JP2006058395A - Sound signal input/output device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a sound signal input/output device for voice recognition devices of various systems each of which is equipped with a speech recognition function. <P>SOLUTION: The sound signal input/output device 30 is equipped with a sound signal input means 1 which is constituted by combining microphones 1a to 1d with prescribed intervals d apart, an A/D converter 2 which converts analog sound signals Sa and Sd of the respective microphones 1a to 1d to digital sound signals Da to Dd, a sound source localization means 3 which detects the direction of a sound source by analyzing the converted digital sound signals Da to Dd, a sound source angle deciding means 11 which decides whether the sound source A localized by the sound source localization means 3 is the sound source from a direction within a preset angle range δ, a voice detecting means 12 which decides whether the sound source A is human voice or not by analyzing the digital sound signals Dd, and a gate circuit 20 which outputs the digital sound signals Dd only when the sound source A is in the direction within the preset angle δ and the human voice is included therein. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a technology of an input / output interface of an acoustic signal to a voice recognition device of various systems in which recognition of a human voice and a direction of a sound source are important, such as a car navigation system or a human authentication system having a voice recognition function. It is.

  Conventionally, the input of an acoustic signal to a voice recognition device (speech recognition circuit) built in a car navigation system or a human authentication system having a voice recognition function generally includes the position of the face of a person who is in a predetermined place. The acoustic signal picked up by one microphone directed in the expected direction is input as it is.

  For example, in a car navigation system having a voice recognition function, a voice recognition device is provided by picking up a driver's voice with a directional microphone directed to the vicinity of the driver's mouth disposed on the ceiling, dashboard or steering wheel of the vehicle. The voice (driver's voice (instruction command)) picked up by the microphone is analyzed and transmitted to the system controller (control microcomputer) to control the system.

  Therefore, the acoustic signal picked up by the microphone may have some voices from other people (passengers and passengers in the front seat), radio sound, and other noises, no matter how directional. This is included in the voice recognition device unconditionally.

  In addition, as a known technique related to speech recognition, a technique related to a speech analysis apparatus that determines whether or not an input acoustic signal is a human voice is described in [Patent Document 1] below.

  The first speech analyzer described in [Patent Document 1] detects the minimum value of the normalized average amplitude difference function and the latest maximum value with a detector, and detects the true value from the detected minimum value. The minimum value is obtained by an interpolator and input to a pitch detector, and a guide pitch that gives an expected area of the pitch detected by the pitch detector is obtained by a guide pitch calculator and input to the pitch detector. Is configured to introduce a weight variable to the pitch, and the error rate of pitch extraction can be significantly reduced, and deterioration of sound quality can be prevented as much as possible.

  In addition, the second speech analysis apparatus includes a speech signal converted by the analog-digital converter, a parameter extracted by the spectrum envelope parameter extractor, and a normalized average amplitude corresponding to the pitch detected by the pitch detector. It is configured to input each difference value between the true minimum value of the difference function and the latest maximum value to the voiced / unvoiced sound discriminator, which can significantly reduce the discrimination error rate of voiced / unvoiced sound. is there.

  Next, the following [Patent Document 2] and [Non-Patent Document 1] include a spatiotemporal differential method (a spatiotemporal gradient method) and a differential as a sound source localization auditory sensor system that determines where sound is heard. A technique related to a three-dimensional sound source localization sensor system using a sound source localization technique based on a sound source localization algorithm using a sum of products is described. The principle of sound source localization is as follows.

  That is, a sound wave propagating as a spherical wave at a constant velocity has a spherical surface area that expands in proportion to the square of the distance, so that the amplitude of the sound wave attenuates in inverse proportion to the distance. Therefore, when the amplitude of the sound wave is observed at four points of four microphones installed on adjacent squares, the wavefront of the sound wave reaches a little earlier for a microphone close to the sound source and a little later for a far microphone. Since the propagation speed of the sound wave and the distance between the microphones are known, the wavefront normal, that is, the direction of the sound source can be determined from the time difference between the acoustic signals captured by the four microphones.

