JP2000231399A - Noise reducing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To remove a noise (music) from a car audio device mixing into a voice inputting microphone by estimating a noise (music) signal mixing into the voice inputting microphone based on an audio signal outputted from the car audio device and subtracting the noise (music) signal from a detection signal from the voice inputting microphone. SOLUTION: The audio signal from the car audio device 10 is band limited by a first filter means 30, and is converted to a digital signal by a first A/D conversion means 40 to be inputted to an adaptive filter part 51 and a coefficient updating part 52. The audio signal (d) converted into the digital signal is processed by the adaptive filter part 51 to be outputted to a difference means 53 as an estimated noise (music) signal (y). The difference means 53 outputs a difference between a microphone detection signal (x) outputted from a second A/D conversion means 80 and the estimated noise (music) signal (y) from the adaptive filter part 51 to the coefficient updating part 52 and a voice recognition part 90 as an error signal (e).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to, for example, a voice recognition device for inputting information by voice used in a noisy environment, particularly a vehicle-mounted voice recognition device such as a car navigation system having a voice recognition function, a voice recognition word processor, and the like. The present invention relates to a noise reduction device that reduces noise to a vehicle.

[0002]

2. Description of the Related Art With the high integration, high speed, and low price of LSIs and the rapid spread of computers, products using digital signal processing technology are being actively developed. Even in cars, including car audio systems, car phones,
Many devices for improving comfort in a vehicle, such as a car navigation system, have been commercialized and mounted. In particular, automobile telephones and car navigation systems are being developed for intelligent transportation systems (ITS: Intelligent Transport System).
rt System). However, it is not only troublesome to operate these devices manually in a running vehicle by the driver, but also involves a danger. Therefore, a hands-free function is required from the viewpoint of safety. As a hands-free function, a voice input operating system capable of operating a device by voice has attracted attention. Although a speech input operating system using speech recognition technology is useful for safety measures as well as convenience, there is a problem that the currently proposed speech recognition device cannot provide sufficient recognizability in a noise environment and is not practical. is there. For this reason, in order to improve speech recognition accuracy in an in-vehicle environment, for example, a document (Acoustic Society of Japan, 3-5-9, “Configuration of a noise suppression system for a vehicle-mounted speech recognition device”)
Thus, a noise suppression system for a vehicle-mounted speech recognition device has been proposed.

[0003]

However, in the method proposed in the above-mentioned document, when the car audio device is operating, the voice input microphone of the voice recognition device not only outputs the voice but also the voice from the car audio device. Is also detected at the same time, there is no effect at all, and there is a fatal problem that the speech recognition accuracy is significantly reduced. For this reason, when performing voice recognition, the audio device must be stopped when the audio device is not operating, or each time the audio device is operating, and warping may occur in consideration of the actual driving environment. The operability becomes worse,
There was a fatal problem that it was not practical.

[0004] The present invention solves the problems of the above-mentioned conventional method, and provides a noise reduction device for removing noise from a car audio device mixed into a voice input microphone and putting a vehicle-mounted voice recognition device into practical use. The purpose is to do.

[0005]

For this purpose, in the noise reduction apparatus of the present invention, a first filter means for limiting a band of an audio signal output to a speaker from a car audio apparatus provided in an automobile, A first A / D converter for converting an analog output signal output from the first filter into a digital signal;
An adaptive filter means for generating an estimated music signal from a digital audio signal output from the D conversion means, and an adaptive filter means provided in the vicinity of an operator operating a voice input device in an automobile to detect a voice generated by the operator. , A second filter means for limiting a band of a detection signal from the sound input microphone, and a second filter for converting an analog output signal output from the second filter means into a digital signal. A / D conversion means, difference means for obtaining a difference between an estimated music signal output from the adaptive filter means, and a digitized microphone detection signal output from the second A / D conversion means, From the error signal obtained by the means and the digitally converted audio signal output from the first A / D converter, the error signal is reduced. Coefficient updating means for updating the coefficient of the adaptive filter, and estimating a noise (music) signal mixed into the audio input microphone based on an audio signal output from the car audio device, A noise (music) signal is subtracted from a detection signal from a voice input microphone.

[0006]

Embodiments of the present invention will be described below in detail with reference to the drawings. In the following description, an example of application to an in-vehicle voice recognition device (for example, a car navigation system with a voice recognition function) will be described, and description of other examples will be omitted.

