CN102305925A - Robot continuous sound source positioning method - Google Patents

Abstract

The invention discloses a robot continuous sound source positioning method which comprises the following steps of: (1) acquiring a sound source signal by adopting a microphone array consisting of at least two pairs of microphones, converting the signal into a digital signal and storing the signal by a robot; and (2) carrying out effective audio signal detection on the acquired signal and carrying out sound source positioning operation on the signal containing effective audio frequency to obtain the position of a sound source relative to the robot. The method disclosed by the invention can divide the frequency domain signal obtained after Fourier transformation as for different application scenes and the frequency characteristic and noise characteristic of the sound source when the method is applied into a sound source positioning system of the robot and the calculating capability of the robot is limited, and thereby, the positioning operation time of the robot is greatly reduced, and the positioning speed is high. Meanwhile, as for the characteristic of movement of the robot, after the displacement angle is subjected to motion modification, the accuracy and instantaneity of positioning are greatly improved.

Description

The continuous sound localization method of a kind of robot

Technical field

The present invention relates to a kind of robot field's auditory localization treatment technology, especially based on the continuous sound localization method in the robot that is applied in of microphone array.

Background technology

The world that robot is faced with is the world that information complicated and changeable can be provided; Make that robot is that human service just needs it that various applied environments are all had corresponding information obtain manner and disposal route, also needs it to make different decision-makings according to the variation of environment simultaneously.And robot generally imports external environmental information through sensor, and most important is exactly vision sensor and hearing transducer two big information input sources.In recent years, the robot vision technology had had significant progress, expanded the range of application of robot greatly, had improved the work efficiency of robot.But visually-perceptible is subjected to the restriction of factors such as illumination, barrier and visibility, and under or the situation that barrier stops relatively poor in light condition, visually-perceptible will lose efficacy.Auditory system can not be subjected to above-mentioned these factor affecting fully as the important component part of human sense organ, just for the research of robot cognition technology new field is provided.The robotic vision and the sense of hearing simultaneously can combine, and learning from other's strong points to offset one's weaknesses provides service for human society jointly.

General robot auditory localization technology is through the simulation to human auditory system mechanism; Utilize robot on one's body or the sensor devices such as microphone array that are placed among the robot environment receive sound wave; Because microphone is in the same environment; Certainly existing certain correlativity between the voice signal that these microphones collect; Through electronic installation or algorithm; Utilize this correlativity that acoustical signal is handled, thereby realize that sound source position is surveyed, speech recognition or the location and the tracking of target.This technology receives noise and reverberation influence in the robot environment of living in, receive robot motion's body Effects of Noise and receive sound source through barrier the reflection and the influence of refraction and some environmental parameters, make sound source accurately located to become very difficult.The restriction of the maximum that self had of mobile robot is exactly high to the rate request of auditory localization in addition.Robot is positioned at current certain position; A sounds; Suppose that robot is with linear uniform motion; Movement velocity is v; Be t positioning time, in air, do not propagate the time that is spent robot has produced v*t between actual sound source and the robot when obtaining positioning result displacement even if do not consider sound; This situation requires careful consideration, particularly when robot and sound source position have relative motion.

Existing sound localization method adopts the mistiming technology that arrives more.At first, the microphone in the microphone array model is divided into per two one group, it is several right to be divided into, right for each microphone, calculates the mistiming of voice signal from two microphones of sound source arrival, poor according to this mistiming computed range; Utilize topological structure parameter, the range difference of microphone array, the speed of sound transmission to estimate the position of sound source then.The core of this method is exactly to calculate sound to arrive the right mistiming of each microphone.Fig. 1 is based on time delay and estimates the location algorithm schematic diagram.Here adopt that (Cross-power spectrum phase, CSP) the time delay algorithm for estimating calculates based on a kind of cross-power spectrum phase place.This method is to utilize the phase information of cross-power spectrum to come estimation time delay, and it to the weighting function of signal cross-power spectrum is:

$W_{\mathrm{csp}}\left(\mathrm{\ω}\right)=\frac{1}{\left|G_{12}\left(\mathrm{\ω}\right)\right|}---\left(1\right)$

G wherein ₁₂(ω) be the cross-spectral density function of signal

The signal that the sound that sends from sound source arrives two microphones can be expressed as:

x ₁(t)＝s ₁(t)+n ₁(t) (2)

x ₂(t)＝αs ₁(t+τ)+n ₂(t) (3)