  Further, because of the property that the amplitude of the sound wave attenuates in inverse proportion to the distance, the distance to the sound source can also be easily obtained from the amplitude ratio between the microphones close to and far from the sound source and the distance difference between the microphones.

  The sound source localization sensor system based on the above principle is a system using a spatio-temporal differential method in which the sound source is localized by controlling the azimuth angle of the sound collector, and an acoustic signal output from the sound source reaches the sound collector. A time difference calculating means for calculating a time difference until a predetermined time, a self-evaluation amount calculating means for calculating a self-evaluation amount serving as a reference for determining the effectiveness of the calculated time difference, and the calculated self-evaluation A determination unit that determines the effectiveness of the calculated time difference based on a quantity; an azimuth angle control unit that changes the azimuth when it is determined that the effectiveness is low; and a determination that the effectiveness is high And a sound source localization unit that localizes the sound source based on the time difference.

Japanese Patent Publication No. 13-3195700

Japanese Patent Publication No. 9-264417 Shigeru Ando, Hiroyuki Shinoda, Katsuya Ogawa, Kunomi Mitsuyama, Proceedings of the Society of Instrument and Control Engineers, Vol. 29, No. 5, pp. 520-528 (1993) “3D sound source localization sensor system based on spatiotemporal gradient method” .

  As described above, in a car navigation system or a human authentication system having a conventional voice recognition function, other than the voice of the person to be targeted, other people's voice in the vicinity or noise that is not the voice of the person is unavoidable. It was input to the voice recognition device, and it caused a number of misrecognitions by the voice recognition device, which in turn caused a malfunction of the system.

  The present invention provides a speech recognition apparatus for a system in advance in view of the problem of speech other than a person to be input to a speech recognition apparatus (speech recognition circuit) in various systems having a speech recognition function and a system malfunction caused thereby. It is an object of the present invention to provide an acoustic signal input / output device capable of reducing malfunctions of a system having the voice recognition function as much as possible by selecting and selecting conditions for acoustic signals picked up by an input microphone.

The present invention
(1) An acoustic signal input means 1 formed by combining a plurality of microphones 1a, 1b, 1c, 1d with a predetermined interval d, and each of the microphones 1a, 1b, 1c, 1d obtained by the acoustic signal input means 1 An A / D converter 2 that converts analog acoustic signals Sa, Sb, Sc, and Sd into digital acoustic signals Da, Db, Dc, and Dd, and the microphones 1a, 1b, and 1c converted by the A / D converter 2 A sound source localization means 3 for analyzing the 1d digital acoustic signals Da, Db, Dc, Dd and detecting the direction of the sound source; Sound source angle determining means 11 for determining whether the sound source is from the direction of the sound, and the microphone converted by the A / D converter 2 whether the sound source A is a human voice. sound detecting means 12 for analyzing and determining the digital sound signal Dd of d, and the sound source A obtained by the sound signal input means 1 is in a direction within the preset angle range δ and includes human speech The above-mentioned problem is solved by providing an acoustic signal input / output device 30 including a gate circuit 20 that outputs the digital acoustic signal Dd only when it is detected.

  (2) Also, an acoustic signal input means 1 formed by combining a plurality of microphones 1a, 1b, 1c, 1d with a predetermined interval d, and the microphones 1a, 1b, 1c, 1d obtained by the acoustic signal input means. A / D converter 2 that converts analog acoustic signals Sa, Sb, Sc, and Sd into digital acoustic signals, and the digital acoustic signals of the microphones 1a, 1b, 1c, and 1d converted by the A / D converter 2 Sound source localization means 3 that detects the direction of the sound source by analyzing Da, Db, Dc, Dd, and whether the sound source localized by the sound source localization means 3 is a sound source from a direction within a preset angle range δ. And the sound source angle determination means 11 for determining whether the sound source A is a human voice or not, and the digital acoustic signal of the microphone 1d converted by the A / D converter 2. Only when the sound source 12 obtained by analyzing and determining Dd and the sound signal input means 1 is in the preset angle range δ and includes human speech, the analog The above problem is solved by providing an acoustic signal input / output device 40 including a gate circuit 21 that outputs an acoustic signal Sd.

