JP4587941B2 - Speech correction system and adaptive filter used therefor - Google Patents

Description

  The present invention relates to a sound correction system and an adaptive filter used therefor, and is particularly suitable for use in a system that suppresses specific sound from mixed sound in which a plurality of sounds are mixed.

  Most recent vehicles are equipped with various electronic devices such as an audio device, an air conditioner, and a navigation device. Recently, in order to avoid one-handed operation or the like when operating these electronic devices, a system is also provided that allows the electronic devices to be operated by voice recognition. By using this voice recognition technology, the driver can operate various electronic devices without removing his / her hands from the steering wheel (without manually operating an operation unit such as a remote controller or an operation panel).

  A voice recognition system usually inputs a specific word, idiom or simple command sentence uttered by a user from a microphone and recognizes it as an utterance command. And the process according to the recognized utterance command is performed. Here, when the user utters a command in the presence of the output sound of the audio device, the audio sound is also input to the microphone in addition to the speaker sound, and the audio sound becomes noise for speech recognition. The rate will drop.

  Accordingly, an ASC (Audio Sound Cancellation) system is provided that suppresses only audio sound from mixed sound (speaker sound and audio sound) input from a microphone. In general, an adaptive filter is often used to realize an ASC system. An N-LMS (Normalized-LMS) algorithm is often used as an adaptive filter algorithm.

  FIG. 5 is a diagram showing a schematic configuration of a conventional ASC system using an adaptive filter. In FIG. 5, 101 is a speaker for outputting audio sound, 102 is a microphone for inputting sound, 103 is an adaptive filter, and 104 is a subtractor. The microphone 102 is for speaker voice input provided for voice recognition processing. However, when the audio source is reproduced by the audio device, not only the speaker voice but also the audio output from the speaker 101 is used. Audio is also input from the microphone 102.

  When the algorithm is the N-LMS algorithm, the adaptive filter 103 indicates the audio sound x (n) that is input as a reference signal to be controlled (Equation 1-1) to (Equation 1-3) (hereinafter, When there is no particular distinction, these are corrected together (denoted as (Equation 1)). (Equation 1) represents an arithmetic expression when the audio sound x (n) is one channel.

  In this (Formula 1), w (n), α (n), e (n), and x (n) are all matrices, and “T” on the right shoulder indicates a transposed matrix. “−1” in FIG. 4 indicates an inverse matrix. L represents the tap length of the adaptive filter, and μ represents the step size parameter. The step size parameter indicates the magnitude of correction of the filter coefficient, and is a parameter for controlling the convergence of the adaptive process. “Α (n) × e (n)” multiplied by the step size parameter μ is called a correction term. In this specification, α (n) included in the correction term is referred to as “correction data”.

  The subtractor 104 subtracts the corrected audio sound y (n) output from the adaptive filter 103 from the mixed sound d (n) of the audio sound and the speaker sound input from the microphone 102, thereby generating an error e ( n) to calculate only the speaker's voice. The speaker voice extracted by the subtracter 104 is supplied to a voice recognition engine (not shown). Thereby, processing corresponding to the utterance command is executed.

  The adaptive filter 103 includes a reference signal calculation unit 103a, an update filter coefficient calculation unit 103b, and an audio correction filter 103c. The reference signal calculation unit 103a calculates the audio sound x (n) input as the reference signal as shown in (Equation 1-2), and calculates correction data α (n). The update filter coefficient calculation unit 103b performs an operation of (Equation 1-1) according to the N-LMS algorithm, operates so as to minimize the power of the error e (n) output from the subtractor 104, and operates as a voice correction filter. The filter coefficient w (n) of 103c is identified.