Description of (First Embodiment) Description of (Configuration) FIG. 1 is a block diagram of a noise reduction device for a vehicle-mounted voice recognition device according to a first embodiment of the present invention. In FIG.
Reference numeral 10 denotes a car audio device provided in the vehicle, which is composed of, for example, a radio, a cassette deck, a CD player, and the like. Reference numeral 20 denotes a car audio speaker, which is connected to the car audio device 10 and converts an audio signal from the car audio device 10 into sound and outputs the sound. Reference numeral 30 denotes first filter means, which is connected to the car audio device 10 and limits a band of an audio signal output from the car audio device 10. Reference numeral 40 denotes a first A / D conversion unit, which is connected to the first filter unit 30 and converts the band-limited audio signal output from the first filter 30 from an analog signal to a digital signal. Reference numeral 50 denotes a noise removing unit, which includes an adaptive filter unit 51, a coefficient updating unit 52, and a difference unit 53. Reference numeral 60 denotes a microphone for voice input, not shown in the present description, but for a vehicle-mounted voice recognition device.

[0008] A detection signal from the voice input microphone 60 is output as a digital signal to the noise removing unit 50 via a second filter means 70 and a second A / D conversion means 80. The voice recognition unit 9 is provided after the noise removal unit 50.
0 is provided, and speech recognition is performed from the signal from which the noise has been removed by the noise removing unit 50. Although not shown, the voice recognition result is configured to be displayed on a monitor screen of a car navigation system provided in the vehicle, or output with voice guidance, for example.

(Description of Operation) The operation will be described with reference to FIGS. First, the operation when noise removal is not performed will be described. An operator in the vehicle (generally considered a driver, henceforth called a driver)
By activating the car navigation device with the voice recognition function, for example, the destination is uttered as “Sapporo”. The uttered voice signal is detected by a voice input microphone 60 provided near the driver, and is converted into an electric signal.
At this time, when the audio device 10 is operated, music is output from the speaker 20, so that the music output from the speaker 20 and the voice uttered by the driver are simultaneously input to the voice input microphone 60. Therefore, for example, a signal having a time-series waveform as shown in FIG. 2A is output from the voice input microphone 60 to the second filter means 70.

The second filter means 70 is a low-pass filter for anti-aliasing (hereinafter referred to as LPF).
That is, high-frequency components of a specific band or more of the output signal from the audio input microphone 60 are removed, and the signal is output to the second A / D converter 80. The second A / D converter 80 converts the band-limited analog signal output from the second filter 70 into a digital signal and performs voice recognition. However, in the voice recognition in this case, the recognition must be performed not only from the voice signal but also from the signal including the music signal from the car audio device 10.
The music signals output from the car audio device 10 and the speaker 20 often include not only musical instruments but also voice signals as vocals, and the frequency component is the frequency of the voice spoken by the driver to the voice recognition device. Since it is exactly the same as the band, the recognition accuracy is greatly degraded.

Next, the case where the noise reduction device according to the present embodiment is used will be described. The audio signal from the car audio device 10 is output to the speaker 20 and
It is also output to the first filter means 30. The first filter means 30 is an LPF for anti-aliasing, and has the same characteristics as the second filter means 70.
The first filter means 30 removes high-frequency components of a specific band or more from the audio signal output from the car audio device 10 and outputs the signal to the first A / D conversion means 40. The first A / D converter 40 converts the audio signal band-limited by the first filter 30 from an analog signal to a digital signal, and outputs the signal to the noise removing unit 50. The noise removing unit 50 includes, for example, an adaptive filter unit 51 composed of an FIR filter, a coefficient updating unit 52 for updating the coefficients of the adaptive filter as needed, and a difference unit 53. The audio signal d which is band-limited by the first filter means 30 and converted to a digital signal by the first A / D conversion means 40 is
It is input to the adaptive filter unit 51 and the coefficient updating unit 52. The digitally converted audio signal d is applied to the adaptive filter unit 5
1 and the difference means 5 as an estimated noise (music) signal y.
3 is output. The difference means 53 is a digitally converted microphone detection signal x output from the second A / D conversion means 80 and an estimated noise (music) from the adaptive filter unit 51.
The difference from the signal y is output as an error signal e to the coefficient update unit 52 and the speech recognition unit 90. That is, when the car audio device 10 is operating, the microphone detection signal x is (voice + music signal), and the error signal e obtained by subtracting the estimated noise (music) signal y from this signal is the voice signal. Becomes