  This acoustic signal input / output device is applicable only to the sound of the sound source within the range of the set angle and when it is a human voice when the input set angle is set to a predetermined angle. Therefore, only when there is a sound source of human voice in a limited range as described above, the sound source is not transmitted to the voice recognition device built in a system such as a car navigation system or a human authentication system. Since the sound signal is input and the voice recognition is activated, the erroneous recognition of the voice recognition device is greatly reduced.

  The acoustic signal input / output device according to claim 1 of the present invention can reduce misrecognition of a digital speech input speech recognition device in the target system, and thereby reduce malfunction of the target system.

  The acoustic signal input / output device according to claim 2 of the present invention can reduce erroneous recognition of the analog speech input speech recognition device in the target system, and thus can reduce malfunction of the target system.

  An embodiment of an acoustic signal input / output device according to the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing a configuration example of a first acoustic signal input / output device according to the present invention. FIG. 2 is a block diagram showing a configuration example of a second acoustic signal input / output device according to the present invention. FIG. 3 is a perspective view showing the configuration of an acoustic signal input means formed by combining a plurality of microphones according to the present invention at a predetermined interval. FIG. 4 is a diagram for explaining a system in which a digital output acoustic signal input / output device of the present invention and a car navigation system incorporating a digital input speech recognition device are combined. FIG. 5 is a diagram for explaining a system in which an analog output acoustic signal input / output device of the present invention and a human authentication system incorporating an analog input speech recognition device are combined.

  In FIG. 1, an acoustic signal input / output device 30 is formed by combining four microphones 1a, 1b, 1c, and 1d with a predetermined interval d as shown in FIG. The sound signal input means 1 and the analog sound signals Sa, Sb, Sc, of the microphones 1a, 1b, 1c, 1d obtained by the sound signal input means 1 are arranged in parallel at the positions. An A / D converter 2 that converts Sd into digital acoustic signals Da, Db, Dc, and Dd (for example, 12 bits), and digital signals of the microphones 1a, 1b, 1c, and 1d converted by the A / D converter 2 The sound source localization means 3 for detecting the direction of the sound source A by analyzing the acoustic signals Da, Db, Dc, Dd, and the sound source A localized by the sound source localization means 3 are preset. The sound source angle determination means 11 for determining whether the sound source is from a direction within the degree range δ, and the digital sound of the microphone 1d converted by the A / D converter 2 whether the sound source A is a human voice. Only when the sound detection unit 12 that analyzes and determines the signal Dd and the sound source A obtained by the acoustic signal input unit 1 is in the predetermined angle range δ and includes human speech. And a gate circuit 20 that outputs a digital acoustic signal Dd.

  Hereinafter, in detail, first, the digital acoustic signal input to the sound detection means 12 is not limited to the digital acoustic signal Dd of the microphone 1d, but the digital acoustic signals Da, Db of the other microphones 1a, 1b, 1c. , Dc, or a plurality of Dc. Further, the digital acoustic signal output to the gate circuit 20 is not limited to the digital acoustic signal Dd in FIG. 1, but may be any one of the digital acoustic signals Da, Db, and Dc of the other microphones 1a, 1b, and 1c. There may be more than one. However, any one is sufficient in a general system.

  In the acoustic signal input / output device 30, the gate circuit 20 is configured by an AND logic circuit that inputs the output signal E1 of the sound source angle determination means 11, the output signal E2 of the sound detection means 12, and the digital acoustic signal Dd.