  The sound correction filter 103c performs a filter operation on the audio sound x (n) to be controlled using the filter coefficient w (n) determined by the update filter coefficient calculation unit 103b. Specifically, the corrected audio sound y (n) is obtained by giving the transfer function identical to the transfer function of the audio sound transmitted from the speaker 101 to the microphone 102 to the audio sound x (n) to be controlled. Get. The audio sound y (n) after filter control output from the sound correction filter 103c is supplied to the subtractor 104, and the error e (n) calculated here is fed back to the update filter coefficient calculation unit 103b.

  The ASC system as described above is a technique for enabling voice recognition without muting the audio voice, and is indispensable for realizing a comfortable voice HI (Human Interface). However, the nature of audio speech varies from source to source and is non-stationary. When handling a signal having such a property, control of the correction data α (n) calculated by the reference signal calculation unit 103a and the N-LMS algorithm of the update filter coefficient calculation unit 103b using the correction data α (n) is performed by the ASC. This is effective in maintaining good system performance.

  However, in order to keep the performance of the ASC system good, it is necessary to obtain the filter coefficient finely (with a large number of taps) at a high sampling frequency, and the amount of calculation becomes large. In particular, when the audio sound is multi-channeled, the amount of calculation becomes very large.

A technique has been proposed in which required coefficient data is calculated in advance and accumulated in a filter coefficient database, and the coefficient data is read from the database and set as a tap coefficient of a control filter (for example, a patent). Reference 1). If the technique described in Patent Document 1 is applied to an ASC system, it is possible to significantly reduce the amount of calculation required for calculating filter coefficients.
Japanese Patent Laid-Open No. 10-307592

  However, the filter coefficient described in Patent Document 1 is for creating a sound field characteristic corresponding to the vehicle type, and is fixed regardless of the contents of the input audio sound. On the other hand, in the case of the ASC system, as described above, the property of the audio sound x (n) to be controlled differs depending on each source, and is temporally unsteady. Also, the environment in which the audio sound x (n) is reproduced changes from time to time. For this reason, the filter coefficient must be changed as needed in accordance with the changing nature or environment of the audio sound x (n). Therefore, the technique described in Patent Document 1 cannot be applied to an ASC system as it is.

  The present invention has been made to solve such a problem, and an object of the present invention is to reduce the amount of calculation required for calculating the filter coefficient of an adaptive filter used in an ASC system.

  In order to solve the above-described problems, in the present invention, correction data is calculated in advance based on the audio sound to be controlled and stored as a database, and the audio sound that is actually used for control is reproduced. Sometimes, the correction data corresponding to the audio sound is read from the database and output in synchronization with the time code of the audio sound to be controlled. Then, using the correction data output in this way, a filter coefficient is obtained by an adaptive algorithm, and audio sound filtering processing is performed using the obtained filter coefficient.

  According to the present invention configured as described above, at least correction data that has been calculated in advance is stored as a database and can be read and used. It is no longer necessary to obtain correction data in real time when is being played. In addition, an adaptive algorithm is applied to the read correction data, and filter coefficients corresponding to the properties of the audio sound to be controlled and the circumstances at that time are obtained in real time. Therefore, even when the present invention is applied to the ASC system, it is possible to calculate an appropriate filter coefficient of the adaptive filter with a small amount of calculation.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a diagram showing a configuration example of an ASC system that implements an adaptive filter and a sound correction system of the present invention. As shown in FIG. 1, the ASC system of the present embodiment includes an in-vehicle user terminal 100 and a data bank device (server device) 200 installed outside the vehicle, and the user terminal 100 and the data bank device 200 include It is configured to be mutually connectable via a communication network.

  In the configuration of the user terminal 100, reference numeral 1 denotes a sound source, which is constituted by a recording medium such as a CD (Compact Disc), a DVD (Digital Versatile Disk), a hard disk, a semiconductor memory, or an MD (Mini Disc) in which audio data is stored. ing. The audio data of the sound source 1 is stored for each time code indicating the elapsed time of the audio data, as shown in FIG. Here, the time code may be a code representing an elapsed time from the head of the track, or may be a code representing a continuous elapsed time from the head of the sound source 1.