X wherein ₁(t), x ₂(t),, a microphone is expressed as s if being received the sound that sound source sends for arriving the voice signal of two microphones respectively at moment t ₁(t), then we can receive an other microphone sound that sound source sends and be expressed as α s ₂(t+ τ), n ₁(t), n ₂(t) be two noises that microphone receives respectively, τ is the mistiming that sound that sound source is sent arrives two microphones, and α is an attenuation coefficient.Known the mistiming τ that arrives; Distance D 1 between while two microphones is known value; Can calculate the sound source position right with respect to this microphone; But only calculating two mistimings between the microphone will obtain an eligible bi-curved information on three dimensions; Can not be with the sound source accurate in locating; Therefore in practical application, adopt many forms to collect voice signal to microphone array; Calculate sound source with respect to the hyperboloid of every pair of microphone on three dimensions, with many be exactly the accurate position of sound source to the corresponding respectively bi-curved point of crossing of microphone.

CSP method difference computing time τ step is following:

1. at first the signal of microphone collection is carried out analog to digital conversion, convert digital signal into, again digital signal is carried out the computing of simple crosscorrelation at frequency domain.

2. two paths of signals is carried out Fourier transform respectively, the result to the Fourier transform of calculating two paths of signals carries out computing cross-correlation then:

$S\left(\mathrm{\ω}\right)=\underset{m=1}{\overset{N}{\mathrm{\Σ}}}x\left[n\right]\exp (-\mathrm{jwm})---\left(4\right)$

${R^{\′}}_{x_1{x}_2}\left(\mathrm{\ω}\right)S\left({\mathrm{\ω}}_1\right)S{\left({\mathrm{\ω}}_2\right)}^{*}---\left(5\right)$

X[n wherein] be the sampled signal of time domain, S (ω) for time-domain signal through the frequency-region signal after the Fourier transform.

3. for a series of Effect of Environmental such as the reflection that reduces ground unrest, sound, refraction, reverberation, improve bearing accuracy, signal is carried out carrying out whole weighted at frequency domain before the computing cross-correlation, promptly

${R^{\′}}_{x_1{x}_2}\left(\mathrm{\ω}\right)=W_n\left(\mathrm{\ω}\right)S\left({\mathrm{\ω}}_1\right)S{\left({\mathrm{\ω}}_2\right)}^{*}---\left(6\right)$

W wherein _n(ω) be weighting function, S (ω ₁), S (ω ₂) ^*Be respectively the frequency-region signal that two paths of signals obtains through Fourier transform.

4. result of calculation is carried out inverse Fourier transform and transform to time domain again, the result of inverse transformation is searched for peaked position, can confirm the mistiming τ of two paths of signals.

Existing method has only been considered a series of Effect of Environmental such as ground unrest, reverberation, does not consider the characteristic of sound source itself.Existing simultaneously method is all carried out correlation operation to all Frequency points, and complexity is high, and has wasted a large amount of assessing the cost and the time.And in the continuous auditory localization process of robot, continuous meaning is the sound source position that real-time positioning is correct.In the continuous sonic location system of robot, topmost problem is a real-time, therefore needs robot can calculate azimuth information fast, therefore just need improve existing algorithm, reduces the complexity of algorithm, and arithmetic speed is accelerated.

A problem of prior art existence is in addition; The mobile robot is when having relative motion with sound source; Because the existence of calculation delay; Suppose that calculation delay is Td; Then current position of calculating is the relative position of mobile robot and sound source Td times prior; Can not reflect truly that this problem was more unobvious when the arithmetic capability of relative position, particularly robot of current robot and sound source was high.

Summary of the invention

To the shortcoming that prior art exists, we have proposed a kind of method of new auditory localization, have overcome shortcoming of the prior art, and have obtained good effect.

Core concept of the present invention is that the signal that Fourier transform is later carries out the frequency spectrum fragment process; Only the frequency chip to part carries out correlation operation; And different bursts is added different weight processing, and irrelevant frequency chip is changed to zero, do not carry out correlation operation.The characteristics that have a relative motion to mobile robot and sound source are estimated the motion of robot simultaneously, and according to prediction result current position location are compensated.

