A broadband signal source positioning method and system

Abstract

The present invention provides a method and system for locating a broadband signal source, relating to the field of speech signal processing. The method comprises obtaining microphone signals; performing a short-time Fourier transform on the microphone signals to obtain time-frequency domain signals; processing the time-frequency domain signals of each pair of microphones using a recursive least squares method to obtain an early reverberation arrival path relative transfer function; constructing a broadband signal source covariance matrix based on the early reverberation arrival path relative transfer function; and determining the broadband signal source position using a MUSIC algorithm based on the broadband signal source covariance matrix. The present invention can improve the accuracy of broadband signal source positioning.

Description

Technical Field

[0001] The present invention relates to the field of speech signal processing, and in particular to a broadband signal source positioning method and system. Background Art

[0002] The covariance matrix of the signal received by the microphone array is very critical for realizing functions such as broadband signal source positioning and signal enhancement. Traditional methods will be interfered by reverberation and noise when calculating the broadband signal source covariance matrix. Traditional methods calculate the broadband signal source covariance matrix directly through the observation signal. However, the observation signal not only contains the direct wave of the broadband signal source signal, but also contains the echo component and noise component after the broadband signal source signal is reflected. If the above-mentioned broadband signal source covariance matrix is ​​used to locate the broadband signal source, it may be detected that the broadband signal source signal arrives from multiple directions, affecting the positioning performance. Therefore, a more accurate broadband signal source positioning method is needed. Summary of the Invention

[0003] The object of the present invention is to provide a method and system for locating a broadband signal source, so as to improve the accuracy of locating the broadband signal source.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A broadband signal source positioning method, comprising:

[0006] Get microphone signal;

[0007] Performing short-time Fourier transform on the microphone signal to obtain a time-frequency domain signal;

[0008] Processing the time-frequency domain signals of each pair of microphones using a recursive least squares method to obtain an early reverberation arrival path relative transfer function;

[0009] Constructing a broadband signal source covariance matrix according to the relative transfer function of the early reverberation arrival path;

[0010] The broadband signal source position is determined using a MUSIC algorithm according to the broadband signal source covariance matrix.

[0011] Optionally, performing short-time Fourier transform on the microphone signal to obtain a time-frequency domain signal specifically includes:

[0012] Performing a short-time Fourier transform on the microphone signal to obtain an observation signal in the STFT domain;

[0013] The observation signal in the STFT domain is vector-constructed and phase-shifted to obtain a time-frequency domain signal.

[0014] Optionally, the processing of the time-frequency domain signals of each pair of microphones by using a recursive least squares method to obtain an early reverberation arrival path relative transfer function specifically includes:

[0015] determining a fitting error based on the time-frequency domain signals of each pair of microphones;

[0016] determining a cost function of the recursive least squares method according to the fitting error;

[0017] The early reverberation arrival path relative transfer function is determined by iterating using a recursive least square method according to the cost function and the time-frequency domain signals of each pair of microphones.

[0018] Optionally, determining the broadband signal source position by using a MUSIC algorithm according to the broadband signal source covariance matrix specifically includes:

[0019] Performing eigenvalue decomposition on the covariance matrix of the broadband signal source to obtain a noise subspace and a signal subspace;

[0020] Determine an array space spectrum function using a MUSIC algorithm according to the noise subspace and the signal subspace;

[0021] The location of the broadband signal source is determined according to the array spatial spectrum function.

[0022] A broadband signal source positioning system, comprising:

[0023] An acquisition module is used to acquire microphone signals;

[0024] A short-time Fourier transform module, configured to perform short-time Fourier transform on the microphone signal to obtain a time-frequency domain signal;

[0025] an early reverberation arrival path relative transfer function determination module, configured to process the time-frequency domain signals of each pair of microphones using a recursive least squares method to obtain an early reverberation arrival path relative transfer function;

[0026] A construction module, configured to construct a broadband signal source covariance matrix according to the early reverberation arrival path relative transfer function;

[0027] The broadband signal source position determination module is used to determine the broadband signal source position using a MUSIC algorithm according to the broadband signal source covariance matrix.

