JPH02184898A - Method and device for voice input - Google Patents

Abstract

PURPOSE:To allow the input of voice command words at high S/N in the sound field of noise environment by dispersely disposing one piece of main microphone and plural pieces of reference microphones in the sound field, then selecting the output of the reference microphone having the highest intercorrelativeness with the main input and executing the noise cancellation of these two inputs. CONSTITUTION:This device is constituted of an intercorrelation deciding unit 20 and an adaptive noise cancellation unit 30 connected behind the unit 20. One piece of the main microphone 1 and plural pieces of the reference microphones 2 are dispersedly disposed in the sound field. The output of the reference microphone 2 decided to have the highest intercorrelativeness with the noise component of the main microphone 1 is used as the reference input and the output of the main microphone as the main input. These two inputs are further subjected to the cancellation of the noise components and are inputted to a void recognizing device 40. The inputting of the voice command words in the sound field of the noise environment with the high S/N is possible in this way.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to an improvement in a voice input method and device, and more specifically, the present invention relates to an improvement in a voice input method and device. The present invention relates to a voice input method and device.

[Prior Art] As a recent trend, audio equipment such as radios and car stereos and air conditioners have come to be mounted on moving vehicles, such as passenger cars. Conventionally, this was done by operating a switch, but recently the idea of operating by voice command has been proposed.

However, these audio equipment and air conditioners emit noise in the same frequency band as the voice frequency, so when attempting to operate the drive using voice commands, the noise from the onboard audio equipment and air conditioners overlaps with the voice command words, making accurate detection difficult. and may not be recognized, resulting in malfunction or non-operation.

In addition to the noise emitted from in-vehicle devices, external noise is also known to enter the interior of the car through the windows and sunroof when the car is driven through relatively noisy areas in urban areas. .

As a voice input method for accurately recognizing voice command words in such a noisy environment, there is a "voice load drive device" described in Japanese Patent Publication No. 63-29755. In order to lower the volume level of all in-vehicle devices that emit sounds that affect voice commands, and to prevent external noise from entering the cabin,
Close opening/closing devices such as windows and sunroofs, and
It is taught that an input delay means should be provided to delay the input of voice commands until the volume level of equipment outside the vehicle and noise outside the vehicle is reduced.

In addition, the academic journal published by the Institute of Electronics and Communication Engineers, Journal of the Institute of Electronics and Communication Engineers.
Volume 69, No. 6 (June 1986) pp601-60
In the technical outlook described in Section 4, ``Latest audio equipment and signal processing technology for the realization of high-quality public address communication systems,'' microphones M and Ml are arranged with an interval d as shown in Figure 9. At the same time, an adaptive filter (Adapt-ive Digital F
ilter (hereinafter abbreviated as rA, D, FJ))61
．． 62 was connected and the input was made to the microphones M, , M2 (
When noise is added through A, D, F6+, 62 (by adder 7) at angle θ), the sound is received with a time delay (te=dsinθ/c, c: speed of sound) with respect to microphones M+ and Ma.

Therefore, if the filter coefficients are set so that H, (ω) has a transfer characteristic representing a delay and a negative phase, and Hz (ω) has a transfer characteristic of 1, then A, D, F6 . ．． 6□
It is taught that the noise can be completely eliminated by adding the passed signals using an adder 7.

According to an adaptive noise cancellation microphone array of the type described above, a directional pattern is formed with valleys in the direction of arrival of the noise.

Therefore, the noise is eliminated, but the audio is also distorted, so
Not pleasant in terms of pitch. Therefore, as shown in FIG. 10, the above-described adaptive noise/sound canceling microphone array is constructed using four-element microphones M1, Ma, Ms, Ma, and each microphone M direct, M2, M3,
Transfer coefficients H+ (ω) and H are placed after M4, respectively.
2(ω).

H3(ω), H4(ω) A, D, F6+, 6L
6s and 64 in parallel in multiple stages, and furthermore, it is preferable to provide a filter coefficient control section that can control the transmission coefficient so that the audio deterioration is at a level below a certain tolerance value and the noise component is minimized. is taught.

