JP2005260744A - Method and apparatus for synchronizing phases of microphone reception signals in microphone array - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method and apparatus for synchronizing phases of microphone reception signals in a microphone array in which phases of reception signals in microphones constituting the microphone array can be highly accurately synchronized. <P>SOLUTION: The apparatus for synchronizing phases of microphone reception signals in a microphone array including three or more microphones mounted in portable equipment and disposed on an xy-plane of a three-dimensional coordinate system with the portable equipment as a reference, includes a first means for calculating a coordinate estimate of a sound source in the three-dimensional coordinate system at predetermined time intervals and a second means for converting reception signals of the microphones into phase-synchronized signals based on the coordinate estimate of the sound source calculated by the first means, wherein the first means utilizes a natural gradient learning method to estimate coordinates of the sound source so as to minimize a difference of phase synchronized signals obtained by the second means. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to an in-phase method and an in-phase apparatus for microphone received signals in a microphone array.

  The inventors of the present application are studying for the purpose of receiving a high quality sound of a remote utterance voice in a situation where the spatial position of the microphone and the arrival direction of the sound source change every moment.

  The inventors of the present application have proposed a method for estimating the direction of the microphone array relative to the direction of arrival of the sound source and correcting the time difference between the received signals (see reference [1]). This method adaptively estimates the rotation angle of two coordinate axes representing the spatial position of the microphone based on the minimization of one evaluation function.

  Reference [1]: Horiuchi, Mizumachi, Nakamura, “Sequential time difference correction algorithm for multiple microphones,” Vol. I, pp. 691-692, Mar. 2003.

  Most of the conventional direction estimation methods up to that point (see reference [2]) perform estimation on a frame basis, but the above proposed method proposed by the inventors of the present application performs estimation for each sample. Yes. This is one of the features of the proposed method.

  Reference [2]: Oga, Yamazaki, Kaneda, “Acoustic system and digital processing,” IEICE, 1995.

  However, the proposed method proposed by the inventors of the present application uses an LMS algorithm, and the stability and tracking speed largely depend on the step size parameter. is required.

  By the way, in Reference [3], Amari et al. Considered the reason why the steepest descent method (probability descent method) using a gradient on a space with a singular point does not give the optimum gradient direction, and used Riemannian metric. A natural gradient learning method is proposed.

Reference [3]: Amari, “Natural Gradient Works Efficiently in Learning,” Neural Computation, vol. 10, pp. 251-276, 1998.
Horiuchi, Mizumachi, Nakamura, “Sequential Time Difference Correction Algorithm for Multiple Microphone Signals,” Sound Lecture, Vol. I, pp. 691-692, Mar. 2003. Oga, Yamazaki, Kaneda, "Acoustic systems and digital processing," IEICE, 1995. Amari, “Natural Gradient Works Efficiently in Learning,” Neural Computation, vol. 10, pp. 251-276, 1998.

  An object of the present invention is to provide a method and an apparatus for in-phase a microphone sound reception signal in a microphone array, which can make the sound reception signal of each microphone constituting the microphone array in phase with high accuracy.

  According to the first aspect of the present invention, in-phase microphone sound reception signals in a microphone array having three or more microphones mounted on a portable device and arranged on the xy plane of a three-dimensional coordinate system based on the portable device. In the control apparatus, the first means for calculating the coordinate estimate value of the sound source in the three-dimensional coordinate system at a predetermined time, and the sound reception of each microphone based on the coordinate estimate value of the sound source calculated by the first means A second means for converting the signal into an in-phase signal, wherein the first means uses a natural gradient learning method so that the difference between the in-phase signals obtained by the second means is minimized; It is characterized by estimating the coordinates of the sound source.

  The invention according to claim 2 is the same phase of the microphone sound reception signal in the microphone array having three or more microphones mounted on the portable device and arranged on the xy plane of the three-dimensional coordinate system with reference to the portable device. In the conversion method, the first step of calculating the coordinate estimate value of the sound source in the three-dimensional coordinate system at a predetermined time, and the sound reception of each microphone based on the coordinate estimate value of the sound source calculated by the first step A second step of converting the signal into an in-phase signal, wherein the first step uses a natural gradient learning method so that the difference in the in-phase signal obtained by the second step is minimized. It is characterized by estimating the coordinates of the sound source.

  According to the present invention, the received sound signals of the microphones constituting the microphone array can be in phase with high accuracy.

  Embodiments of the present invention will be described below with reference to the drawings.

[1] Definition of Microphone Array Coordinate System FIG. 1 shows a microphone array coordinate system.

The microphone array is mounted on a portable device. Three or more microphones M _i (i = 1, 2,..., M) are arranged at arbitrary positions on the xy plane of a three-dimensional space (three-dimensional coordinate system) based on the portable device. The coordinates (polar coordinates) (r _i , π / 2, φ _i ) of each microphone M _i are given in advance.

  The coordinates (polar coordinates) of the sound source S are (R, θ, φ). Here, it is assumed that the sound source is single.

Arrival time tau _i from the sound source S to the microphone M _i is the distance d (S, M _i) between the sound source S and Ikurohon M _i using a can be expressed by the following equation (1).

  Where c is the speed of sound.

Further, using this arrival time τ _i , the sound reception signal x _i (t) of each microphone M _i can be expressed as the following equation (2).

Here, s (t) represents a sound source signal, and n _i (t) represents observation noise.

[2] Description of the in-phase circuit for in-phase the microphone sound reception signal in the microphone array FIG. 2 shows an in-phase circuit for in-phase the microphone sound reception signal in the microphone array.

