JPH02254499A - Voice input device - Google Patents

Abstract

PURPOSE:To reduce other noise component than an object voice and to output a clear sound signal by analyzing a frequency characteristic of a sound signal, and adjusting a frequency characteristic of a variable filter means in accordance with a result of analysis. CONSTITUTION:A signal mainly composed of a voice detected by a front microphone 2 is inputted to a noise canceller 200 through an LPF 10, an A/D converter 20, a BPF 30 and an amplifier 40. On the other hand, a signal mainly composed of a noise detected by a rear microphone 6 is inputted to the canceller 200 through an LPF 50, an A/D converter 60, a BPF 70 and an amplifier 80. A CPU 90 analyzes a voice of a drive within a prescribed time, when an ignition switch IGS is turned on, and adjusts a frequency characteristic of the BPF 30 and 70. In such a way, an unnecessary noise component is eliminated, and a clear sound signal can be outputted to a navigation unit 400, etc.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] [Industrial Application Field] The present invention relates to a voice input device, and can be used, for example, when controlling on-vehicle equipment by voice recognition.

[Prior Art] For example, when a driver operates various in-vehicle devices in a vehicle, and there are a large number of operable switches, he has to search for the position of the switch to be operated and check until his finger is positioned on the switch. Since this must be done visually, there are safety issues when the vehicle is in motion.

Therefore, in order to eliminate the need for visual operations, it has been proposed to use a voice recognition device to control in-vehicle equipment through voice input.

However, especially in a car, the level of external noise that is input to the recognition device together with the input voice may be large, which may significantly impede voice recognition. External noise is the sound from outside the car.

There are the engine sounds of your own troops, the sounds from the sound equipment inside the car, etc.

Therefore, the technique disclosed in Japanese Patent Publication No. 63-29755 is used to reduce the noise level in preparation for voice input when a specific switch is operated.

Lower the output of the sound device! ! , trying to close the glass.

[Problems to be Solved by the Invention] According to the conventional example (Japanese Patent Publication No. 63-29755), it is possible to reduce the influence of noise and improve the recognition rate of the speech recognition device. While doing so.

Each time voice recognition is performed, the driver must operate a switch and wait until the window glass is fully closed, so it takes time before voice input can begin, and the driver is forced to perform cumbersome operations.

SUMMARY OF THE INVENTION An object of the present invention is to output clear audio signals by reducing noise components other than target audio in various audio input devices.

[Structure of the Invention] [Means for Solving the Problems] In order to solve the above problems, the present invention provides a voice input means for converting voice into an electric signal; Variable filter means for adjusting frequency characteristics; Switch means; 1 for analyzing the electrical signal output from the audio input means and detecting its frequency characteristics;
frequency characteristic detection means; and in response to the operation of the switch means, the electrical signal output from the audio input means is analyzed by the frequency characteristic detection means, and the frequency characteristic of the variable filter means is determined according to the analysis result. A control means for adjusting is provided.

[Function] In the present invention, audio is converted into an electrical signal by audio input means such as a microphone, and then passed through variable filter means and output.

The variable filter means constitutes a bandpass filter, for example, and reduces noise components outside the frequency band of the target voice. Further, the frequency characteristics of the variable filter means are adjusted in the adjustment mode activated in response to the operation of the switch means.
Automatically set by the control means.

That is, when an actual speaker (e.g., a driver) activates the adjustment mode by operating a switch means in a state where there is substantially no external noise (e.g., before starting the engine of a car), the speaker makes a sound. Then, the frequency characteristics of only the voice components that do not include noise are analyzed, and a variable filter is applied so as to correspond to the frequency characteristics, that is, to pass only the signal in the voice band of the speaker and reduce other frequency components. When the frequency characteristics of the means are adjusted to perform 4° speech recognition, it has been conventionally practiced to provide a filter to reduce noise components. However, since there are large individual differences in the frequency characteristics of human speech, it is not possible to narrow the passband of the filter when performing speech recognition for unspecified speakers. It was not possible to sufficiently reduce noise components outside the user's frequency band.

In contrast, if the present invention is used, the frequency characteristics of the actual speaker's voice are measured in response to the operation of the switch means,
Since the frequency characteristics of the variable filter means are automatically adjusted to suit it.

The band of the filter can be narrowed, and noise components at frequencies outside the frequency band of one speaker can be significantly reduced.

Note that the voice input device of the present invention is applicable not only to voice recognition, but also to various applications where it is necessary to extract only the voice signal of a specific speaker under noisy conditions (for example, telephone, broadcasting, etc.).

Needless to say, it can also be used for recording, etc.).

Other objects and features of the present invention will become apparent from the following description of embodiments with reference to the drawings.