[0028] Optionally, the short-time Fourier transform module specifically includes:

[0029] A short-time Fourier transform unit, configured to perform a short-time Fourier transform on the microphone signal to obtain an observation signal in the STFT domain;

[0030] The time-frequency domain signal determination unit is used to construct a vector and perform phase shift on the observation signal in the STFT domain to obtain a time-frequency domain signal.

[0031] Optionally, the early reverberation arrival path relative transfer function determination module specifically includes:

[0032] a fitting error determining unit, configured to determine a fitting error based on the time-frequency domain signals of each pair of microphones;

[0033] a cost function determining unit, configured to determine a cost function of the recursive least squares method according to the fitting error;

[0034] The early reverberation arrival path relative transfer function determination unit is configured to iteratively determine the early reverberation arrival path relative transfer function by using a recursive least squares method according to the cost function and the time-frequency domain signals of each pair of microphones.

[0035] Optionally, the broadband signal source location determination module specifically includes:

[0036] an eigenvalue decomposition unit, configured to perform eigenvalue decomposition on the broadband signal source covariance matrix to obtain a noise subspace and a signal subspace;

[0037] an array space spectrum function determining unit, configured to determine the array space spectrum function using a MUSIC algorithm according to the noise subspace and the signal subspace;

[0038] A broadband signal source position determining unit is configured to determine the broadband signal source position according to the array space spectrum function.

[0039] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0040] The present invention obtains microphone signals; performs short-time Fourier transform on the microphone signals to obtain time-frequency domain signals; processes the time-frequency domain signals of each pair of microphones using a recursive least squares method to obtain an early reverberation arrival path relative transfer function; constructs a broadband signal source covariance matrix based on the early reverberation arrival path relative transfer function; and determines the position of the broadband signal source using a MUSIC algorithm based on the broadband signal source covariance matrix. The present invention reconstructs the broadband signal source covariance matrix using the early reverberation arrival path relative transfer function of the broadband signal source signal, thereby reducing the interference of reverberation on the covariance matrix, making it more conducive to broadband signal source positioning, and thus improving the accuracy of broadband signal source positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 Flowchart of the broadband signal source positioning method provided by the present invention;

[0043] Figure 2 This is a flow chart for constructing the covariance matrix of a broadband signal source provided by the present invention. DETAILED DESCRIPTION

[0044] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0045] The object of the present invention is to provide a method and system for locating a broadband signal source, so as to improve the accuracy of locating the broadband signal source.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] To introduce existing methods for estimating the covariance of wideband signal sources, it is necessary to introduce the representation of wideband signal sources in the time-frequency domain (STFT), which inevitably requires the introduction of some mathematical formulas. In the field of speech signal processing, the microphone received signal is often represented in the STFT domain as follows:

[0048] Y(p,k)=X(p,k)h(k)+V(p,k) (1)

[0049] in

[0050] Y(p,k)=[Y1(p,k),Y2(p,k)…,Y M (p,k)] T ,

[0051] h(k)=[h1(k),h2(k)…,h M (k)] T ,

[0052] V(p,k)=[V1(p,k),V2(p,k)…,V M (p,k)]T .

[0053] Wherein, the subscript M represents the number of microphones, p represents the number of frames, k represents the frequency, Y(p,k) is the observation vector, h(k) is the steering vector, V(p,k) is the noise vector, and Y m (p,k),V m (p, k) represent the short-time Fourier transform of the observation signal and noise signal of the mth microphone at the pth frame, respectively, h m (k) represents the Fourier transform of the impulse response function of the broadband signal source arriving at the mth microphone. X(p,k) is the short-time Fourier transform of the broadband signal source, a scalar quantity. The superscript T denotes the transpose. A fundamental assumption in speech signal processing is that X(p,k) and V(p,k) are uncorrelated.

[0054] The covariance matrix of the observed signal can be expressed as follows:

[0055] φ YY (p,k)=E[Y(p,k)Y H (p,k)] (2)

[0056] φ YY (p,k_=φ XX (p,k)h(k)h H (k)+φ VV (p,k) (3)

[0057] Where E stands for expectation, superscript H stands for conjugate transpose, φ XX (p,k)=E[X(p,k)X T (p,k)] is the variance of the broadband source signal, φ VV (p,k)=E[V(p,k)V H (p,k)] is the covariance matrix of the noise.