The sound reception signal x _i (t) of each microphone M _i is sent to the in-phase circuit 10. The in-phase circuit 10 corrects the sound reception signal x _i (t) so that the delay time difference between the sound reception signals x _i (t) becomes zero. That is, the in-phase circuit 10 in-phases each sound reception signal x _i (t). Then, the in-phase signal is output as y _i (t). Hereinafter, processing performed by the in-phase circuit 10 will be described.

First, consider in-phase the received sound signal x _i (t). Using the Sinc function, the in-phased signal y _i (t) can be expressed as the following equation (3).

However, Sinc (x) = sin (πx) / πx. N is a filter length, D is a fixed delay, and T is a sampling interval. ^ Τ _i is an arrival time τ _i calculated from estimated coordinates (^ R, ^ θ, ^ φ) of the sound source S described later.

With respect to y _i (t), a square error E like the following equation (4) is defined.

  The estimated coordinates (^ R, ^ θ, ^ φ) of the sound source S are given by the following equation (5) as R, θ, and φ when the square error E is minimized.

  In order to minimize the square error E, a natural gradient learning method is introduced. Estimated coordinates (^ R, ^ θ, ^ φ) of the sound source S are defined as in the following equation (6).

  Here, T represents transposition.

  The minute change amount dω of the coordinates of the sound source S is given by the following equation (7).

When the natural gradient learning method is used, the (n + 1) -th estimated coordinate ω _{n + 1} is given by the following equation (8) using the coordinate ω _n estimated at the n-th time.

Here, G ⁻¹ is given by the following equation (9).

In the above equation (8), ∇ (nabla) is a differential operator. G ⁻¹ ∇E (ω _n ) in the above equation (8) corresponds to a minute change amount of the coordinates of the sound source S.

First, the in-phase circuit 10 calculates the estimated coordinates (^ R, ^ θ, ^ φ) of the current sound source S based on the above equation (8). Next, based on the estimated coordinates (^ R, ^ θ, ^ φ) of the obtained sound source S and the coordinates (r _i , π / 2, φ _i ) of each microphone M _i , the above equation (1) is used. , to calculate the arrival time τ _i from the sound source S to the microphone M _i. Then, based on the calculated arrival time τ _i , the in-phase signal y _i (t) is calculated from the above equation (3).

[3] Verification of basic performance The basic performance of the algorithm for estimating the position of the sound source S used in the above embodiment is verified by computer simulation.

First, several conditions were set for computer simulation. The microphone has coordinates (−d, 3 ^1/2 d / 3, 0), (d, 3 ^1/2 d / 3, 0) (−d, 2 × 3 ^1/2 d /) on the xy plane. 3, three). However, d is 0.05 m.

  For the target sound source signal, shaped gaussian noise having an average of 0, a variance of 0.05, and a band of 125 Hz-6 kHz was used. The noise signal uses random band noise in the same band as the target sound source signal, and is uncorrelated between the channels and the target sound source signal. The signal used was quantized with a sampling frequency of 48 kHz and 16 bits.

It is assumed that the target sound source signal has a constant direction of θ of 45 deg and φ alternately arrives from directions of 30 deg and −30 deg, and the sound reception signal x _i (i = 1, 2, 3) is received by each microphone M _i. t) was created by giving appropriate time shifts to the target sound source signal and noise signal on the computer and adding them together.

  The initial values of θ and φ were 0 deg, and the initial value of R was 1. Moreover, it compared with the conventional method (method shown in the reference [1]) based on the LMS algorithm. The step size parameter μ in the conventional method was set to 0.01.

  FIG. 3 shows a direction estimation result when the SNR of the target sound source signal with respect to the noise signal is 20 dB. The lower part of FIG. 3 shows an enlarged part of the upper part of FIG. However, the horizontal axis represents time, and the vertical axis represents the estimated arrival direction.

  From FIG. 3, it can be confirmed that the method (proposed algorithm) according to the present embodiment estimates the sound source direction at a higher speed than the conventional method (Conventional) based on the LMS algorithm. In general, the selection of the step size parameter is not easy as it relates to system stability and convergence speed. In the proposed algorithm, the natural gradient learning method is used, so that the coefficient update amount can be appropriately varied at each time and estimated coordinates, and the tracking speed can be increased.

It is a schematic diagram which shows the coordinate system of a microphone array. It is a block diagram which shows the structure of the circuit for making in-phase the microphone sound-receiving signal in a microphone array. It is a graph which shows a simulation result.

Explanation of symbols

M _i microphone S sound source 10-phase circuit

Claims (2)

In an apparatus for phase-synchronizing a microphone sound signal in a microphone array having three or more microphones mounted on a portable device and arranged on an xy plane of a three-dimensional coordinate system based on the portable device.
The first means for calculating the coordinate estimate value of the sound source in the three-dimensional coordinate system at every predetermined time, and the sound reception signal of each microphone based on the coordinate estimate value of the sound source calculated by the first means Second means for converting into a normalized signal, and the first means uses a natural gradient learning method to determine the coordinates of the sound source that minimizes the difference in the in-phase signal obtained by the second means. An in-phase apparatus for receiving a microphone sound signal in a microphone array, wherein the apparatus is an estimation device. In the method of phase-synchronizing a microphone sound reception signal in a microphone array having three or more microphones mounted on a portable device and arranged on an xy plane of a three-dimensional coordinate system based on the portable device.
The first step of calculating the coordinate estimate value of the sound source in the three-dimensional coordinate system at predetermined time intervals, and the sound reception signal of each microphone in-phase based on the coordinate estimate value of the sound source calculated by the first step A second step of converting the signal into a normalized signal, and the first step uses a natural gradient learning method to determine the coordinates of the sound source that minimizes the difference in the in-phase signal obtained in the second step. An in-phase method for receiving a microphone sound signal in a microphone array, wherein the method is an estimation method.

Images