[Example] Fig. 1 shows the appearance of the interior of an automobile equipped with a device implementing the present invention. This embodiment is equipped with a system for operating the navigation device through voice input.

Referring to FIG. 1, a monitor television 3 of a navigation device is incorporated in the center console portion, and a navigation eye unit 4, which is the main body of the navigation device, is provided in a hidden position near the monitor television 3. In addition, the approximate center of the dash board 1 above the monitor television 3 (
A microphone 2 is incorporated in the center of the vehicle in the left-right direction. This microphone 2 is provided to detect the voice uttered by the driver.

Further, speakers 7 and 8 are installed on the rear tray 5 at mutually symmetrical positions on the left end and the right end. These speakers 7.8 are used to convert sound signals from an on-board audio device and a television receiver into sound and output the sound. Furthermore, on the rear tray 5, at the center in the left and right direction (at a position equidistant from the speakers 7 and 8),
A microphone 6 is incorporated. This microphone 6
is provided mainly to detect the sound output from the speaker 7.8 as noise during speech recognition.

In other words, the front microphone 2 simultaneously detects the driver's voice, the sound coming from the speaker 7.8, and various air vibrations, so the electrical signal detected by the microphone 2 contains a relatively large noise component. is often included. Therefore, by detecting the noise component signal with the rear microphone 6 and synthesizing the electrical signals detected by the two microphones 2 and 6, an audio signal with a relatively high clarity and a low noise component level is extracted. I have to.

By the way, when playing the sound of a stereo sound source.

The two speakers 7.8 generate mutually different sounds. Therefore, in order to remove each noise component emitted from the two speakers 7.8 from the signal detected by the microphone 2, it is necessary to remove each detected noise component and the noise component detected by the microphone 2. It is necessary to match the time and amplitude of -, respectively.

However, in this embodiment, the front microphone 2 is placed at the center of the front, and the speakers 7.8, which are noise sources, are placed at symmetrical positions in the vehicle interior.
The propagation delay time until the sound emitted from the left speaker 7 is detected by the microphone 2 and the propagation delay time until the sound emitted from the right speaker 8 is detected by the microphone 2 substantially match. Moreover, the rear microphone 6
is placed in the center of the two speakers 7 and 8, so
The propagation delay time until the sound emitted from the left speaker 7 is detected by the microphone 6 matches the propagation delay time until the sound emitted from the right speaker 8 is detected by the microphone 6.

Therefore, the time difference TdL between when the same sound emitted from the left speaker 7 is detected by the front microphone 2 and when it is detected by the rear microphone 6,
The time difference Tdr between when the same sound emitted from the microphone 2 is detected by the front microphone 2 and when the same sound is detected by the rear microphone 6 is the same.

In other words, if the signal detected by microphone 6 is delayed by a specific time difference Td (=Tdl=Tdr), the timing of the delayed signal can be made to match the timing of the noise component of the signal detected by microphone 2. By extracting the difference between them, it is possible to reduce the noise emitted from both the left and right speakers and make the audio component clear. In this case, only 1ffi is required for the microphone for detecting the noise component and the electric circuit for processing the signal detected by the microphone, which simplifies the configuration of the device.

Description will be made with reference to FIG. 2, which shows the configuration of an electrical component of the voice input device mounted on the vehicle of FIG. 1. The signal mainly consisting of voice detected by the front microphone 2 is
After high-frequency components (frequency equal to or higher than half of the sampling frequency of A/D conversion) are removed by a low-pass filter 10, the signals are converted into digital signals by an A/D (analog/digital) converter 20, and then passed through a band-pass filter 30. The signal passes through the amplifier (multiplier) 40 and is applied to one input terminal of the noise canceller 200. Further, a signal mainly composed of noise detected by the rear microphone 6 is converted into a digital signal by an A/D converter 60 after high frequency components are removed by a low-pass filter 50, and then passed through a band-pass filter 70. street, amplifier (multiplier)
80 and is applied to the other input terminal of the noise canceller 200.

The bandpass filters 30 and 70 are digital filters that pass only signal components of frequencies that match the frequency band of the voice of one speaker, that is, the driver. Of course, since there are large individual differences in the frequency band of human speech, adjustment work is required to adapt the characteristics of the bandpass filters 30 and 70 to the characteristics of the actual speaker.

However, in this embodiment, the bandpass filter 30
．． Since the frequency characteristics of 70 are configured to be adjusted automatically, there is no need for adjustment during normal audio input.

That is, a microcomputer (hereinafter referred to as CPU) 90 provided in this device controls the A/D under certain conditions.
Analyzing the frequency characteristics of the signal output from the converter 20,
The characteristics of the bandpass filters 30.70 are adjusted according to the results.

To the CPU 90, to give instructions to it.