[Problems to be Solved by the Invention] However, the voice load drive device disclosed in the specification of Japanese Patent Publication No. 63-29755 reduces the volume level of in-vehicle equipment every time a voice command is given, and It is necessary to close the opening/closing device in the part where the noise enters, or to provide an input delay means to delay the input of the voice command until the volume level of the onboard equipment and the noise outside the vehicle has decreased.

Furthermore, according to an improved adaptive noise-cancelling microphone array reported in the Technical Outlook of the Journal of the Institute of Electronics and Communication Engineers, it was observed that a directional pattern with reduced sensitivity was realized in the direction of noise arrival. However, in order to operate this device, the direction from which the noise is coming must be confirmed in advance.

However, in a narrow space such as the cabin of a car, where the number of people increases and decreases as people get on and off the vehicle, and within the sound field where the noise conditions change, the optimal placement position of the microphone also changes. Therefore, even if the sound field conditions change, it is difficult to sufficiently cancel noise using the conventional adaptive noise canceling microphone array in which the microphones are installed at fixed positions.

This invention solves the problems of conventional adaptive noise canceling microphone arrays in which the microphone installation position is fixed regardless of changes in sound field conditions, and the incompleteness and complicated noise cancellation of audio load drive devices that respond to changes in sound field. This was done in order to eliminate the drawback of requiring an operating procedure, and even if the noise source in the sound field changes, the noise can be canceled sufficiently, and the voice command word is input to the voice recognition device with a high S/N. It is an object of the present invention to provide a voice input method and device that can perform voice input.

[Means for Solving the Problems] The present invention has been made in view of the above-mentioned circumstances, and the voice input method includes a voice command word from which noise components have been removed according to the procedure shown in FIG. This is input to a speech recognition device.

That is, one main microphone and a plurality of reference microphones are distributed in a sound field, and then the degree of cross-correlation between the noise component of the distributed main microphone and the noise component of the plurality of reference microphones is sequentially calculated. Thereafter, the output of the reference microphone determined to have the highest degree of cross-correlation with the noise component of the main microphone by the above calculation is used as the reference input, the output of the main microphone is used as the main input, and the noise components of these two inputs cancel each other out. The reference input side transfer characteristic is controlled in this way, and the added value with the noise component canceled is input to the speech recognition device.

Further, one of the audio input devices according to the present invention has the configuration shown in FIG. In the apparatus, the cross-correlation determination unit 20 includes a main microphone (1) 1
and a plurality of distributed reference microphones 2, and an interrelationship that calculates the degree of cross-correlation between the noise component of the main microphone 1 and the noise component of the reference microphone 2, which are arranged on the downstream side of the main microphone 1 and the reference microphone 2. a correlation determiner 3; a changeover switch means 4 for selectively connecting the output paths of the plurality of reference microphones 2 to the input paths of the cross-correlation determiner 3; and a degree of cross-correlation determined by the cross-correlation determiner 3. The adaptive noise canceling unit 30 is configured to switch between the main microphone 1 and the reference microphone 2 to connect the output path of the reference microphone 2 with the highest output to the reference microphone 2 input path of the adaptive noise canceling unit 30. A, D, F6 having an input path and provided in the input path of the reference microphone 2; an adder 7 provided between the downstream side of the A, D, F6 and the input path of the main microphone; a noise level detector 8 arranged after the adder 7; a filter coefficient control section 9 arranged between the latter stage of the adder 7 and the A, D, and F6; and noise in the noise level detector 8. Changeover switch means 1 for switching and connecting the input path of the main microphone l to the input path of the speech recognition device 40 in accordance with the output of the cancel signal.
It consists of 0.

Furthermore, as shown in FIG. 3, another audio input device is an audio input device consisting of a cross-correlation determining unit 20 and an adaptive noise canceling unit 30 connected after this cross-correlation determining unit 20. The cross-correlation unit includes 11 main microphones and a plurality of distributed reference microphones 2. The cross-correlation unit calculates the degree of cross-correlation between the noise component of the main microphone L and the noise component of the reference microphone 2, which are arranged on the downstream side of the plurality of distributed reference microphones 2. a changeover switch means 4 for selectively connecting the outputs of the plurality of reference microphones 1 to the input paths of the cross-correlation determiner 3; means for storing a reference microphone number; and inputting the reference microphone number saved in the storing means into a noise level detector of the adaptive noise cancellation unit, and determining the cross-correlation degree according to a response signal in the noise level detector. a changeover switch means for selectively connecting the output path of the reference microphone with the highest number to the reference microphone input path of the adaptive noise cancellation unit, the adaptive noise cancellation unit having two input paths of a main microphone 1 and a reference microphone 2; , an adaptive filter 6 disposed on the input path of the reference microphone 2, an adder 7 disposed between the downstream side of the adaptive filter 6 and the input path of the main microphone 1, and a downstream side of the adder 7. a noise level detector 8 disposed; a filter coefficient control section 9 disposed between the downstream stage of the adder 7 and the adaptive filter 6; and a zero signal of the noise cancellation signal output detected by the noise level detector 8. The configuration includes a changeover switch means 10 for connecting the output path of the added value to the input path of the speech recognition device 40 in accordance with the above.