[0058] Usually use and Approximately estimate the covariance matrix of the observed signal and noise:

[0059]

[0060]

[0061] where λ∈(0,1] is the forgetting factor.

[0062] like Figure 1 As shown, the present invention provides a broadband signal source positioning method, comprising:

[0063] Step 101: Acquire microphone signals.

[0064] Step 102: Perform short-time Fourier transform on the microphone signal to obtain a time-frequency domain signal.

[0065] Step 102 specifically includes:

[0066] Performing short-time Fourier transform on the microphone signal to obtain an observation signal in the STFT domain.

[0067] The observation signal in the STFT domain is vector-constructed and phase-shifted to obtain a time-frequency domain signal.

[0068] Step 103: Process the time-frequency domain signals of each pair of microphones using a recursive least squares method to obtain an early reverberation arrival path relative transfer function.

[0069] Step 103 specifically includes:

[0070] A fitting error is determined based on the time-frequency domain signals of each pair of microphones.

[0071] A cost function of the recursive least squares method is determined according to the fitting error.

[0072] The early reverberation arrival path relative transfer function is determined by iterating using a recursive least square method according to the cost function and the time-frequency domain signals of each pair of microphones.

[0073] Step 104: Constructing a broadband signal source covariance matrix according to the early reverberation arrival path relative transfer function.

[0074] like Figure 2 As shown, the present invention generally reconstructs the broadband signal source covariance matrix from the observed signal in three steps.

[0075] Step 1: transform the observed signal from the time domain to the time-frequency domain

[0076] Assume that the input signal of a microphone is y m (t), then its short-time Fourier transform can be expressed as follows:

[0077]

[0078] in

[0079]

[0080] t is the discrete time point of the time domain discrete signal, t is an integer, is the analysis window function of length T, N is the length of the window function, p is the current frame number, k is the current frequency, L is the sliding step size of the analysis window, and * represents the complex conjugate. Similarly, the STFT domain representation of the signals received by other microphones can be obtained. for The analysis window function after shifting pL discrete time points to the right.

[0081] Using the convolution transfer function (CTF) approximate estimation method, under the condition of single broadband signal source and no noise, the received signal y m (t) is expressed in the STFT domain as:

[0082]

[0083] The value of coefficient Q is related to the reverberation time and the length of the analysis window function, h m (p, k) is the short-time Fourier transform of the impulse response function of the broadband signal source reaching the mth microphone, and q is the coefficient of the frame. Similarly, the other microphones also have the above form.

[0084] First, we analyze a pair of microphones (i, j) and use formula (8) to get: Y i (p,k)*h j (p,k)=Y j (p,k)*h i (p,k). Among them, Y i (p,k) is y i (t) Representation in the STFT domain, h j (p,k) is the unit impulse response from the broadband signal source to the jth microphone represented in the STFT domain, h i (p, k) is the unit impulse response from the broadband signal source to the i-th microphone expressed in the STFT domain, Y j (p,k) is y j (t) is represented in the STFT domain and written as a vector:

[0085]

[0086] in

[0087] Y i (p,k)=(Y i (p,k),…,Y i (p-Q+1,k)) T

[0088] Y j (p,k)=(Y j (p,k),…,Y j (p-Q+1,k)) T

[0089] h j(k)=(h j (0,k),…,h j (Q-1,k)) T

[0090] h i (k)=(h i (0,k),…,h i (Q-1,k))

[0091] The impulse response from the broadband signal source to all microphones is written as a vector:

[0092]

[0093] For M microphones, we get M(M-1) / 2 microphone pairs, and use coefficients (i, j) to represent a microphone pair, where i = 1, ..., M-1, j = i+1, ..., M. In order to unify the form of all microphone pairs, we can construct a column vector Y for each pair of microphones ij (p,k), whose dot product with the column vector h(k) is 0.

[0094]

[0095] in

[0096]

[0097] To avoid the solution h(k) = 0 in the above equation, first divide both sides by h1(0,k), and then transform the column vector Y ij The first coefficient in (p,k) is called -Z ij , perform phase shift, and finally get:

[0098]

[0099] in

[0100]

[0101]

[0102] The above formula can be understood as filtering the input signal to obtain the desired output. is the input signal, which is Y ij (p,k) is in the form after the first term is removed, are the filter coefficients and are a vector associated with the DP-RTF.