Ignition switch IGS and cancel switch C
3 is connected.

An outline of the operation of the CPU 90 is shown in FIG. 3, and the operation of the CPU 90 will be described with reference to FIG. Step 1 identifies whether the electrical contacts of the ignition switch IGS are in the accessory position ACC. When the IGS reaches the ACC position, the process proceeds to the next step.

In step 2, timer 1 is started, and in step 3, time t1 counted by timer 1 is referred to.

When tl reaches 0.5 seconds, that is, when 0.5 seconds have passed since the ignition switch was placed in the ACC position, the process proceeds to step 4.

In step 4, start timer 2 and proceed to step 5.
Proceed to. In step 5, digital signals outputted from the A/D converter 20 at a constant sampling period are sequentially input and stored on the internal memory. The processing in step 5 is performed at t2. That is, the process is repeated until the count value of timer 2 reaches 2 seconds. In other words, digital data of acoustic information detected for two seconds is stored in the memory.

While the CPU 90 executes steps 5 and 6, the driver, who is the speaker, needs to utter arbitrary words.

When the driver speaks for two seconds, information on the audio waveform is measured and the results are stored in memory.

Proceeding to step 7, the CPU 90 executes a known fast Fourier transform (FFT) process on the audio waveform information stored in the memory. As a result, data on the frequency distribution of the measured audio energy of the driver is obtained.

This data is further processed to determine the upper limit frequency and lower limit frequency of the audio frequency band. Specifically, in this example,
The upper and lower limits of all frequency ranges in which energy of a predetermined ratio or higher is detected relative to the maximum value of the energy distribution are determined.

In the next step 8, a selection signal matching the upper and lower frequency limits determined in step 7 is generated.

Referring to FIG. 2, the ROM connected to the CPU 90
In 100, data of coefficients that determine the frequency characteristics of the bandpass filters 30 and 70 are stored in advance in different address areas, divided into frequency bands.

Therefore, the CPU 90 generates a signal SEL for selecting the memory address storing the coefficient data assigned to the frequency band determined in step 7, and applies it to the address terminal of the ROM 100.

In the next step 9, the state of the cancel switch C8 is checked. If the cancel switch SC is on, the process returns to step 2 and the analysis process is performed again.

In the next step 10, the ignition switch IGS
is at the 08 position. If the ON position is detected, the process proceeds to step 11.

In step 11, the bandpass filters 30 and 70 are controlled to cause each filter to latch the coefficient data Pf output from the ROM 100 as a coefficient.

In step 12, the ignition switch ■GS is turned to O.
Identifies whether it is in the FF position or not. ○When the FF position is detected,
Then return to step 1.

In other words, the car engine is stopped (IGS
When the ignition key is moved from the OFF position to the ACC position, the system enters the mode that analyzes the frequency characteristics of the driver's voice. The characteristics of the bandpass filter 30.70 are automatically adjusted to suit.

Note that the CPU 90 includes a nonvolatile memory therein, and the state of the selection signal SEL output by the CPtJ 90 is saved even when the power of the device is turned off. In this embodiment, once the characteristics of the bandpass filter are adapted to the characteristics of a single speaker, there is no need to make the adjustment again unless the speaker, ie, the driver, changes.

In other words, if the bandpass filter has been adjusted, the driver only needs to perform the normal operation when starting the engine of the stopped suit. If the driver switches the ignition key from the OFF position to the ACC position and then immediately switches it to the ON position without waiting, the process of the CPU 90 will be performed in step 1-2-3- of FIG.
13 and proceeds to 11, the 1 selection signal SEL is not updated, and the selection signal SEL is the same as before the power was turned off.
Accordingly, the characteristics of the bandpass filter 30.70 are set to the same characteristics as the previously adjusted frequency band characteristics.

Referring again to FIG. 2, the noise canceller 200 includes a delay circuit 210. The adaptive filter 220 and the coefficient control circuit 230 are different. The delay circuit 210 outputs a signal dk obtained by delaying the input signal for a predetermined period (6,4 m5ec). The adaptive filter 220 has 127 stages of delay elements (Z
""), 128 stage variable amplification element (A1-A128:
The filter includes a multiplier) and 128 stages of addition elements, and the characteristics of this filter can be varied in various ways by adjusting the coefficients set for each of the variable amplification elements.

The coefficient control circuit 230 outputs the output from the delay circuit 210 (,
The difference ek between the No. 1 dk and the signal yk output by the adaptive filter 220 is input, a group of coefficients whose root mean square value is minimized is generated, and these coefficients are applied to each of the variable amplification elements. In this example, the coefficient control circuit 230 is a known LMS.
It is designed to run an algorithm.