[Embodiments] Hereinafter, embodiments of the voice input method and device according to the present invention will be described. However, since the voice input device is configured to implement the voice input method according to the present invention,
The example of the voice input device will be replaced with the example of the voice input method.

FIG. 4 is a perspective view of a main part showing the structure inside a passenger car in which the voice input device according to the present invention is installed.

In Fig. 4, l is the main microphone for inputting the speaker's voice command, 2 is the reference microphone placed at the location where the noise is generated inside the vehicle or the location where outside noise flows in, and 41 is the monitor speaker connected to the voice recognition device. , 51 is a navigation monitor television connected to the navigation unit.

The system configuration of the voice input device for a passenger car in this embodiment is as follows:
As shown in FIG. 5, it consists of a cross-correlation determining unit 20 and an adaptive noise canceling unit 30 disposed after this unit 20. a cross-correlation determiner 3 that calculates and determines the degree of cross-correlation of the reference microphones 2.
2, . . . 2 to the input path of the cross-correlation determiner 3, and the output signal from the noise level detector 8 of the adaptive noise canceling unit 30 described later. a changeover switch 5 that switches between and connects the output path of the reference microphone 2 with a high noise level and the reference input path of the adaptive noise 3o; and an output path that connects the output from the cross-correlation determiner to the noise level detector 8 of the adaptive cancellation unit 30; Main microphone 1 and cross-correlation determiner 3
bandpass filters (hereinafter referred to as “B, P, F
(abbreviated as J)1,2. Analog-digital converter (hereinafter abbreviated as rA/DJ) la, 2a; and delay circuit 1c,
2c; RAM 1 is used as a means for storing the reference microphone number connected to the cross-correlation determiner 3 and having a high degree of cross-correlation.
d is provided.

Further, the adaptive noise canceling unit 30 has two inputs that receive outputs from the main microphone 1 and the reference microphone 2, and has connection paths A, D, F1a, and A, DF2d that receive the output of the reference microphone (reference input).
An adder 7 provided between the rear stage side and the input path that receives the output of the main microphone 1, a noise level detector 8 disposed after the adder, and A, D, F a filter coefficient control section 9 provided between the filter 2d and a changeover switch lO that switches and connects the output path of the adder 8 to the input path of the speech recognition device 40 in accordance with the noise canceling signal output of the noise level detector 8; The speech recognition device 40 is a recognition unit for limited languages and unspecified speakers. The second voice recognition device 40 includes, for example, the navigation map ``Kafudai'', ..., ``Kaiten'', etc.
Voice recognition such as Hidari j is performed, and the next stage navigation unit (not shown) is driven while hand-joking with the speaker (driver) using the monitor speaker 41.

When inputting a voice command word to the voice recognition device 40 using the voice input device configured as described above, when the input switch of the power source (not shown) is turned on, the cross-correlation determination unit side executes the flowchart shown in FIG. As shown in , the device is initialized (Step S1), and the reference microphone number immediately before the changeover switch 10 was turned off in the previous operation is loaded from the RAM 1d into the cross-correlation determiner 3 (Step S2). ), this reference microphone number is used in the adaptive noise cancellation unit 3 after the cross-correlation determiner unit.
0 (step S3), and is also input to the changeover switch 4 of the cross-correlation determining unit 20, setting the switch (step S4).

