JPH0695694A - Noise reducing circuit for voice recognition device - Google Patents

Abstract

PURPOSE:To eliminate noises included in a voice so as to improve a recognition rate in a voice recognition device which recognizes the voice and controls equipment. CONSTITUTION:A voice recognition device noise reducing circuit, which cancels noises mixed with the voice so as to control equipment by the voice of the speaker in a vehicular room, is provided with an adaptive filter 22 which has a backup that stores and holds the filter coefficients to be updated so as to cancel stationary noises by a backup power supply, an adaptive algorithm forming means 23 which sequentially updates filter coefficients of the filter 22 against a stationary noise input and an adaptive speed control means 25 which makes the update adaptive speed of the filter coefficients formed by the means 23 smaller when nonstationary noise is inputted and the update adaptive speed of the filter coefficients, that are formed, is made larger when only stationary noise is inputted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device for recognizing a voice and controlling a device, and more particularly, the present invention relates to improving a recognition rate by reducing noise in a noisy environment.

[0002]

2. Description of the Related Art Heretofore, as such a technique, the recognition rate is extremely reduced in a noisy environment when performing voice recognition, and there has been a technique described below in order to prevent this reduction.
FIG. 3 is a diagram showing a conventional noise reduction circuit for a voice recognition device. The noise reduction circuit for a voice recognition device shown in the figure includes a delay unit 21 for delaying a signal from a voice capturing microphone for capturing a voice into which noise to be removed is mixed and converting the voice into an electric signal, and the delay unit 21. FIR (Fini) to form a cancellation signal that removes the noise that is connected and mixed into the voice
te Impulse Response) and an adaptive filter 22,
An adaptive algorithm forming means 23 for updating the filter coefficient forming the adaptive filter 22 by the least squares method or the learning identification method using the canceled voice signal described later.
And an adder 24 that adds the signal from the voice capture microphone and the cancel signal from the adaptive filter 22 to form only a voice output.

The adaptive filter 22 is designed so that the difference between the original signal and the delayed signal becomes zero, that is, the original signal is predicted from the delay information. Although noise is generated in the vehicle interior when a vehicle or the like travels, the noise is periodic and steady when the vehicle travels constantly. The adaptive algorithm forming means 26 learns noise and forms filter coefficients of the adaptive filter 22. When a voice is generated to control the device in the vehicle cabin, the voice is sudden and non-stationary, so that the filter coefficient is not updated by this voice and is canceled by the adder 24 by the adaptive filter 22. Is output without any output. Thus, when the voice recognition device performs the voice recognition, the noise mixed in the voice in the noise environment is removed, and the recognition rate is improved. That is, noise reduction is performed in front of the voice recognition device, and only the voice buried in the noise is extracted and input to the voice recognition device. By the way, in the reduction circuit for the voice recognition device,
As described above, digital signal processing is always performed by the adaptive filter, but if the adaptation speed is too fast, the filter coefficient is updated even in the voice section and adaptation is performed, and not only the noise signal but also part of the voice is canceled. There were times when it ended up. Specifically, there is a repetitive component in the latter half of the voice such as "aa", and therefore a part of the latter half is canceled. In order to prevent the partial cancellation of the voice, it is necessary to monitor the output of the adder 24 and change the update speed according to the output. Therefore, an adaptive speed control means 25 for controlling the adaptive speed of the adaptive algorithm forming means 23 by receiving the input signal and the output signal of the adder 24 is provided. Thus, when the voice section is detected, the update of the filter coefficient of the adaptive filter 22 is stopped so that the noise is canceled by the filter coefficient learned so far. In this case, the update of the filter coefficient is almost stopped, so that the cancellation of the voice is prevented from being stopped.

[0004]

However, in the conventional noise reduction circuit for a voice recognition device, since the filter coefficient of the adaptive filter is not learned so that the noise is canceled in the initial state when the power is turned on, the initial value is reduced. Equivalent time is required, but it is not possible to determine whether there is output by voice or whether noise is not canceled and there is output, and on the other hand, the adaptive speed control means determines that voice is input and the update is almost stopped. However, there is a problem in that the update speed becomes slow despite the noise input and the time required to obtain satisfactory performance becomes long. That is, each time the system is started up, there is a constraint that it cannot be used satisfactorily for a certain period of time.

Therefore, in view of the above problems, it is an object of the present invention to provide a noise reduction circuit for a voice recognition device which can reduce the initial equivalent time each time the system starts operating.

[0006]

In order to solve the above problems, the present invention has a backup in a noise reduction circuit for a voice recognition device for canceling noise mixed in voice for controlling equipment by voice of a speaker in a vehicle compartment. An adaptive filter, adaptive algorithm forming means and adaptive speed control means are provided.

The backup adaptive filter stores and retains a filter coefficient, which is updated to cancel the stationary noise, by a backup power supply. The adaptive algorithm forming means sequentially updates the filter coefficient of the adaptive filter with respect to a steady noise input. The adaptive speed control means reduces the adaptive speed of updating the filter coefficient formed by the adaptive algorithm forming means when the unsteady voice is input, and when only the stationary noise is input, The adaptive speed for updating the formed filter coefficient is increased.

[0008]

According to the noise reduction circuit for a voice recognition apparatus of the present invention, the backup adaptive power supply stores and holds the filter coefficient updated in the adaptive filter with backup so that the stationary noise is canceled. Therefore, the filter coefficient updated when the system power is turned off is stored and held by the backup power supply, and when the system power is turned on, the stored filter coefficient is immediately used. Therefore, when the system power is turned on, the learning-created filter coefficient is formed. There is no need to improve the performance at the start of adaptation. In the adaptive algorithm forming means, the filter coefficient of the adaptive filter is sequentially updated with respect to steady noise input, and the adaptive speed control means uses the adaptive algorithm forming means when the non-steady voice is input. It is prevented that the adaptive speed of the update of the formed filter coefficient becomes small and the voice is prevented from being canceled, and when only the stationary noise is input, the adaptive speed of the update of the formed filter coefficient becomes large and the noise becomes high. Only canceled.