  Reference numeral 2 denotes an audio source playback unit, which includes, for example, a CD player, a DVD player, an MP3 player, and an MD player. The audio source playback unit 2 plays back and outputs the audio sound (reference signal) x (n) to be controlled from the sound source 1. The audio source reproduction unit 2 reproduces and outputs a time code representing the elapsed time of the audio sound x (n) from the sound source 1.

  Reference numeral 3 denotes a speaker, which outputs the audio sound x (n) reproduced by the audio source reproducing unit 2. Reference numeral 4 denotes a microphone for inputting voice. The microphone 4 is originally for speaker voice input provided for voice recognition processing. However, when the sound source 1 is played back by the audio source playback unit 2, not only the speaker voice but also the microphone 4 is used. The audio sound output from the speaker 3 is also input from the microphone 4.

  Reference numeral 5 denotes an update filter coefficient calculation unit, 6 denotes an audio correction filter, and 7 denotes a subtracter. The update filter coefficient calculation unit 5 and the audio correction filter 6 constitute a part of the adaptive filter according to the present embodiment. When the algorithm is an N-LMS algorithm, the adaptive filter according to the present embodiment corrects the audio sound x (n) to be controlled according to the algorithm similar to (Equation 1) described above.

  The update filter coefficient calculation unit 5 performs the calculation of (Equation 1-1) according to, for example, the N-LMS algorithm using the correction data α (n) acquired by the procedure described later, and feeds back from the subtractor 7. The filter coefficient w (n) of the sound correction filter 6 is identified by operating so that the power of the error e (n) to be performed is minimized. The update filter coefficient calculation unit 5 corresponds to the filter coefficient calculation unit of the present invention.

  The audio correction filter 6 corresponds to the filter processing unit of the present invention, and uses the filter coefficient w (n) obtained by the update filter coefficient calculation unit 5 to control the audio sound x (n) to be controlled. A filtering process is performed on the image. Specifically, the corrected audio sound y (n) is obtained by giving the transfer function identical to the transfer function of the audio sound transmitted from the speaker 3 to the microphone 4 to the audio sound x (n) to be controlled. Get.

  The audio sound y (n) after the filter control output from the sound correction filter 6 is supplied to the subtractor 7. The subtracter 7 subtracts the corrected audio sound y (n) output from the sound correction filter 6 from the mixed sound d (n) of the audio sound and the speaker sound input from the microphone 4 to generate an error e. (N) is calculated, and only the speaker voice is extracted. The speaker voice extracted by the subtracter 7 is supplied to a voice recognition engine (not shown). The error e (n) is fed back to the update filter coefficient calculation unit 5.

  Reference numeral 8 denotes an audio track information extraction unit which extracts audio track information from the audio sound to be controlled recorded in the sound source 1. The audio track information extraction unit 8 corresponds to the identification information extraction unit of the present invention. The audio track information to be extracted includes information for identifying audio sound, and corresponds to audio sound identification information according to the present invention.

  A request transmission unit 9 transmits the audio track information extracted by the audio track information extraction unit 8 to the data bank apparatus 200 via the communication network, and the correction data corresponding to the audio sound indicated by the audio track information. Request acquisition of α (n). Reference numeral 10 denotes a correction data acquisition unit that acquires correction data α (n) returned as a response from the data bank device 200 according to the audio track information transmitted from the request transmission unit 9 to the data bank device 200. To do.

  Reference numeral 11 denotes a correction data output unit, which holds the correction data α (n) acquired by the correction data acquisition unit 10. Then, a time code representing the elapsed time of the audio sound x (n) to be controlled is acquired from the audio source reproduction unit 2, and the correction data α (n) is synchronized with the elapsed time represented by the time code. It outputs to the update filter coefficient calculation part 5 sequentially. That is, as will be described later, a time code is associated with the correction data α (n) acquired by the correction data acquisition unit 10. The correction data output unit 11 sequentially outputs the correction data α (n) corresponding to the time code to the update filter coefficient calculation unit 5 every time the time code is acquired from the audio source reproduction unit 2. The update filter coefficient calculation unit 5 uses the correction data α (n) output from the correction data output unit 11 to calculate the filter coefficient w (n) of the sound correction filter 6 according to the above (Equation 1-1). Ask.