As a result, the microphone output path of the reference microphone number is connected to be input to the cross-correlation determiner 3, and a signal is sent from the noise level detector 8 of the adaptive noise canceling unit 30 to the changeover switch 5, and the changeover switch 5 connects the output path of the reference microphone with that number and the reference input path on the adaptive noise cancellation unit side, and noise cancellation is performed between the input of the main input path of the adaptive noise cancellation unit. Then, while the monitor sound is being output from the speech recognition device 40, the cross-correlation calculation is stopped so as not to misidentify it as a noise source (step S5).

If the monitor sound is not output, the cross-correlation degree is calculated by the cross-correlation determiner 3 (step S6), and after the calculation, check again that the monitor sound is not output.
Step 7), and save the result to RAM together with the microphone number at that time (Step S8).

Next, a repetition counter is set (step S9).The repetition counter indicates how many times the correlation calculation of the reference microphone is performed before the cross-correlation degree data is reviewed. When reviewing n
, 2n is set when reviewing every two rounds.

Next, the reference microphone number is increased by 1 (SIO step), and when the microphone number becomes n or more, the reference microphone number is cleared, and the process returns to the string and repeats the above process. If the number is not greater than or equal to n (step S11), the selector switch 4 is set (step S12), and it is determined whether or not the monitor audio is being output (step S11).
13 steps).

If the determination result is rNOJ, the cross-correlation determiner 3 determines whether the human input from the main microphone and 1. : The degree of cross-correlation of the reference microphone number is calculated (Step S14), and it is determined again whether the monitor audio is being output (Step S15). If the determination result in Step S15 is rNOJ, The reference microphone number and the cross-correlation degree determination result are saved in the RAM (step S16), and it is determined whether the level upL is higher than the previous cross-correlation degree (step S17).

If the judgment result in step S17 is rYEsJ, that microphone number is output to the adaptive noise cancellation unit (step S l 8), and a repetition counter is set (step S l 9). to the RAM (S20 step) and repeat the process from the StO step onwards.

Also, if the judgment result in step S17 is "NO", the repetition counter is incremented by one (step S21).
, it is determined whether the repetition counter is correct (S2
2 steps).

If the number of calculation repetitions is less than a certain value, the processes after the SIO step are repeated.

When the number of calculation repetitions is exceeded, the correlation data with the highest cross-correlation value within the number of repetitions and the microphone number at that time are loaded (step S23), and the processes from step S18 onward are repeated. The reason for this is that if the highest value happens to occur after the ignition is turned on, the reference microphone will not be switched (by switch 5) unless further cross-correlation data is obtained. If the maximum value momentarily appears, the microphone change (switch 5) will not be turned on even if the conditions inside the vehicle change and the overall cross-correlation value (or that of the reference microphone) decreases. .

On the other hand, in the adaptive noise cancellation unit 30, the power supply (
When the switch (not shown) is turned on, the device is initialized (Tl
Step), the selector switch 10 is turned off to prevent noise from appearing after the power is turned on. Then, by the notification of the reference microphone number from the cross-correlation determiner 3 of the cross-correlation determination unit 20 described above (step S3), the optimum reference microphone number is loaded to the noise level detector 8 (step T3), and the optimum reference microphone number is loaded to the noise level detector 8 (step T3). The changeover switch 4 of the cross-correlation determination unit is set to connect to the road microphone number (step T4).

Then, an interruption of the adaptive noise canceling operation of the adaptive noise canceling unit 30 is permitted (step T5).
. Also, the noise level detector 8 waits until ε of the output from the adder 7 converges to a certain level (step T6).
.

When the noise has converged to a certain level, the changeover switch 10 of the adaptive noise cancel/cell unit 30 is turned on, and the noise-cancelled audio signal is output to the subsequent speech recognition device 40 (step T7).

Next, the optimum reference microphone number (318 steps) is loaded from the cross-correlation determining unit 20 (step T8). If there is no change to this optimal reference microphone number, repeat step T8 (step T9), and wait until voice input is finished (step TlO).
, If the convergence level hardly changes during this time, it is determined that there is no voice input.

After the above step, in order to set a new microphone position, the changeover switch 10 is turned off (rT11 step), and after the changeover switch 5 is further set (T12 step), the steps from T6 onwards are repeated.

Interrupt processing is the processing itself based on the fixed learning method, and the seventh
Processing is performed according to the flowchart shown in FIG.