[0103] Step 2: Apply the recursive least squares (RLS) algorithm to estimate the early path relative transfer function (DP-RTF)

[0104] The DP-RTFs for each microphone pair can be obtained from Therefore, to obtain the DP-RTF estimate, we need to solve Equation (12). For the convenience of formula expression, let c = 1, ..., C be the number of each pair of microphones, C = M (M-1) / 2, and use c to replace the coefficient ij. The fitting error is defined as:

[0105]

[0106] Among them, Z c (p,k) is Z ij (p,k), that is, the column vector Y ij The first coefficient in (p,k).

[0107] When the current frame number is p, for microphone pair c, the RLS cost function can be written as follows:

[0108]

[0109] Here, p' is the coefficient of the past frame and c' is the coefficient of the microphone pair.

[0110] The cost function combines the fitting errors of microphone pairs in all past frames and the fitting errors of the 1st to cth microphone pairs in the current frame, where λ∈(0,1]. To minimize J c (p,k), the cost function is Take the derivative and set it to 0. Finally we get The expression when calculating the c-th microphone pair in the p-th frame is:

[0111]

[0112] in,

[0113]

[0114]

[0115] R c (p,k) is the autocorrelation matrix of the input, r c (p,k) is the cross-correlation vector between the input and the expected output.

[0116] make The Sherman-Morrison formula is used to indirectly invert the matrix, thereby greatly reducing the amount of calculation. The RLS iterative algorithm is shown in Table 1, where the RLS iterative algorithm is an existing technology.

[0117] Table 1 RLS iterative algorithm table

[0118]

[0119] Take out all The first term in , the early reverberation arrival path relative transfer function (DP-RTF) can be obtained as follows:

[0120]

[0121] is the DP-RTF referenced to the first microphone.

[0122] Step 3: Construct the broadband signal source covariance matrix based on DP-RTF

[0123] After obtaining the estimation of DP-RTF with different microphones as reference, the covariance matrix of the broadband signal source, i.e., the broadband signal source covariance matrix, is calculated according to the following formula:

[0124]

[0125] is the DP-RTF with the i-th microphone as the reference.

[0126] Step 105: Determine the location of the broadband signal source using a MUSIC algorithm according to the broadband signal source covariance matrix.

[0127] Step 105 specifically includes:

[0128] The broadband signal source covariance matrix is ​​subjected to eigenvalue decomposition to obtain a noise subspace and a signal subspace.

[0129] An array space spectrum function is determined by using a MUSIC algorithm according to the noise subspace and the signal subspace.

[0130] The location of the broadband signal source is determined according to the array spatial spectrum function.

[0131] The covariance matrix is ​​used in the MUSIC algorithm for positioning. First, the covariance matrix Φ dd (p,k) is decomposed into two subspaces, which are as follows:

[0132]

[0133] Among them, Σ S Corresponding to the larger eigenvalue of the covariance matrix, Σ N For the remaining eigenvalues, U S and U N Corresponding to Σ S and Σ N The eigenvector of U SThe spanned subspace is called the signal subspace, which is represented by U N The spanned subspace is called the noise subspace. Based on the relevant knowledge of eigenvalues ​​and eigenvectors, it can be known that the signal subspace and the noise subspace are orthogonal to each other.

[0134] From the MUSIC algorithm, we know that the signal arrival direction vector a(k) is orthogonal to the noise subspace. Based on the orthogonal relationship between the noise subspace and the signal subspace, we can get the array space spectrum function:

[0135]

[0136] Among them, θ d is the signal arrival direction, a(k,θ d ) is the arrival direction of the signal at frequency k, θ d The direction vector of the broadband signal source reaching the microphone. According to formula (19), all the θ d , looking for the peak to estimate the arrival direction of the broadband signal source, thereby determining the location of the broadband signal source.

[0137] The present invention also provides a broadband signal source positioning system corresponding to the broadband signal source positioning method, comprising:

[0138] The acquisition module is used to obtain microphone signals.