  The microphones 1a, 1b, 1c, and 1d are arranged in parallel so that, for example, the adjacent distance d is positioned at each corner of a square having a side length d = 30 mm. Further, in consideration of correspondence with human hearing, a square diagonal line having a side length d may be spaced apart by about d = 106 mm so that the length of a diagonal line of 150 mm is approximately equal to the distance between human ears. Good.

  As the sound source localization means 3, the auditory sensor system described in [Patent Document 2] and [Non-Patent Document 1] of publicly known technology can be used.

  That is, in FIG. 1, A / D converted digital acoustic signals Da, Db, Dc, Dd are respectively stored in the sampling data buffer 4,..., And the sum (synthetic sound field f) of these data is stored. The synthesized sound field f is input to the low-pass filter 6 and input to the time differentiation calculator 7 to determine the time gradient ft.

  Further, the spatial gradients fx and fy in the X direction and the Y direction are obtained by the X direction spatial gradient calculator 5b and the Y direction spatial gradient calculator 5c, respectively, and the output fx of the X direction spatial gradient calculator 5b is sent to the low-pass filter 8. The output fy of the Y-direction spatial gradient calculator 5 c is input to the low-pass filter 9.

  Next, the sound source angle calculator 10 uses the output signals f, ft, fx, and fy of the low-pass filter 6, the time differentiation calculator 7, the low-pass filter 8, and the low-pass filter 9, respectively, An angle θ from the polar axis of the sound source displayed in three-dimensional spherical coordinates (r, θ, φ) having the sound center position as the origin and the microphone direction as the polar axis is calculated.

  The sound source angle determination means 11 is a comparator for determining whether or not the calculated direction of the sound source is within a preset angle range δ and outputting a logic signal E1. The cone angle θ1 from the predetermined reference line (polar axis) of the sound source obtained by the sound source localization means 3 (the three-dimensional cone angle (coordinate latitude) θ and the polar angle of the spherical coordinates (r, θ, φ)) (Longitude) The direction is displayed with φ.) A logic circuit that can determine whether or not is within the range of the predetermined angle δ by subtraction to obtain the difference (δ−θ1) between the two can be applied. A “1” signal is output, and if it is out of the angle range, a logic “0” signal is output.

  The voice detection means 12 preferably uses a voice analysis device described in the above-mentioned [Patent Document 1].

  That is, in FIG. 1, the A / D converted digital acoustic signal Dd is input to the linear prediction analyzer 13, and the partial autocorrelation function K [i] and the residual EN are calculated. The prediction residual ed [i] is calculated by the PARCOR analysis filter 14, and the average amplitude difference function ra [k] is calculated by the average amplitude difference function calculator 16 via the low-pass filter 15. Next, the minimum point detector 17 detects the minimum point ramin of the average amplitude difference function, and the pitch extractor 18 extracts the pitch pitch. The speech section determiner 19 determines whether or not it is a human speech from the data of the partial autocorrelation function K [i], the residual EN, the pitch pitch, and the minimum point ramin of the average amplitude difference function, and outputs a logical signal E2. Output. For example, “1” is output for a human voice, and “0” is output for other sounds.

  Thus, from the digital acoustic signal Dd and the logic signals E1 and E2 input to the AND logic circuit of the gate circuit 20, the output of the gate circuit 20 is a sound from a sound source within a predetermined angle range δ, and The digital sound signal Dd is output only in the case of the sound of the sound, and as shown in FIG. 4, the sound signal input / output device is connected to the sound recognition device 31 of the car navigation system 32 having the sound recognition function. What is input via 30 is the voice of the person who is the sound source A within the predetermined angle range δ. In addition, it is desirable to remove sounds other than the human voice band through a filter (not shown) as much as possible.