In fact, the thought of this method comes from the calculating of signal to noise ratio (S/N ratio), is different through calculating the signal to noise ratio (S/N ratio) of finding the different frequency sheet; When particularly the sound that sends of sound source was voice, experimental observation was thought, when the signal to noise ratio (S/N ratio) of signal is higher than certain threshold value; It is a voice signal, otherwise is noise.Therefore; We find; The different frequency sheet also is different carrying out computing cross-correlation time institute role; Be that its signal energy value of the higher frequency chip of signal to noise ratio (S/N ratio) is big more; The cross-spectral density function more can characterize the characteristic of voice signal part, so this part frequency chip just needs the weighting function of increasing.

Method of the present invention is:

1) the robot microphone array that adopts at least two pairs of microphones to constitute is gathered sound-source signal, and is to store after the digital signal with conversion of signals;

2) signal of gathering is carried out effective sound signal and detect, the signal that will include effective audio frequency carries out Fourier transform, converts frequency domain into from time domain;

3) signal that converts frequency domain into is carried out fragment process, the signal data that the every pair of microphone is obtained carries out correlation calculations respectively, and the result of correlation calculations is carried out inverse Fourier transform, is transformed into time domain from frequency domain;

4) carry out the maximum value position search, obtain sound source to the right mistiming τ of each microphone;

7) calculate the position of sound source with respect to robot according to angle θ and mistiming τ, wherein θ is initial point and the line of sound source and the angle between the coordinate system x axle forward of microphone array coordinate system of living in.

Specifically describe step involved in the present invention below:

1, the robot microphone array that adopts at least two pairs of microphones to constitute is gathered sound-source signal, and is to store after the digital signal with conversion of signals;

2, the signal of gathering is carried out effective sound signal and detect, the signal that comprises effective audio frequency is carried out the auditory localization computing, obtain the position of sound source with respect to robot.If do not have effective sound signal to exist then do not position computing.Operational method is following:

A) signal that will include effective audio frequency carries out Fourier transform, converts frequency domain into from time domain:

$S\left(\mathrm{\ω}\right)=\underset{m=1}{\overset{N}{\mathrm{\Σ}}}x\left[n\right]\exp (-\mathrm{jwm});---\left(7\right)$

X[n wherein] be the signal of time domain, ω is a frequency variable, and j representes a unit of imaginary part, and m is a variable, the number of expression from 1 to n, S (ω) is the signal through the frequency field that obtains after the Fourier transform.

B) signal that converts frequency domain into is carried out fragment process,, does not carry out the computing of correlativity the extraneous data zero setting of setpoint frequency:

$S^{\′}\left(\mathrm{\ω}\right)=\underset{m=a}{\overset{b}{\mathrm{\Σ}}}x\left[n\right]\exp (-\mathrm{jwm})---\left(8\right)$

A～b is meant the setpoint frequency scope.

C) the frequency-region signal data that respectively every pair of microphone obtained are carried out correlation calculations:

${R^{\′}}_{x_1{x}_2}\left(\mathrm{\ω}\right)=W_{\mathrm{csp}}\left(\mathrm{\ω}\right)S^{\′}\left({\mathrm{\ω}}_1\right)S^{\′}{\left({\mathrm{\ω}}_2\right)}^{*}---\left(9\right)$

Wherein, W _Csp(ω) be weighting function

$W_{\mathrm{csp}}\left(\mathrm{\ω}\right)=\frac{1}{\left|G_{x_1{x}_2}\left(\mathrm{\ω}\right)\right|}---\left(1\right)$

for the cross-power spectral density function

$G_{x_1{x}_2}\left(\mathrm{\ω}\right)=\underset{\mathrm{\ω}=0}{\overset{n}{\mathrm{\Σ}}}f\left(\mathrm{\ω}\right){\mathrm{\α}}_{x_1}{\mathrm{\α}}_{x_2}{S}^{\′}\left(\mathrm{\ω}\right)S^{\′*}\left(\mathrm{\ω}\right){\exp }^{-\mathrm{j\ω\τ}}$

Wherein τ is that signal is to the right mistiming of microphone.F (ω) is the weighted value of each frequency chip.ω is divided into different sheets, and for different sheets, f (ω) gets different values.

D) result to the correlation calculations of c step carries out inverse Fourier transform, is transformed into time domain from frequency domain:

$x^{\′}\left[n\right]=\frac{1}{2\mathrm{\π}}{\∫}_{-\∞}^{\∞}{R^{\′}}_{x_1{x}_2}\left(\mathrm{\ω}\right)e^{\mathrm{j\ωt}}\mathrm{d\ω}$

The time-domain signal that obtains for inverse Fourier transform of x ' [n] wherein, t is constantly.