The voice input device of this embodiment has been described above using the configuration of the second aspect of the present invention as a representative example. In the first invention, the cross-correlation determiner 3 is configured to directly send a signal to the changeover switch 5, whereas in the device of this embodiment, the noise level detector 8 of the adaptive noise canceling unit 30 The difference is that the changeover switch 5 is changed over by a signal sent from a noise level detector.

Further, in the audio input device of the present invention, the adaptive noise canceling unit 20 is connected to a portion that performs an adaptive noise canceling operation, that is, an adaptive filter 6, an adder 7, a filter coefficient control section 9, and a level of the output (ε). In order to reduce the operation of determining and switching the switch 5 and the unnecessary load on the speech recognition device 40, the reference microphone is switched to a switching output (ε
Both inventions have in common that the main part is a noise level detector that operates to inhibit data output while the noise level has converged to a certain level.

In addition, although a case has been described in which A, D, and F are used as the adaptive filters 6 of the adaptive noise canceling unit 30 of the audio input device of this embodiment, the A, D, and F filters minimize the output squared error. , it functions to update the filter coefficients.

In addition, adaptive noise and noise source 1 of the voice input device of this embodiment
3 and the sounds coming from the main microphone 1 and the reference microphone 2, respectively, are added as two inputs (main input and reference input) by an adder 7.

At this time, the adaptive filter 6 is arranged in the passage path of the reference input, and the filter coefficients of the adaptive filter 6 are controlled so that the noise component of the output added by the adder 7 becomes zero. Unlike an adaptive noise canceller configured with a microphone array, the principle is not to form a directivity pattern with valleys in the direction of noise arrival reaching the main microphone 1 and the reference microphone 2, so the direction of noise arrival is learned in advance. There's no need to.

[Effects of the Invention] As is clear from the above description, the audio input method of the present invention distributes one main microphone and a plurality of reference microphones in a sound field, and then selects a microphone that has the highest cross-correlation with the main input. Since the output of the reference microphone is selected and the noise of these two inputs is canceled, there is no need to re-learn the direction of incidence of the noise arriving at the microphone every time the 5 4 H sound source moves.

Since the mouth and input to the voice input device are input with noise components canceled, there is no need to provide a special low S/N device like a conventional voice recognition device.

C. Among the outputs from the reference microphones distributed at multiple locations in the sound field, the output of the reference microphone with the highest degree of cross-correlation with the main input is selected to cancel the two-person noise, so the sound field in the noisy environment Under the circumstances, the highest S
/N voice command words can be input.

2. Input high S/N voice commands to the voice recognition device by quickly responding to changes in the spatial state within the sound field (for example, changes in the number of passengers due to boarding and alighting, movement of the noise source, and changes in the noise state) Furthermore, the audio input device according to the present invention does not have a structure in which the inputs from a plurality of microphones are passed through respective adaptive filters to cancel noise, but the inputs from the main microphone and the plurality of microphones are Since the configuration is provided with a cross-correlation determination unit that selects the reference input and main input with the highest cross-correlation degree of the noise components of the input of the reference microphone, the noise Even if the source moves,
There is no need to re-learn the direction of noise arrival.

Further, unlike conventional voice input devices, there is no need to provide special low S/N means for the voice recognition device.

Furthermore, even if the noise environment changes, we can respond to the change.
A high S/H voice command word can be quickly input to the voice recognition device in accordance with the noise conditions of the sound field.

Furthermore, the second audio input device of the present invention temporarily saves the number of the reference microphone having the highest cross-correlation degree determined by the cross-correlation determiner in the storage means, and The number is input to the noise level detector of the adaptive noise canceling unit, and depending on the response signal in the noise level detector, the output path of the reference microphone with the number having a high degree of cross-correlation is input to the reference microphone input path of the adaptive noise canceling unit. Since the structure is such that the device can be switched and connected to the
The reference microphone with the highest cross-correlation degree when used last time can be used. Therefore, convergence on the adaptive noise cancellation unit side is performed quickly. Therefore, there is an advantage that the device can be reused immediately.