  Next, the acoustic signal input / output device 40 shown in FIG. 2 is similar to the acoustic signal input / output device 30 described above, and an acoustic signal input is formed by combining a plurality of microphones 1a, 1b, 1c, 1d with a predetermined distance d apart. Means 1, and an A / D converter 2 for converting the analog sound signals Sa, Sb, Sc, Sd of the microphones 1a, 1b, 1c, 1d obtained by the sound signal input means 1 into digital sound signals; Sound source localization means 3 that analyzes the digital acoustic signals Da, Db, Dc, Dd of the microphones 1a, 1b, 1c, 1d converted by the A / D converter 2 to detect the direction of the sound source, and the sound source localization A sound source angle determining means 11 for determining whether the sound source A localized by the means 3 is a sound source from a preset angle range δ; and whether the sound source A is a human voice. Voice detecting means 12 for analyzing and determining the digital acoustic signal Dd of the microphone 1d converted by the A / D converter 2, and the sound source A obtained by the acoustic signal input means 1 And a gate circuit 21 that outputs the analog acoustic signal Sd only when the direction is within the set angle range δ and human speech is included.

  In this acoustic signal input / output device 40, the gate circuit 21 takes the logical product of the output signal E1 of the sound source angle determination means 11 and the output signal E2 of the sound detection means 12 by the AND logic circuit 21b, and the microphone is output by the AND logic output E3. The output gate switch 21a (configured by a switching transistor or the like) for the 1d analog acoustic signal Sd is opened and closed. The difference from the acoustic signal input / output device 30 is a gate circuit 20 that outputs a digital acoustic signal Dd. Instead of this, a gate circuit 21 for outputting an analog acoustic signal Sd is provided.

  The digital sound signal input to the sound detection means 12 is not limited to the digital sound signal Dd of the microphone 1d in FIG. 2, but any one of the digital sound signals Da, Db, Dc of the other microphones 1a, 1b, 1c. One or more. The analog acoustic signal output to the gate circuit 21 is not limited to the analog acoustic signal Sd in FIG. 2, and may be any of the analog acoustic signals Sa, Sb, Sc of the other microphones 1 a, 1 b, 1 c, But you can. However, any one is sufficient in a general system.

  The digital acoustic signal Dd or the analog acoustic signal Sd, which is the output of the acoustic signal input / output device 30 or 40, is a digital or analog input speech recognition as in the car navigation system 32 in FIG. 4 or the human authentication system 42 in FIG. Input to speech recognition devices (speech recognition circuits) 31 and 41 of various systems having functions.

  Thus, sounds from the sound source B and the sound source C outside the angle range δ are output from the sound signal input / output devices 30 and 40 to the speech recognition devices 31 and 41 even if they are human voices. Therefore, the erroneous recognition of the speech recognition devices 31 and 41 and the malfunction of various systems such as the car navigation system 32 and the human authentication system 42 are greatly reduced.

  In FIGS. 4 and 5, the angles δ and δ ′ of the setting range are shown in two dimensions, but the actual input setting angles δ and δ ′ are the cone angles θ from the polar axis Z of the three-dimensional spherical coordinates. Is set as an angle display.

  In the acoustic signal input / output device 30 or 40, the set angle range δ can be set not only on the front side of the microphone 1a as shown in FIG. 4, but also in an arbitrary angle range δ ′ as shown in FIG. is there.

  For example, the sound signal input / output device 30 is connected to the sound signal input interface of the voice recognition device 31 in the car navigation system 32 of FIG. 4, and the sound source A is the voice of the driver, the sound source B is the voice of the passenger in the passenger seat, and the sound source If C is the output sound of a radio speaker and the input setting angle is set as shown in the figure, this device is activated only by the sound of the sound source A emitted by the driver, and only the sound emitted by the driver of the sound source A Is input to the voice recognition device 31 of the car navigation system 32. Therefore, the voice of the passenger seat of the sound source B and the sound of the sound source C of the radio are out of the range of the angle δ, and thus are not input as an acoustic signal to the speech recognition device 31, and malfunction of the speech recognition device 31 is prevented. It is. If the voice of the passenger in the passenger seat (sound source B) is used as the target sound source, it can be easily changed by simply changing the set angle range δ of the sound source localization means 3.