E) result of calculation of d step is carried out the maximum value position search, can obtain sound source to the right mistiming τ of said microphone.

F) calculate the angle θ of sound source with respect to microphone array, θ is initial point and the line of sound source and the angle between the coordinate system x axle forward of microphone array coordinate system of living in, and θ is the angle on the space.

G) calculate the position of sound source according to angle θ and mistiming τ with respect to robot.

When said robot and sound source have relative motion; Because the calculating sound source needs the regular hour with respect to the angle of microphone array; And at this moment in the process; Between robot and the sound source also relative motion is taking place; Therefore, also need according to the computing interval robot and sound source to the angular velocity of motion to f) the angle θ that calculates revises.

The method that current angle θ is revised is:

A, transfer angle when preceding the N time computing acquisition of time auditory localization computing;

B, acquisition are as the angle of time computing and the differential seat angle Ad of preceding the N time computing, and the angular velocity Ad/ (N*Tf) of calculating sound source and robot relative motion, and Tf is for working as the mistiming of time computing and preceding the N time computing;

C, change Td*Ad/ (N*Tf) with the angle between sound source in the Td time and the robot and revise current angle, that is, the angle of current calculating adds that the angle of correction is real angle θ, i.e. θ+Td*Ad/ (N*Tf).Wherein θ is f) angle that calculates, Td is that robot calculates the needed time of angle at every turn.

Effective audio signal data that said auditory localization computing whenever collects set amount in time domain just carries out the computing of an auditory localization.

Said effective audio signal data is a speech data.

The data of said set amount are the data of frame length size.

Said robot can be the mobile robot, also can be the stationary machines people.

Said sound source can be stationary sound source, also can be for moving sound source.

Said Fourier transform can be Fast Fourier Transform (FFT).

The microphone array of said robot can be the array of cruciform array, triangular array, circular array, linear array or other types.

Described fragment process is that all Frequency points after the Fourier transform are divided into different groups, divides into groups according to the frequency range of effective sound signal.

What of effective sound signal of being comprised are relevant in said f (ω) and each group, the frequency band that useful signal is concentrated, and the weight of f (ω) weighting is big, the frequency band that useful signal is few, the weight of f (ω) weighting is little.

Technique effect of the present invention:

In the sonic location system of robot, when particularly the robot computing power is limited.To different application scenes, the frequency-region signal that the frequency characteristic of sound source and the characteristic of noise can be later with Fourier transform carries out burst, and the frequency spectrum of different sheets is carried out different processing, has significantly reduced the location operation time of robot, and locating speed is fast.Simultaneously to the characteristics of robot movement, to its move the parallactic angle degree move revise after, the accuracy that has significantly improved the location is with real-time.

Description of drawings

Fig. 1 is based on time delay and estimates the location algorithm schematic diagram;

Fig. 2 is a microphone array model synoptic diagram;

Fig. 3 is the floor map of robot and point source of sound;

The practical implementation method

According to specific embodiment and Figure of description method of the present invention is described in detail below.

1, to gather sound-source signal respectively and carry out analog to digital conversion be digital signal to two pairs of microphone arrays, and storage.

2,1 digital signal of gathering is carried out effective sound signal and detect, the signal that comprises effective audio frequency is carried out the auditory localization computing, obtain the position of sound source with respect to robot.Effectively sound signal detects and is prior art; Can detect according to the characteristic of sound-source signal, in this embodiment, effectively sound signal detects and is the sound end detection; If have voice to exist then begin to carry out the auditory localization computing, if do not have voice to exist then do not position computing.The voice signal of gathering is carried out the branch frame in time domain handle, promptly whenever collect the computing that a certain amount of data (data of frame length size) are just carried out an auditory localization

3, the acquired signal after handling in 2 is carried out Fast Fourier Transform (FFT)

$S\left(\mathrm{\ω}\right)=\underset{m=1}{\overset{N}{\mathrm{\Σ}}}x\left[n\right]\exp (-\mathrm{jwm})---\left(7\right)$

4, the signal in 3 is carried out fragment process in frequency field, to the frequency characteristic of sound source, only get the data of the frequency chip in the setpoint frequency scope and carry out correlation operation, the computing of correlativity is not carried out in the extraneous data zero setting of setpoint frequency

$S^{\′}\left(\mathrm{\ω}\right)=\underset{m=a}{\overset{b}{\mathrm{\Σ}}}x\left[n\right]\exp (-\mathrm{jwm})---\left(8\right)$

A～b is meant the setpoint frequency scope.