[0139] The short-time Fourier transform module is used to perform short-time Fourier transform on the microphone signal to obtain a time-frequency domain signal.

[0140] The early reverberation arrival path relative transfer function determination module is used to process the time-frequency domain signals of each pair of microphones using a recursive least squares method to obtain the early reverberation arrival path relative transfer function.

[0141] A construction module is used to construct a broadband signal source covariance matrix according to the early reverberation arrival path relative transfer function.

[0142] The broadband signal source position determination module is used to determine the broadband signal source position using a MUSIC algorithm according to the broadband signal source covariance matrix.

[0143] As an optional implementation manner, the short-time Fourier transform module specifically includes:

[0144] A short-time Fourier transform unit is used to perform short-time Fourier transform on the microphone signal to obtain an observation signal in the STFT domain.

[0145] The time-frequency domain signal determination unit is used to construct a vector and perform phase shift on the observation signal in the STFT domain to obtain a time-frequency domain signal.

[0146] As an optional implementation manner, the early reverberation arrival path relative transfer function determination module specifically includes:

[0147] The fitting error determining unit is configured to determine a fitting error according to the time-frequency domain signals of each pair of microphones.

[0148] A cost function determining unit is used to determine the cost function of the recursive least squares method according to the fitting error.

[0149] The early reverberation arrival path relative transfer function determination unit is configured to iteratively determine the early reverberation arrival path relative transfer function by using a recursive least squares method according to the cost function and the time-frequency domain signals of each pair of microphones.

[0150] As an optional implementation manner, the broadband signal source location determination module specifically includes:

[0151] The eigenvalue decomposition unit is used to perform eigenvalue decomposition on the covariance matrix of the broadband signal source to obtain a noise subspace and a signal subspace.

[0152] The array space spectrum function determining unit is configured to determine the array space spectrum function by using a MUSIC algorithm according to the noise subspace and the signal subspace.

[0153] A broadband signal source position determining unit is configured to determine the broadband signal source position according to the array space spectrum function.

[0154] The present invention uses the convolution transfer function (CTF) to calculate the covariance matrix of a broadband signal source for broadband signal source localization. This method can effectively reduce the reverberation component in the broadband signal source covariance matrix when the indoor impulse response (RIR) time is long, making it more suitable for functions such as localization and speech enhancement. The broadband signal source covariance matrix is ​​re-estimated using the early path transfer function of the broadband signal source, reducing the interference of reverberation on this covariance matrix. This results in fewer echo components in the broadband signal source covariance matrix, improving robustness in reverberant environments and facilitating broadband signal source localization and speech enhancement.

[0155] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0156] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A broadband signal source positioning method, characterized in that: include: Get microphone signal; Performing short-time Fourier transform on the microphone signal to obtain a time-frequency domain signal; Processing the time-frequency domain signals of each pair of microphones using a recursive least squares method to obtain an early reverberation arrival path relative transfer function; A broadband signal source covariance matrix is ​​constructed based on the early reverberation arrival path relative transfer function; wherein, after obtaining an estimate of the DP-RTF with reference to different microphones, the broadband signal source covariance matrix is ​​calculated according to the following formula: is the DP-RTF with the i-th microphone as the reference; Determine the location of the broadband signal source using a MUSIC algorithm according to the broadband signal source covariance matrix; The time-frequency domain signals of each pair of microphones are processed using a recursive least squares method to obtain an early reverberation arrival path relative transfer function, specifically comprising: determining a fitting error based on the time-frequency domain signals of each pair of microphones; determining a cost function of the recursive least squares method according to the fitting error; The recursive least squares method is used to iterate the cost function and the time-frequency domain signals of each pair of microphones to determine the relative transfer function of the early reverberation arrival path; wherein, in order to unify the form of all microphone pairs, a column vector Y is constructed for each pair of microphones. ij (p,k), Y ij The dot product of (p,k) and the column vector h(k) is 0: in, Q is the coefficient. To avoid the solution h(k)=0 in the above equation, first divide both sides by h1(0,k), and then ij The first coefficient in (p,k) is called -Z ij , perform phase shift, and finally get: in, The above formula is understood as filtering the input signal to obtain the desired output. is the input signal, which is Y ij (p,k) is in the form after the first term is removed, is the filter coefficient and is a vector related to DP-RTF; DP-RTFs for each microphone pair from Therefore, to obtain the DP-RTF estimate, we need to Solve; take out all The first item in , we get DP-RTF as follows: is the DP-RTF referenced to the first microphone.