The coefficient updating section 52 updates the filter coefficients of the adaptive filter section 51 from time to time so that the error signal e becomes smaller from the digitally converted audio signal d and error signal e. As a coefficient updating method, for example, LMS (L
east mean square) algorithm or the like is used. Thus, it is possible to estimate how the audio signal output from the car audio device 10 is detected via the speaker 20, the interior space, and the microphone 60 for voice input, so that the estimated noise (volume) signal y If a difference between the actually detected signals x is obtained, a time-series waveform from which noise (music signal) is removed as shown in FIG. 2B can be obtained. Therefore, even when the car audio is operating, the voice input microphone 60 is used. A very high recognition rate can be realized because the background music signal mixed into the image can be removed.

As described above, according to the first embodiment of the present invention, an audio signal output from a car audio device provided in a vehicle and a voice input microphone provided near a speaker are detected. Since the music signal from the car audio device mixed into the audio input microphone is estimated from the detected signal and removed from the detected signal, the following effects can be obtained.・ There is no need to stop car audio for voice recognition. -Since a voice recognition device that can obtain a high recognition rate can be constructed even while listening to car audio, a practical vehicle-mounted voice recognition device can be realized.・ Since the microphone for voice input does not require special characteristics and can be used without worrying about unidirectional and omnidirectional characteristics, there is no need to spend man-hours on microphone selection and low-cost microphones can be used. Since it can be used, an increase in cost can be suppressed.

(Second Embodiment) (Description of Configuration) FIG. 3 is a block diagram of a noise reduction device according to a second embodiment of the present invention. In FIG. 3, reference numeral 21 denotes a music signal detection microphone, and other components are the same as those in the first embodiment.

Description of (Operation) The operation of the second embodiment will be described with reference to FIG. Elements that are assigned the same numbers as the components of the first embodiment are:
The third embodiment has the same configuration and operation as the first embodiment.

A feature of the second embodiment resides in a method for acquiring an audio signal. In the first embodiment, the digitally converted audio signal d is directly obtained from the electric signal before being converted into a sound. In the second embodiment, however, the audio signal d is output from the car audio device 10 and converted into a sound by the speaker 20. The music signal is obtained from a microphone 21 for music signal detection provided near the speaker 20. That is, in the second embodiment, the audio signal output from the car audio device 10 is output from the speaker 20 as sound into the vehicle. The audio signal output to the vehicle as a sound is detected by a music signal detection microphone 21 provided in the vicinity of the speaker 20, converted again into an electric signal, and converted into an electric signal. The noise signal is processed by the noise removing unit 50 as the digitally converted audio signal d via the conversion unit 40. The noise removal method is the same as the method described in the first embodiment.

As described above, according to the second embodiment of the present invention, the acoustic signal output as sound from the speaker 20 into the vehicle is transmitted to the microphone 21 for music signal detection provided near the speaker 20. Since the audio signal d is detected, converted into an electric signal, filtered, and A / D converted to obtain a digitally converted audio signal d, in addition to the effects of the first embodiment, the audio signal d is directly transmitted from the car audio device. It is not necessary to obtain a reference signal for noise removal by a very simple method of attaching a microphone for detecting a music signal afterwards near the speaker. By adding the noise reduction device of the present invention to a car audio device, a highly accurate voice recognition device can be realized, and cost increase can be suppressed. That.

(Third Embodiment) (Explanation of Configuration) FIG. 4 is a block diagram of a noise reduction device according to a third embodiment of the present invention. In FIG. 4, 100
Is an absolute value processing means, 110 is a peak value detection means, 120
Is a determination means, and the output signal from the second filter means 70 is output to the absolute value processing means 100, the peak value detection means 11
0, other components are the same as those of the first and second embodiments, except that they are output to the coefficient update unit 52 of the noise reduction unit 50 as the coefficient update processing control signal via the determination unit 120.

Description of (Operation) The operation of the third embodiment will be described with reference to FIG. First,
Elements assigned the same reference numerals as those of the second embodiment have the same configuration and operation in the third embodiment as in the first and second embodiments.