5, different frequency chips is carried out the correlation calculations that different weights is handled:

${R^{\′}}_{x_1{x}_2}\left(\mathrm{\ω}\right)=W_{\mathrm{csp}}\left(\mathrm{\ω}\right)S^{\′}\left({\mathrm{\ω}}_1\right)S^{\′}{\left({\mathrm{\ω}}_2\right)}^{*}---\left(9\right)$

Wherein, W _Csp(ω) be weighting function

$W_{\mathrm{csp}}\left(\mathrm{\ω}\right)=\frac{1}{\left|G_{x_1{x}_2}\left(\mathrm{\ω}\right)\right|}---\left(1\right)$

for the cross-power spectral density function

$G_{x_1{x}_2}\left(\mathrm{\ω}\right)=\underset{\mathrm{\ω}=0}{\overset{n}{\mathrm{\Σ}}}f\left(\mathrm{\ω}\right){\mathrm{\α}}_{x_1}{\mathrm{\α}}_{x_2}{S}^{\′}\left(\mathrm{\ω}\right)S^{\′*}\left(\mathrm{\ω}\right){\exp }^{-\mathrm{j\ω\τ}}$

Wherein τ is that signal is to the right mistiming of microphone.F (ω) is the weighted value of each frequency chip.ω is divided into different sheets, and for different sheets, f (ω) gets different values.Signal carries out fragment process in frequency field and is meant according to different application scenes; With after the Fourier transform be divided into different groups a little; A point after the Fourier transform is represented a frequency; Burst is exactly that different points is divided into different groups; Group unit minimum is a point; Maximum is whole point, handles not carrying out different weights on the same group.The principle of burst weighting is to classify according to the frequency range of useful signal in the signal, the frequency band that useful signal is concentrated more, and the weight of weighting is big more; The frequency band that useful signal is few more; The weight of weighting is more little, reduces the influence of non-useful signal to integral body like this, improves signal to noise ratio (S/N ratio).(for example: if when sound source is voice; Generally concentrate on the frequency band of 100-4kHz at the frequency field valid data; But major part all concentrates on 200-1200Hz; Therefore can simply be divided into 100-200Hz; 200-1200Hz; Three groups of 1200-4000Hz carry out different weighted to every group).

6, the result of calculation in 5 is carried out inverse fast Fourier transform, result of calculation is transformed to time domain.

$x^{\′}\left[n\right]=\frac{1}{2\mathrm{\π}}{\∫}_{-\∞}^{\∞}{R^{\′}}_{x_1{x}_2}\left(\mathrm{\ω}\right)e^{\mathrm{j\ωt}}\mathrm{d\ω}$

The time-domain signal that obtains for inverse Fourier transform of x ' [n] wherein.

7, the result of calculation in 6 is searched for peaked position, can obtain sound source to the right mistiming τ of microphone

8, obtain the angle θ with respect to microphone array respectively according to the result of calculation of step 7, θ is initial point and the line of sound source and the angle between the coordinate system x axle forward of microphone array coordinate system of living in.

9, when between robot and the sound source relative motion taking place, need revise angle theta

The a} sound end detects when sound is arranged, and stores N frame before the current location respectively to every pair of microphone, below the angle of (being designated as F5) is described as an example with 5 frames.

B} calculates the poor Ad of angle of angle and the F5 of present frame (F0), and according to the object of which movement characteristics of stationarity in short-term, we think that F5 is mild to the motion of these continuous 6 frames of F0.Calculate the poor of these two angles, if the mistiming between two frames is Tf, the angular velocity of then predicting current sound source and robot relative motion is Ad/ (5*Tf).

C) the needed time of angle of calculating every frame is Td, then Td in the time angle between sound source and the robot be changed to Td*Ad/ (5*Tf)

D) carry out above-mentioned computing to two groups of microphones respectively, and revise the relative angle of current sound source and robot with it.

Revise later angles according to two and obtain the position of sound source, and can be directed against different application, the frequency chip of voice signal is carried out different processing with respect to microphone array.