2. The broadband signal source positioning method according to claim 1, wherein: The performing of a short-time Fourier transform on the microphone signal to obtain a time-frequency domain signal specifically includes: Performing a short-time Fourier transform on the microphone signal to obtain an observation signal in the STFT domain; The observation signal in the STFT domain is vector-constructed and phase-shifted to obtain a time-frequency domain signal.

3. The broadband signal source positioning method according to claim 1, wherein: Determining the location of the broadband signal source by using a MUSIC algorithm according to the broadband signal source covariance matrix specifically includes: Performing eigenvalue decomposition on the covariance matrix of the broadband signal source to obtain a noise subspace and a signal subspace; Determine an array space spectrum function using a MUSIC algorithm according to the noise subspace and the signal subspace; The location of the broadband signal source is determined according to the array spatial spectrum function.

4. A broadband signal source positioning system, characterized in that: include: An acquisition module is used to acquire microphone signals; A short-time Fourier transform module, configured to perform short-time Fourier transform on the microphone signal to obtain a time-frequency domain signal; an early reverberation arrival path relative transfer function determination module, configured to process the time-frequency domain signals of each pair of microphones using a recursive least squares method to obtain an early reverberation arrival path relative transfer function; A construction module is used to construct a broadband signal source covariance matrix based on the early reverberation arrival path relative transfer function; wherein, after obtaining an estimate of the DP-RTF with different microphones as references, the broadband signal source covariance matrix is ​​calculated according to the following formula: is the DP-RTF with the i-th microphone as the reference; A broadband signal source position determination module, configured to determine the broadband signal source position using a MUSIC algorithm based on the broadband signal source covariance matrix; The early reverberation arrival path relative transfer function determination module specifically includes: a fitting error determining unit, configured to determine a fitting error based on the time-frequency domain signals of each pair of microphones; a cost function determining unit, configured to determine a cost function of the recursive least squares method according to the fitting error; The early reverberation arrival path relative transfer function determination unit is used to iteratively determine the early reverberation arrival path relative transfer function using a recursive least squares method based on the cost function and the time-frequency domain signals of each pair of microphones; wherein, in order to unify the form of all microphone pairs, a column vector Y is constructed for each pair of microphones. ij (p,k), Y ij The dot product of (p,k) and the column vector h(k) is 0: in, Q is the coefficient. To avoid the solution h(k)=0 in the above equation, first divide both sides by h1(0,k), and then ij The first coefficient in (p,k) is called -Z ij , perform phase shift, and finally get: in, The above formula is understood as filtering the input signal to obtain the desired output. is the input signal, which is Y ij (p,k) is in the form after the first term is removed, is the filter coefficient and is a vector related to DP-RTF; DP-RTFs for each microphone pair from Therefore, to obtain the DP-RTF estimate, we need to Solve; take out all The first term in the early reverberation arrival path relative transfer function DP-RTF is obtained as follows: is the DP-RTF referenced to the first microphone.

5. The broadband signal source positioning system according to claim 4, characterized in that: The short-time Fourier transform module specifically includes: A short-time Fourier transform unit, configured to perform a short-time Fourier transform on the microphone signal to obtain an observation signal in the STFT domain; The time-frequency domain signal determination unit is used to construct a vector and perform phase shift on the observation signal in the STFT domain to obtain a time-frequency domain signal.

6. The broadband signal source locating system according to claim 4, characterized in that: The broadband signal source position determination module specifically includes: an eigenvalue decomposition unit, configured to perform eigenvalue decomposition on the broadband signal source covariance matrix to obtain a noise subspace and a signal subspace; an array space spectrum function determining unit, configured to determine the array space spectrum function using a MUSIC algorithm according to the noise subspace and the signal subspace; A broadband signal source position determining unit is configured to determine the broadband signal source position according to the array space spectrum function.

Images