectrogram. (c) and (d) are the results of removing noise through KLT based filter in time domain and spectrogram. (e) and (f) show the results of SDW-MWF in time domain and spectrogram. (g) and (h) show the result of removing noise through KLT-based filter and SDW-MWF in time domain and spectrogram.

(a) and (b) are noise environments before filtering, and the SNR is -4.83 dB. As a result of (c) and (d), the SNR was increased to -0.37 dB. In addition, (e) and (f) show that the SNR is 1.47 dB as a result of SDW-MWF. Finally, (g) and (h) are performed in accordance with the present invention. As a result of filtering using the KLT-based filter and the SDW-MWF, the SNR is 5.8 dB and the noise is removed most.

FIG. 6 is a graph comparing the proposed algorithm with the False Positive, False Negative, and the success rate in the case of using the algorithm previously used according to the filtering method to confirm the performance of the direction detection.

First, False Positive means an error situation in which angles are estimated in a non-ignited section. False Negative is the situation where the angle is not estimated or the actual angle is not estimated in the section where the speech exists. The success rate is the number of times the actual angle and the estimated number of results between +10 degrees and -10 degrees when five utterances are performed at a given angle. That is, when the success rate is 5, it means that all the utterances are detected five times, which means that the success rate is 100%. The higher the success rate, the better. The lower the False Positive and the False Negative, the better.

 FIGS. 6 and 7 show the average value of the results of removing noise at SNRs of 4dB, 1dB, -3dB, -5dB, and -7dB and performing direction detection at each angle. FIG. 6 shows results of sorting the direction detection results according to the filtering method, and FIG. 7 shows results obtained by classifying the same data according to the VAD method.

Referring to FIG. 6, when a filter according to the present invention using a KLT-based filter and a SDW-MWF together, which is a single filter, is used rather than a baseline data that does not perform filtering, Can be confirmed.

And we can confirm that the KLT based filter among the two noise canceling filters has a low success rate or error occurrence rate (actually SNR is good but MWF is good but direction detecting performance is better than KLT based filter).

In fact, if the noise cancellation performance increases, the distortion of the signal may become worse and the performance of the direction detection may be deteriorated. However, since the two filters used here are selected based on this, the direction detection performance is not deteriorated.

Fig. 7 shows the results of classifying the direction detection results largely according to the VAD method. Here, a general VAD means a VAD method using basic energy, and the VAD proposed in the present invention means a combination of a VAD method using a harmonic characteristic and a VAD method using a bandpass filter as described above.

6 and 7, it can be confirmed that the VAD method using the harmonic characteristic and the VAD method using the bandpass filter have better performance than the general VAD method. In addition, we confirmed that the VAD method using the harmonic characteristic has better performance than the VAD method using the bandpass filter in terms of the error rate and the success rate. Finally, since the proposed VAD method uses both of the advantages of both methods by using both the harmonic characteristic and the bandpass filter, the success rate of voice detection can be increased, and as a result, it is possible to confirm that the performance of direction detection improves

While the invention has been shown and described with reference to certain embodiments thereof, it will be understood by those skilled in the art that various changes and modifications may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. However, it should be understood that such modifications are within the technical scope of the present invention. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

100: microphone
200: Noise canceler
300:
400: sound source position estimating unit
500:
1000: a device for estimating a sound source position

Claims (12)

A plurality of microphones for receiving sounds including noises and sounds generated from the sound sources and converting the sounds into electric sound signals;
A noise eliminator for filtering the sound signal to remove the noise;
A voice detector for detecting the voice in a filtered sound signal; And
And a sound source position estimator for analyzing the detected sound to estimate a position of the sound source,
The noise-
A KLT (Karhunen-Loeve Transform) -based filter and a SDW-MWF (Speech Distortion Weighted Multi-channel Wiener Filter) are sequentially applied using a Kuhn-Tuker condition and a Lagrangian multiplier And estimating the position of the sound source in a noisy environment.
The method according to claim 1,
Wherein the noise includes noise due to operation of the cleaner.
delete delete The method according to claim 1,
The voice detection unit detects a voice interval in the voice signal through VAD (Voice Activity Detection)
Wherein the sound source position estimating unit estimates a direction of a sound source by using a time difference between sounds received by the respective microphones.
The method according to claim 1,
And a driving unit for moving the sound source position estimating apparatus in the direction of the estimated sound source.
Receiving a sound including a noise and a sound generated from the sound source through a microphone;
Converting the received sound into an electrical sound signal through a microphone;
Filtering the sound signal by the processor to remove noise from the sound signal;
Detecting, by the processor, the speech in the filtered sound signal; And
Analyzing the detected speech by the processor to estimate the position of the sound source,
Wherein the step of filtering the sound signal comprises:
Filtering the sound signal using a Kuhn-Tuker condition and a Karhunen-Loeve Transform (KLT) -based filter using a Lagrangian multiplier, and filtering the sound signal using a Speech Distortion Weighted Multi-Channel (MWF) And filtering the filtered sound signal using a Wiener filter to estimate a sound source position in a noisy environment.
8. The method of claim 7,
Wherein the noise comprises noise due to operation of the cleaner.
delete delete 8. The method of claim 7,
The detecting of the voice may include detecting a voice interval in the voice signal through a voice activity detection (VAD)
Wherein the step of estimating the position of the sound source by analyzing the detected sound estimates the direction of the sound source using the time difference of the sound received by each microphone.
Receiving a sound including a noise and a sound generated from the sound source through a microphone; Converting the received sound into an electrical sound signal through a microphone; Filtering the sound signal by the processor to remove noise from the sound signal; Detecting, by the processor, the speech in the filtered sound signal; And analyzing the detected speech by the processor to estimate the position of the sound source, wherein filtering the sound signal comprises:
Filtering the sound signal using a Kuhn-Tuker condition and a Karhunen-Loeve Transform (KLT) -based filter using a Lagrangian multiplier, and filtering the sound signal using a Speech Distortion Weighted Multi-Channel (MWF) A method for estimating a sound source location in a noisy environment, comprising the steps of: filtering the filtered sound signal using a Wiener Filter, the computer program product comprising: Recording medium.

Images