In mobile robot's sonic location system, to the characteristics of robot movement, to its move the parallactic angle degree move revise after, the accuracy that has significantly improved the location is with real-time, when particularly the robot computing power is limited.To different application scenes, the frequency characteristic of sound source and the characteristic of noise can be carried out different processing to different frequency spectrum sheets.Protection scope of the present invention is not limited to following instance.

The microphone array model is a cruciform array that is made up of four microphones, and like Fig. 2, each all is 40cm to the distance between the microphone on the cross, is fixed on the shoulder apart from the robot of ground 100cm, constitutes a planar array model.Experimental situation is the room of a 8m*8m*3m, and as reference field, Fig. 3 is the floor map of robot and point source of sound, the position of the clear direction of figure acceptance of the bid and robot and sound source with the plane at microphone array model place.Robot at first stands in the central authorities in room; Draw on the floor and be used for the people that stands as 72 positions at 72 o'clock; Be the position of test sound source point; Being set to whenever at a distance from 15 of 72 points ± be a Sounnd source direction; Totally 24 directions; On each direction according to distance microphone array center 1m, 2m and 3m three places, totally 72 point sources of sound.People and robot can do relative motion.Ask for help and just can pronounce when being positioned on the point shown in Figure 3, purpose is to make things convenient for test result.

In the experiment to 28 groups of speech datas of each point source of sound station acquisition; Totally 2016 groups, each distance has 672 groups of data, for the basic improved CSP method of weighting function of not carrying out; It directly utilizes original weighting function when calculating, experiment finds that the effect of this location algorithm is better than the CSP method.Experimental data is seen table 1.

The experimental data comparison sheet of table 1 CSP method and the inventive method

Claims (10)

1. the continuous sound localization method of robot comprises the steps:

1) the robot microphone array that adopts at least two pairs of microphones to constitute is gathered sound-source signal, and is to store after the digital signal with conversion of signals;

2) signal of gathering is carried out effective sound signal and detect, the signal that will include effective audio frequency carries out Fourier transform, converts frequency domain into from time domain;

3) signal that converts frequency domain into is carried out fragment process, the signal data that the every pair of microphone is obtained carries out correlation calculations respectively, and the result of correlation calculations is carried out inverse Fourier transform, is transformed into time domain from frequency domain;

4) carry out the maximum value position search, obtain sound source to the right mistiming τ of each microphone;

5) calculate the position of sound source with respect to robot according to angle θ and mistiming τ, wherein θ is initial point and the line of sound source and the angle between the coordinate system x axle forward of microphone array coordinate system of living in.

2. the continuous sound localization method of robot according to claim 1 is characterized in that, when said robot and sound source have relative motion, to f) the angle θ that calculates of step revises, and modification method is:

A, transfer angle when preceding the N time computing acquisition of time auditory localization computing;

B, obtain to calculate the angular velocity Ad/ (N*Tf) of sound source and robot relative motion as the angle θ of time computing and the poor Ad of the angle of preceding the N time computing, Tf is the mistiming of working as inferior computing and preceding the N time computing;

C, the current angle of correction are:

θ+Td*Ad/(N*Tf)

Wherein, Td is that robot calculates the needed time of angle at every turn.

3. the continuous sound localization method of robot according to claim 1 and 2 is characterized in that, effective audio signal data that said auditory localization computing whenever collects set amount in time domain just carries out the computing of an auditory localization.

4. the continuous sound localization method of robot according to claim 3 is characterized in that said effective audio signal data is a speech data.

5. the continuous sound localization method of robot according to claim 4 is characterized in that, the data of said set amount are the data of frame length size.

6. the continuous sound localization method of robot according to claim 1 and 2 is characterized in that, said machine people is mobile robot or stationary machines people.

7. the continuous sound localization method of robot according to claim 1 and 2 is characterized in that, said sound source is stationary sound source or mobile sound source.

8. the continuous sound localization method of robot according to claim 1 and 2 is characterized in that said Fourier transform is Fast Fourier Transform (FFT).

9. the continuous sound localization method of robot according to claim 1 and 2 is characterized in that the microphone array of said robot is classified cruciform array, triangular array, circular array or linear array as.

10. the continuous sound localization method of robot according to claim 1 and 2 is characterized in that, described fragment process is that all Frequency points after the Fourier transform are divided into different groups.

Images