  On the other hand, in the configuration of the data bank device 200, reference numeral 21 denotes a correction data DB (database), which is correction data α (n) necessary for obtaining the filter coefficient of the sound correction filter 6 and is to be controlled. The correction data α (n) calculated in advance based on the audio sound x (n) is stored together with the identification information (for example, audio track information) of the audio sound. The correction data DB 21 corresponds to the correction data storage unit of the present invention.

  As can be seen from the above (Equation 1-2), the arithmetic expression for obtaining the correction data α (n) does not include the term of the step size parameter μ and the error e (n). The correction data α (n) can be obtained only from the audio sound x (n). Therefore, the correction data α (n) is data that can be specified only from the audio sound x (n), and the value is updated according to the environment when the audio sound x (n) is actually reproduced. It is not necessary.

  Therefore, in the present embodiment, the sound source 1 is obtained in advance, the correction data α (n) is calculated from the audio sound x (n), and this is stored in the correction data DB 21. This is preferably performed for the audio sound x (n) of various sound sources 1. At that time, in order to be able to identify later from which audio sound x (n) which correction data α (n) is obtained, the correction data α (n) is converted to audio sound x (n ) And stored in the correction data DB 21.

  FIG. 3 is a diagram illustrating an example of the correction data α (n) obtained from one audio sound x (n). As shown in FIG. 3, the correction data α (n) is stored in the correction data DB 21 in association with the time code of the audio sound x (n). That is, the calculation shown in (Equation 1-2) is performed on the audio data for each time code as shown in FIG. 2, and the correction data α (n) for each time code obtained as a result is used as the time code. The data is linked and stored in the correction data DB 21.

  Reference numeral 22 denotes a correction data reading unit. When a request for acquisition of correction data α (n) is sent from the request transmission unit 9, the audio indicated by the audio track information included in the acquisition request is sent in response thereto. The correction data α (n) corresponding to the voice is read from the correction data DB 21. Then, the read correction data α (n) is transmitted to the user terminal 100 via the communication network.

  As described above in detail, according to the present embodiment, the correction data α (n) is calculated in advance from the audio sound to be controlled and stored in a database, and when the audio sound is actually reproduced, Since the adaptive filter is updated by using the correction data α (n), the calculation for at least the correction data α (n) is performed when the ASC system performs the process of suppressing only the audio sound. This can be omitted, and the amount of calculation can be reduced. In addition, since the filter coefficients are appropriately updated according to the adaptive algorithm instead of using the fixed filter coefficients, the filter coefficients corresponding to the properties of the audio sound to be controlled and the circumstances at that time can be obtained in real time. The system performance can be kept good.

  In the above-described embodiment, the example in which the correction data α (n) is acquired by communicating with the external data bank apparatus 200 every time audio data is reproduced from the sound source 1 has been described. It is not limited. For example, the user terminal 100 has a local data storage unit (for example, a hard disk), and the correction data α (n) first acquired from the data bank device 200 is stored in the local data storage unit. When the same audio data is reproduced for the second time or later, the correction data α (n) may be read from the local data storage unit and used. In this case, communication with the external data bank device 200 is performed only when the correction data α (n) is not present in the local data storage unit.