In other words, if we consider the driver's audio as the target signal component <S) and all other sounds as noise components (N), the electrical signal applied from the amplifier 40 to the noise canceller 200 contains both S and N. contains the components of amplifier 8
Since the electrical signal applied to the noise canceller 200 from 9 is mainly composed of N components, minimizing the result of combining both means extracting only the S component.

Although the noise components included in the two electrical signals input to the noise canceller 200 are different in time and amplitude,
The difference can be eliminated by passing each signal through a delay circuit 210 and an adaptive filter 220.

The extracted audio signal output by the noise canceller 200 is applied to a voice recognition unit 300, which executes a known recognition algorithm to recognize the voice uttered by the driver; Recognized voice information and predetermined command words (enlarge, reduce, on, off).

(top, bottom, right, left, etc.).
If it matches, the command signal corresponding to the command word is
Output to navigation unit 400.

Therefore, in this example, the driver's voice input can provide instructions to and control the navigation unit 400.

Note that when the ignition switch IGS is in the ACC position, power is also supplied to the accessory audio device, so if the power switch for that device is on, even in the mode that measures the frequency band of the driver's audio, power is supplied from the audio device. Output sound, ie noise, may be detected and cause errors in measurements. Therefore, in that case, it is desirable to change the configuration to the following. That is, in addition to providing a switch such as a relay to turn on/off the power of the audio device, in step 4 of FIG.

Turn off the switch to cut off the power to the audio device,
In step 7, the control is changed to turn on the switch and permit power supply to the audio device.

Note that in the above embodiment, the bandpass filter 30
．． a CPU 90 that adjusts the characteristics of the noise canceller 200; and a coefficient control circuit 2 that controls the adaptive filter of the noise canceller 200;
Although 30 and 30 are provided independently, it is also possible to execute both processes by one CPU.

Further, in the embodiment, the frequency analysis of the driver's voice and the adjustment of the bandpass filter are performed by digital signal processing, but these processes can be performed in the same way by replacing them with analog electric circuits. In the case of replacing the bandpass filters 30 and 70 with analog circuits, for example, if a large number of capacitors are configured to be switched by analog switches, the frequency characteristics of the filters can be switched in the same way as in the case of digital processing.

Furthermore, although the embodiment shows the case of speech recognition, even when the present invention is used for simply processing analog signals, it is possible to identify a specific speaker from acoustic signals containing various noises. It is possible to extract only the audio components.

[Effects] As described above, according to the present invention, the frequency characteristics of the variable filter means can be made to accurately match the frequency band of the actual speaker, so the passband width of the filter can be sufficiently narrowed. All unnecessary frequency components can be removed as noise and only the necessary audio components can be extracted. Moreover, since the characteristics of the variable filter means are automatically adjusted, there is no need for operational hassle.

[Brief explanation of drawings]

FIG. 1 is a perspective view showing the exterior of the interior of an automobile equipped with the device of the present invention. FIG. 2 is a block diagram showing the configuration of the voice input device of the embodiment. FIG. 3 is a flowchart showing an outline of the operation of the CPU 90. 2: Microphone (audio input means) 3: Monitor TV 7, 8: Speaker 10.50:
Low-pass filter 20.60: A/D converter 30.70: Band-pass filter (variable filter means) 90: Microcomputer (frequency characteristic detection means,
Control means) 100: ROM 200: Noise canceller 210: Delay circuit 220: Adaptive filter 230:
Coefficient control circuit 300: Voice recognition unit 400: Navigation unit IGS: Ignition switch (switch means) C8
:Cancel switch

Claims (4)

[Claims] (1) Audio input means that converts audio into an electrical signal; Variable filter means that adjusts the frequency characteristics of the electrical signal output from the audio input means; Switch means; Analyzes the electrical signal output from the audio input means frequency characteristic detection means for detecting the frequency characteristics thereof; and in response to the operation of the switch means, the frequency characteristic detection means analyzes the electrical signal output from the voice input means, and according to the analysis result. A voice input device comprising: control means for adjusting the frequency characteristics of the variable filter means. (2) The control means starts the analysis operation of the frequency characteristic detection means when a first predetermined time has elapsed since the switch means became in the first state, and when a second predetermined time has elapsed. The voice input device according to claim 1, wherein the analysis is terminated at a certain time. (3) The switch means is an ignition switch of the vehicle, and the control means starts an analysis operation of the frequency characteristic detection means in response to switching the switch means to the accessory position, and ends the analysis. 2. The audio input device according to claim 1, wherein if the switch means is previously switched to the on position, the analysis operation is stopped and the frequency characteristic of the variable filter means maintains the previously set state. (4) The voice input device according to claim 1, wherein a voice recognition means is connected to the output terminal of the variable filter means, and the voice recognition means controls onboard equipment according to the recognized voice.