A feature of the third embodiment resides in a control method of the coefficient updating unit 52 for estimating a noise removal signal. In other words, when it is detected that a signal (audio signal) of a certain level or higher is input to the audio input microphone 60,
The coefficient updating unit 52 is configured to temporarily suspend the coefficient updating.

In the noise reduction device according to the present embodiment, a noise (music) removal signal is generated from an electric audio signal output from the car audio device 10 and a voice + music signal detected by the voice input microphone 60. Although only the music signal component is removed from the voice + music signal detected from the voice input microphone 60, when the driver speaks with an unexpectedly loud voice,
When a loud voice signal is input to the voice input microphone 60 due to coughing, sneezing, or the like, the signal is not only distorted or saturated, but also affects the coefficient update in the noise removing unit 50. The error signal e output from the removing unit 50 to the speech recognition unit 90 may be echoed, and as a result, the speech recognition accuracy may be temporarily deteriorated. For this reason, in the third embodiment, the absolute value processing unit 100 that performs full-wave rectification on the detection signal that is detected from the audio input microphone 60 and band-limited by the second filter unit 70, Peak value detecting means 110 for determining the peak value of the output signal
And comparing the level of the output from the peak value detection means 110 with the level of the predetermined determination signal Vth, and generating the coefficient update processing control signal for controlling the coefficient update processing of the coefficient update section 52 by the determination means 120. Provided. When the output signal from the peak value detecting means 110 is smaller than the predetermined determination signal Vth, the determining means 120 causes the coefficient updating unit 52 to continue the coefficient updating process, and
When the output signal of the reference numeral 10 is larger than the determination signal Vth, a coefficient update processing control signal for temporarily stopping the coefficient update processing is generated and output to the coefficient update unit 52 to control the coefficient update processing. For this reason, when a sound signal larger than expected is input to the sound input microphone 60,
Since a signal for temporarily stopping the coefficient update processing is output from the determination unit 120 to the coefficient update unit 52, the adaptive filter unit 51 generates the estimated music signal y while holding the coefficient immediately before the large signal is input. Good noise (music) elimination is performed without malfunction.

As described above, according to the third embodiment of the present invention, when the level of the detection signal detected by the voice input microphone 60 exceeds a certain value, the noise removing unit 50 Since the adaptive filter coefficient updating process performed in step (1) is temporarily stopped, in addition to the effect of the first embodiment, even if a signal larger than expected is input to the microphone Since a stable noise removal signal is always generated and noise removal can be performed without disturbing coefficient updating by the signal, a highly stable apparatus can be realized in any situation.

In the above embodiment, an example in which the present invention is applied to an on-vehicle voice recognition device, particularly a car navigation system with a voice recognition function, has been described. However, all voice recognition devices used in a known noise environment, telephones, etc. Available. That is, even when the audio or TV is turned on at home, a voice recognition word processor, a telephone, or the like can be used without stopping those devices, and information with reduced noise can be provided to the voice recognition unit and the other party.

[0024]

As described above in detail, according to the first aspect of the present invention, the first filter for limiting the band of the audio signal output to the speaker from the car audio device provided in the automobile. Means, first A / D conversion means for converting an analog output signal output from the first filter means into a digital signal, and first A / D conversion means.
From the digital audio signal output from the conversion means,
Adaptive filter means for generating an estimated music signal; a voice input microphone provided near an operator operating a voice input device in a vehicle, for detecting a voice generated by the operator; and the voice input microphone Second filter means for limiting the band of the detection signal from
A second A / D converter for converting an analog output signal output from the filter means into a digital signal; an estimated music signal output from the adaptive filter means;
A difference means for calculating a difference between the digitized microphone detection signals output from the A / D conversion means, an error signal obtained by the difference means, and a digital signal output from the first A / D conversion means. From the converted audio signal, coefficient updating means for updating the coefficient of the adaptive filter so that the error signal is reduced, based on the audio signal output from the car audio device, based on the audio input microphone Estimate the mixed noise (music) signal,
Noise from the detection signal from the voice input microphone
The following effects are achieved by adopting a configuration in which a (music) signal is subtracted.・ There is no need to stop car audio for voice recognition. -Since a voice recognition device that can obtain a high recognition rate can be constructed even while listening to car audio, a practical vehicle-mounted voice recognition device can be realized.・ Since the microphone for voice input does not require special characteristics and can be used without worrying about unidirectional and omnidirectional characteristics, there is no need to spend man-hours on microphone selection and low-cost microphones can be used. Since it can be used, an increase in cost can be suppressed.