  The present invention is a three-dimensional sound source localization sensor that was originally developed as an auditory sensor system for a robot in order to reduce misrecognition of a voice recognition device in various systems having a voice recognition function such as a conventional car navigation system and a human authentication system. Focusing on both the system effects and the effects of the speech analysis device for precise speech analysis, the idea is to organically combine them into a means for improving accuracy for new speech recognition. Needless to say, it is very useful to add to the input side of an existing speech recognition device (speech recognition circuit) and to improve speech recognition accuracy easily and dramatically.

It is a block diagram which shows the structural example of the 1st acoustic signal input / output device which concerns on this invention. It is a block diagram which shows the structural example of the 2nd acoustic signal input / output device which concerns on this invention. It is a perspective view which shows the structure of the acoustic signal input means formed by separating a plurality of microphones according to the present invention at predetermined intervals. It is a figure for demonstrating the system which combined the acoustic signal input / output apparatus of the digital output of this invention, and the car navigation system which incorporates the speech recognition apparatus of a digital input. It is a figure for demonstrating the system which combined the acoustic signal input / output device of the analog output of this invention, and the human authentication system incorporating the analog speech recognition apparatus.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Acoustic signal input means 1a, 1b, 1c, 1d Microphone 2 A / D converter 3 Sound source localization means 4 Sampling data buffer 5a Synthesis | combination calculator 5b X direction space gradient calculator 5c Y direction space gradient calculator 6, 8, DESCRIPTION OF SYMBOLS 9 Low-pass filter 7 Time differentiation calculator 10 Sound source angle calculator 11 Sound source angle determination means 12 Speech detection means 13 Linear prediction analyzer 14 PARCOR analysis filter 15 Low-pass filter 16 Average amplitude difference function calculator 17 Minimum point detector 18 Pitch extractor 19 Voice section decision device 20, 21 Gate circuit 21a Output gate switch 21b AND logic circuit 30, 40 Acoustic signal input / output devices A, B, C Sound source Da, Db, Dc, Dd Digital acoustic signal Sa, Sb, Sc, Sd Analog Acoustic signal E1 Logic signal output from sound source angle determination means E2 Logic signal output from voice detection means E3 AND logic output from AND logic circuit δ, δ ′ Pre-set angle range d Predetermined interval θ1 Cone angle f Synthetic sound field ft Time gradient fx Spatial gradient in X direction fy Y direction Spatial gradient

Claims (2)

An acoustic signal input means formed by combining a plurality of microphones with a predetermined separation, an A / D converter for converting an analog acoustic signal of each microphone obtained by the acoustic signal input means into a digital acoustic signal, and the A / D Sound source localization means for detecting the direction of the sound source by analyzing the digital sound signal of each microphone converted by the converter, and the sound source localized by the sound source localization means from a direction within a preset angle range Sound source angle determining means for determining whether the sound source is sound source; sound detecting means for determining whether the sound source is a human voice by analyzing the digital acoustic signal of the microphone converted by the A / D converter; The sound source obtained by the acoustic signal input means is in the direction within the preset angle range and includes the human voice only. Acoustic signal output device, characterized in that it comprises a gate circuit for outputting a barrel acoustic signal. An acoustic signal input means formed by combining a plurality of microphones with a predetermined separation, an A / D converter for converting an analog acoustic signal of each microphone obtained by the acoustic signal input means into a digital acoustic signal, and the A / D Sound source localization means for detecting the direction of the sound source by analyzing the digital sound signal of each microphone converted by the converter, and the sound source localized by the sound source localization means from a direction within a preset angle range Sound source angle determining means for determining whether the sound source is sound source; sound detecting means for determining whether the sound source is a human voice by analyzing the digital acoustic signal of the microphone converted by the A / D converter; Only when the sound source obtained by the acoustic signal input means is in a direction within the preset angle range and includes human speech. Acoustic signal output device, characterized in that it comprises a gate circuit which outputs the log acoustic signal.

Images