  Instead of preparing the data bank device 200 outside the user terminal 100, the user terminal 100 may be provided with the correction data DB 21 and the correction data reading unit 22 itself. In this case, in order to record the correction data α (n) in the correction data DB 21 for the first time, the user terminal 100 needs to further include the reference signal calculation unit 103a of FIG. The correction data α (n) must be calculated by the unit 103a only when the audio data is reproduced for the first time. That is, the correction data α (n) calculated first for the audio sound x (n) reproduced by the audio source reproduction unit 2 is stored in the correction data DB 21, so that the same audio sound x (n) is stored for the second time. In the subsequent reproduction, the correction data α (n) can be read from the correction data DB 21 and used.

  In the above embodiment, the case where the adaptive algorithm of the update filter coefficient calculation unit 5 is the N-LMS algorithm has been described as an example, but the present invention is not limited to this. For example, a projection algorithm or an RLS (Recursive Least Square) algorithm may be used. In the case of the projection algorithm, the audio sound x (n) input as the reference signal to be controlled is corrected according to the following (Equation 2).

  In the above (Formula 2), β (n) is correction data, which is also obtained from only the audio sound x (n). Therefore, the correction data β (n) is obtained in advance from the audio sound x (n) and stored in the correction data DB 21 as a database, and when the audio sound is actually reproduced, the correction data β (n in the database is stored. ) To update the adaptive filter.

  Similarly, in the case of the RLS algorithm, the audio sound x (n) input as the reference signal to be controlled is corrected according to the following (Equation 3). In (Equation 3), γ (n) is correction data, which is also obtained only from the audio sound x (n). Accordingly, the correction data γ (n) is obtained in advance from the audio sound x (n) and stored in the correction data DB 21 as a database. When the audio sound is actually reproduced, the correction data γ (n) in the database is stored. ) To update the adaptive filter.

  In the above embodiment, the example in which the audio track information is used as the identification information to acquire the correction data α (n) from the correction data DB 21 has been described. However, in addition to this, an ASC system is implemented. The vehicle type information (corresponding to the tap length identification information according to the present invention) of the existing vehicle may be used. The tap length L of the adaptive filter varies depending on the size of the vehicle on which the adaptive filter is mounted, and a value required for maintaining good performance of the ASC system. That is, it is preferable that the tap length L of the adaptive filter is increased as the vehicle is larger.

  When the ASC system is configured using the vehicle type information in addition to the audio track information, the ASC system is configured as shown in FIG. In FIG. 4, components having the same functions as those shown in FIG. 1 are denoted by the same reference numerals. The ASC system shown in FIG. 4 further includes a vehicle type information acquisition unit 12 (corresponding to the identification information acquisition unit of the present invention) that acquires the vehicle type information of the vehicle on which the audio sound x (n) to be controlled is reproduced. Yes. The vehicle type information referred to here may be information that at least indicates the size of the vehicle (it is only necessary to know the rank of the size, not the specific size). As this vehicle type information, for example, information stored in an internal memory of a navigation device (not shown) is used.

  The vehicle type information acquisition unit 12 acquires vehicle type information from a navigation device (not shown), for example, and outputs it to the request transmission unit 9. The request transmission unit 9 transmits the audio track information output from the audio track information extraction unit 8 and the vehicle type information output from the vehicle type information acquisition unit 12 to the data bank device 200 via the communication network, and the audio track information And the acquisition of correction data α (n) indicated by the vehicle type information is requested.

  In the correction data DB 21, correction data α (n) calculated for different tap lengths L based on the audio sound to be controlled is stored together with the identification information of the audio sound and the vehicle type information. That is, in the embodiment of FIG. 1 described above, one correction data α (n) is obtained from one audio sound x (n) and stored in the correction data DB 21. In the embodiment of FIG. A plurality of correction data α (n) is obtained from one audio sound x (n) for different tap lengths L, which correction data α (n) is obtained from which audio sound x (n), which The correction data α (n) is linked with the identification information of the audio sound x (n) and the vehicle type information so that the vehicle type (for which tap length L) can be identified later. Stored in the business data DB 21.