According to a second aspect of the present invention, in the first aspect of the present invention, an absolute value processing means for performing full-wave rectification of the signal band-limited by the second filter processing means; The peak value detecting means for detecting the peak value of the signal output from the absolute value processing means, the signal output from the peak value detecting means, and a predetermined threshold signal are compared. When the output signal is smaller than the threshold signal, the coefficient updating means of the coefficient updating means continues the coefficient updating operation, and when the output signal is larger, the coefficient updating means control signal for temporarily stopping the coefficient updating operation is provided. With this configuration, in addition to the effect of the invention described in claim 1, even if a larger signal than expected is input to the microphone for voice input, the large signal
Since a stable noise removal signal is always generated and noise removal can be performed without disturbing coefficient updating, a highly stable apparatus can be realized in any situation.

According to a third aspect of the present invention, in the first or second aspect of the present invention, a microphone for detecting a music signal is provided near the speaker, and a microphone for detecting a sound signal from the speaker is provided. Since the audio signal is detected, there is no need to obtain the audio signal directly from the car audio device. Because the reference signal can be obtained, without having to replace all the existing devices,
By simply adding the noise reduction device of the present invention to a conventional car audio device, a high-accuracy speech recognition device can be realized, and an increase in cost can be suppressed.

[Brief description of the drawings]

FIG. 1 is a block diagram of a noise reduction device according to a first embodiment.

FIG. 2 is a diagram illustrating an example of noise removal.

FIG. 3 is a block diagram of a noise reduction device according to a second embodiment.

FIG. 4 is a block diagram of a noise reduction device according to a third embodiment.

[Explanation of symbols]

 Reference Signs List 10 car audio device 20 speaker 30 first filter means 40 first A / D conversion means 50 noise removing section 51 adaptive filter section 52 coefficient updating section 53 difference means 60 microphone 70 second filter means 80 second A / D D conversion means 90 Voice recognition unit

 ──────────────────────────────────────────────────の Continuing on the front page (72) Inventor Takashi Ichikawa 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. F-term (reference) 3D020 BA02 BA10 BA11 BC02 BE03 BE04 5D015 DD02 EE04 9A001 EE05 HH16 JJ77 KK31 KK54

Claims (3)

[Claims] A first filter unit for limiting a band of an audio signal output to a speaker from a car audio device provided in an automobile; and a digital signal converting an analog output signal output from the first filter unit to a digital signal. First A / D conversion means for converting the digital audio signal from the first A / D conversion means, adaptive filter means for generating an estimated music signal from the digital audio signal output from the first A / D conversion means, and operating a voice input device in the automobile A voice input microphone for detecting a voice generated by the operator, a second filter means for limiting a band of a detection signal from the voice input microphone, A / D conversion means for converting an analog output signal output from the filter means into a digital signal, and an output from the adaptive filter means Estimated music signal to be
A difference means for calculating a difference between the digitized microphone detection signals output from the second A / D conversion means, an error signal obtained by the difference means, and the first A / D
And a coefficient updating unit that updates coefficients of the adaptive filter so as to reduce an error signal from the digitally converted audio signal output from the converting unit, based on the audio signal output from the car audio device. A noise (music) signal mixed into the voice input microphone, and a noise (music) signal subtracted from a detection signal from the voice input microphone. 2. An absolute value processing means for performing full-wave rectification of a signal band-limited by the second filter processing means, and a peak value detection means for detecting a peak value of a signal output from the absolute value processing means. Comparing the signal output from the peak value detecting means with a predetermined threshold signal, and if the output signal from the peak value detecting means is smaller than the threshold signal, the coefficient updating operation of the coefficient updating means; 2. The noise reduction apparatus according to claim 1, further comprising: a determination unit that outputs a coefficient update unit control signal such that a coefficient update unit control signal is temporarily stopped when the coefficient update operation is large. 3. The noise reduction device according to claim 1, wherein a microphone for detecting a music signal is provided near the speaker, and an audio signal is detected from a sound generated from the speaker.