[0009]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing a control system using a voice recognition device according to an embodiment of the present invention. As shown in the figure, a control system using a voice recognition device includes two voice capturing microphones 200 that capture a voice of a speaker in a vehicle interior 300 of a vehicle, and the voice capturing microphones 20.
A speaker direction distance determination unit 201 that identifies a speaker based on the direction of voice from 0 and a distance from a sound source, and an automatic gain control (AGC) unit 202 connected to the speaker direction distance determination unit 201.
And noise capturing microphone 201 and vibration sensor 20
The noise reduction circuit 203 for voice recognition for canceling the noise mixed in the signal from the automatic gain control (AGC) by 2
And a voice recognition unit 204 which is connected to the voice recognition noise reduction circuit 203 and recognizes which registered word matches the voice whose noise has been canceled, and the voice recognition unit 204.
Various control units 205 that form control signals based on the words recognized in step S, a voice synthesis unit 206 that synthesizes the words recognized through the various control units 205 into speech, and a synthesis unit 206 that is connected to the speech synthesis unit 206 to synthesize the words. Loudspeaker 2 that reproduces the reproduced sound
07, an audio 208 controlled by the various control units 205, an air conditioner 209, and a telephone 21.
0, navigation 211, auto drive 212
Including etc.

FIG. 2 is a diagram showing a noise reduction circuit for a voice recognition apparatus according to an embodiment of the present invention. The noise reduction circuit for a voice recognition device shown in the figure is an A / D converter (Ana) that converts an analog signal from the automatic gain control unit 202 into a digital signal.
log To Digital Converter) 20 and the A / D converter 2
A delay unit 21 for delaying the digital signal from 0, an adaptive filter 22 made of FIR for forming a cancel signal connected to the delay unit 21 for removing noise mixed in the voice, An adaptive algorithm forming means 23 for updating the filter coefficient forming the adaptive filter 22 by the method of least squares of the voice signal, and a signal from the voice capturing microphone and an inverted cancel signal from the adaptive filter 22 are added. And an adaptive filter 22 for inputting an input signal and an output signal of the adder 24 to control an adaptive speed of the adaptive algorithm forming means 23, and the adaptive filter 22. For example, RAM (Random Access Memory) supplies power even when the system power is turned off. ; And a backup power supply 26. The filter coefficients stored and held by the backup power supply 26 are as follows. The filter coefficient of the adaptive filter 22 at time n is a vector Cn, the tap of the filter is a vector Tn, and the convergence coefficient that determines the update speed is α. Where α is the power square of the input | Sn | ² and the power square of the output is | Pn, as shown in FIG.
| ² is determined by the following formula. α = α ₀ / (1 + X · | Pn | ² / | Sn | ² ); X,
α ₀ is an arbitrary constant. Using this α, the vector Cn
Is updated by the learning identification method as follows.

Vector Cn + 1 = vector Cn + αPn
Vector Tn / | Tn | ^{2 The} above filter coefficient is stored and retained in the adaptive filter 22 by the backup power supply 26 when the system power is cut off. When the system power is turned on and then turned on, the above filter coefficient is used as the initial state and the noise cancellation functions from the beginning. Therefore, the square ratio of the output power and the input power in the initial state due to the noise input as in the conventional case | Pn | ²
/ | Sn | ² never increases. Therefore, α does not decrease significantly after the system power is turned on.
That is, the adaptive speed control means determines that the noise input is a voice input and reduces the convergence coefficient so that the adaptive speed is not reduced. For this reason, it becomes unnecessary to perform initial equalization by learning the filter coefficient when the system power is turned on as in the conventional case, and noise can be canceled from the time when the system power is turned on.

[0012]

As described above, according to the present invention, in the adaptive filter with backup, the filter coefficient updated when the system power is cut off is stored and held by the backup power supply, and when the system power is turned off, the filter coefficient is stored. Since the filter coefficient is immediately used, it is not necessary to form the filter coefficient by learning when the system power is turned on, and the performance at the start of adaptation can be improved.

[Brief description of drawings]

FIG. 1 is a diagram showing a control system using a noise reduction circuit for a voice recognition device according to an embodiment of the present invention.

FIG. 2 is a diagram showing a noise reduction circuit for a voice recognition device according to an embodiment of the present invention.

FIG. 3 is a diagram showing a conventional noise reduction circuit for a voice recognition device.

[Explanation of symbols]

 22 ... Adaptive filter 23 ... Adaptive algorithm forming means 24 ... Adder 25 ... Adaptive speed control 26 ... Backup power supply 26

Claims (1)

[Claims] 1. A noise reduction circuit for a voice recognition device for canceling noise mixed in voice for controlling a device by a voice of a speaker in a vehicle interior, which is updated to cancel the constant noise. An adaptive filter with backup (22) that stores and retains the generated filter coefficient by a backup power source, and the adaptive filter (22) for steady noise input
The adaptive algorithm forming means (23) for sequentially updating the filter coefficient of (1) and the adaptive speed for updating the filter coefficient formed by the adaptive algorithm forming means (23) are reduced when the non-stationary voice is input. A noise reduction circuit for a voice recognition device, comprising: an adaptive speed control means (25) for increasing an adaptive speed for updating the formed filter coefficient when only the stationary noise is input.