  Based on the audio track information and the vehicle type information sent from the request transmission unit 9 of the user terminal 100, the correction data reading unit 22 corrects the correction data α (n) corresponding to the audio track information and the vehicle type information. Read from the data storage unit 21. Then, the read correction data α (n) is transmitted to the user terminal 100. The correction data acquisition unit 10 acquires the correction data α (n) sent from the correction data reading unit 22 and outputs it to the correction data output unit 11.

  In addition, although vehicle type information is used here, it is not limited to this. For example, table information indicating the correspondence between the vehicle type and the tap length may be prepared, and the vehicle type information may be replaced with the tap length information by referring to this table information.

  In addition, each of the above-described embodiments is merely an example of actualization in carrying out the present invention, and the technical scope of the present invention should not be construed in a limited manner. In other words, the present invention can be implemented in various forms without departing from the spirit or main features thereof.

  The present invention is useful for, for example, an ASC system that suppresses only audio sound from mixed sound input from a microphone.

It is a figure which shows the structural example of the ASC system which implemented the adaptive filter and audio | voice correction | amendment system of this invention. It is a figure which shows the schematic structure image of the audio data of a sound source. It is a figure which shows the schematic structure image of the data for correction stored in the data DB for correction of this embodiment. It is a figure which shows the other structural example of the ASC system which implemented the adaptive filter and audio | voice correction system of this invention. It is a figure which shows the structural example of the conventional ASC system.

Explanation of symbols

DESCRIPTION OF SYMBOLS 5 Update filter coefficient calculation part 6 Audio | voice correction filter 7 Subtractor 8 Audio track information extraction part 9 Request transmission part 10 Correction data acquisition part 11 Correction data output part 12 Vehicle type information acquisition part 21 Correction data DB
22 data reading unit for correction 100 user terminal 200 data bank device

Claims (3)

A correction data storage unit that stores correction data necessary for obtaining the filter coefficient of the adaptive filter, the correction data calculated based on the audio sound to be controlled, together with the identification information of the audio sound;
An identification information extraction unit that extracts the identification information from the audio sound to be controlled;
A correction data acquisition unit that acquires correction data corresponding to the identification information extracted by the identification information extraction unit from the correction data storage unit;
Correction data for obtaining a time code representing the elapsed time of the audio sound to be controlled and outputting the correction data acquired by the correction data acquisition unit in synchronization with the elapsed time represented by the time code An output section;
A filter coefficient calculation unit for obtaining a filter coefficient of the adaptive filter using the correction data output from the correction data output unit;
An audio correction system comprising: a filter processing unit that performs a filtering process on the audio sound to be controlled using the filter coefficient obtained by the filter coefficient calculating unit. In the correction data storage unit, correction data calculated for different tap lengths based on the audio sound to be controlled is stored together with the identification information of the audio sound and the identification information of the tap length,
An identification information acquisition unit that acquires identification information of a tap length according to the vehicle type of the vehicle on which the audio sound to be controlled is reproduced;
The correction data acquisition unit is a correction data corresponding to the identification information based on the identification information of the audio sound extracted by the identification information extraction unit and the identification information of the tap length acquired by the identification information acquisition unit. The sound correction system according to claim 1, wherein the sound correction system is acquired from the correction data storage unit. The correction data calculated based on the audio sound to be controlled corresponds to the identification information extracted from the audio sound to be controlled from the correction data storage unit stored together with the identification information of the audio sound. A correction data acquisition unit for acquiring correction data;
Correction data for obtaining a time code representing the elapsed time of the audio sound to be controlled and outputting the correction data acquired by the correction data acquisition unit in synchronization with the elapsed time represented by the time code An output section;
A filter coefficient calculation unit for obtaining a filter coefficient using the correction data output from the correction data output unit;
An adaptive filter comprising: a filter processing unit that performs a filtering process on the audio sound to be controlled using the filter coefficient obtained by the filter coefficient calculation unit.

Images