[Brief explanation of the drawing]

FIG. 1 is a process diagram showing the basic configuration of the voice input method according to the present invention, FIG. 2 is a block diagram showing the basic configuration of the first voice input device of the present invention, and FIG. FIG. 4 is a block diagram showing the basic configuration of the voice input device. FIG. 4 is a perspective view of the main parts showing the structure inside the cabin of a passenger car in which the voice input device of the embodiment is installed. FIG. 5 is a voice input device of an embodiment of the present invention. A schematic block diagram of the configuration of the device, FIG. 6 is a flowchart of the cross-correlation determination unit when operating the voice input device of FIG. 5, and FIGS. 7(a) and (b) are adaptations of the voice input device of FIG. 5. FIG. 8 is an explanatory diagram of the principle of two-person noise cancellation in the audio input device of the present invention; FIG. 9 is an explanatory diagram of the principle of noise cancellation of a conventional two-element microphone array; The figure is an explanatory diagram of the principle of noise cancellation using a conventional four-element microphone array. In Figures 1 to 10, 1...Main microphone, 2...Reference microphone, 3...Cross correlation judger, 4.5.10... Changeover switch , 6...Adaptive filter, 7...Adder, 8...Noise level detector, 9......
- Filter coefficient control section, 20... Cross correlation determination unit, 30... Adaptive noise cancellation unit, 40... Voice recognition device.

Claims (3)

[Claims] (1) A. One main microphone and multiple reference microphones are distributed in a sound field, and B. Cross-correlation between the noise components of the output of the distributed main microphone and the noise components of the output of the multiple reference microphones. C. After that, the output of the reference microphone determined to have the highest degree of cross-correlation with the noise component of the main microphone by the above calculation is used as the reference input, and the output of the main microphone is used as the main input. Add these two inputs, and 2. If the added value of the two inputs is zero, input the added value as is to the speech recognition device, and if it is not zero, use the reference so that the added value of the two inputs becomes zero. A voice input method characterized by controlling the transfer characteristic of the input and inputting the added value to a speech recognition device when the added value reaches zero. (2) An audio input device consisting of a cross-correlation determining unit and an adaptive noise canceling unit connected after the cross-correlation determining unit, wherein the cross-correlation determining unit includes one main microphone and a plurality of distributed reference microphones. and a cross-correlation determiner that calculates the degree of cross-correlation between the noise component of the main microphone and the noise component of the reference microphone, which are disposed downstream of the main microphone and the reference microphone, and output paths of the plurality of reference microphones, respectively. , a changeover switch means for switching and connecting to an input path of a cross-correlation determiner; and switching an output path of a reference microphone having the highest degree of cross-correlation determined by the cross-correlation determiner to a reference microphone input path of the adaptive noise cancellation unit. the adaptive noise canceling unit has two input paths for receiving the outputs of the main microphone and the reference microphone, an adaptive filter provided in the input path of the reference microphone, and a downstream side of the adaptive filter. an adder disposed between the adder and the input path of the main microphone; a noise level detector disposed after the adder; and a filter coefficient control disposed between the adder and the adaptive filter. and changeover switch means for selectively connecting an adder output path to an input path of a speech recognition device in accordance with a noise canceling signal output from the noise level detector. (3) An audio input device consisting of a cross-correlation determination unit and an adaptive noise cancellation unit connected to the rear stage of the cross-correlation determination unit, the cross-correlation determination unit comprising one main microphone and a plurality of distributed microphones. a cross-correlation determiner that calculates the degree of cross-correlation between the noise components of a main microphone and the noise components of the reference microphone disposed downstream of the reference microphone; changeover switch means for switching connection to the input path; means for storing the number of the reference microphone having the highest degree of cross-correlation determined by the cross-correlation determiner; and applying the saved reference microphone number to the storing means. Switching that inputs the input to the noise level detector of the noise canceling unit, and switches and connects the output path of the reference microphone with the number with a high degree of cross-correlation to the reference microphone input path of the adaptive noise canceling unit according to the response signal in the noise level detector. The adaptive noise canceling unit has two input paths for receiving the outputs of the main microphone and the reference microphone, and includes an adaptive filter disposed in the input path of the reference microphone, a downstream side of the adaptive filter, and the adaptive noise canceling unit. an adder disposed between the input path of the main microphone; a noise level detector disposed after the adder; and a filter coefficient control section disposed between the adder and the adaptive filter; . A device characterized in that it has a changeover switch means for selectively connecting the adder output to an input path of a speech recognition device in response to a zero signal of the noise canceling signal output detected by the noise level detector.