JPH02244098A - Voice signal processor - Google Patents

Abstract

PURPOSE:To extract voice command words at high S/N from the noises generated by an appliance by providing the processor with an adaptive filter and subtracting the output signal of a band-pass filter connected to a 2nd microphone from the output signal of a band-pass filter connected to a 1st microphone. CONSTITUTION:The microphone 13 which detects the speaker's voice command words and the microphone 19 which detects the noises propagating in a prescribed space are provided. The band-pass filters 211, 221 are respectively connected thereto. A noise cancelation unit 23 having the adaptive filter 232 is connected to the band-pass filters 211, 221 and the output signal of the band-pass filter 221 is adaptively subtracted from the output band-pass filter of the band- pass filter 211. Even if, therefore, the noises are mixed in the electric signal converted by the microphone 13, the noises are converted by the microphone 19 and are offset and erased by the signal past the adaptive filter 232. The voice command words are extracted at the high S/N in this way from the noises generated by the appliance without degrading the functions of the appliance.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Field of Industrial Application) The present invention relates to a voice signal processing device that can extract voice command words with a high S/N ratio even in a noisy environment.

(Prior Art) As a recent trend, various devices such as radios, car stereos, air conditioners, etc. have been mounted on moving objects such as passenger cars. Conventionally, these devices have been operated by changing switches, but recently attempts have been made to operate the devices by voice commands.

By the way, some of these devices can generate noise that includes the same frequency band as the audio frequency. Therefore,
When attempting to operate a device using voice commands, a problem arises in that the noise generated by the device overlaps with the voice command word, making accurate voice recognition impossible.

In order to solve these problems, for example,
Publication No. 29755 etc. states that when performing voice recognition,
A device has been proposed that causes all devices that adversely affect speech recognition to perform an operation to lower the noise level.

(Problems to be Solved by the Invention) However, in such a conventional device, there is a problem in that the function of the equipment that generates the noise deteriorates, and the speaker sometimes feels uncomfortable.

The present invention has been made to solve these problems, and is capable of extracting voice command words from the noise generated by equipment at a high S/N ratio without reducing the functionality of the equipment. Consider it a technical issue.

[Structure of the invention]

(Means for solving the problem) The technical means taken to achieve the above-mentioned technical problem are a first microphone that detects the speaker's voice command, and a first microphone that detects noise propagating in a predetermined space. a second microphone is provided, first and second bandpass filters are respectively connected to both microphones, and a noise canceling unit having an adaptive filter is connected to the first and second bandpass filters. , the output signal of the second band-pass filter is adaptively subtracted from the output signal of the first band-pass filter.

(Operation) According to the above-mentioned technical means, the noise canceling means adaptively subtracts the electrical signal converted by the second microphone from the electrical signal converted by the first microphone. Become. Therefore, even if noise is mixed in the electrical signal converted by the first microphone, this noise is canceled out by the signal converted by the second microphone and passed through the adaptive filter. be done. As a result, only the voice command word is extracted from the electrical signal converted by the first microphone.

Furthermore, according to the above-mentioned technical means, a bandpass filter is connected between the first and second microphones and the noise canceling unit. Therefore, electrical signals unrelated to the extraction of the voice command are removed by the bandpass filter, and only electrical signals relevant to the extraction of the voice command are input to the noise cancellation unit. Hence,
Voice command words can be extracted more reliably.

(Example) Hereinafter, a preferred first example of the present invention will be described with reference to the accompanying drawings.

FIG. 1 is a perspective view depicting the inside of a passenger compartment 16 of a passenger car 10, and shows an example in which a voice input device 15 is used to operate a navigation device 14 of the passenger car 10.

The main microphone 13 is fixed near the space between the driver's seat 11 and the passenger's seat 12 so that the voices of the driver's seat occupant and the passenger's seat occupant can easily be heard. Further, the reference microphone 19 is fixed to the back side of the rear seat 20 so as to prevent the voices of the driver's seat occupant and the front passenger seat occupant from entering as much as possible. Main microphone 13
The reference microphone 19 is fixed to the center of the front console 9 and the center of the rear panel 8, respectively, so that the noise inside the vehicle interior 16 is captured at approximately the same level.

The voice input device 15 is housed under the foot of the passenger seat 12 so as not to be seen. The voice input device 15 is connected to the navigation device 14 . Furthermore, two speakers 17R117L are installed on the rear panel 8. The speakers 17R and 117L are connected to the in-vehicle audio equipment 18, and music, radio broadcasts, etc. are played from the speakers 17R and 17L. Music, radio broadcasting, etc. flowing from the speakers 17R and 17L become noise including the audio frequency band and interfere with the operation of the voice recognition unit 24.

Further, when the engine (not shown) of the passenger car 10 starts, engine noise having a lower frequency than the voice command is generated from the engine. This engine noise also becomes noise including the audio frequency band and interferes with the operation of the voice recognition unit 24. In addition, when the passenger car 10 runs,
Vehicle running noise, wind noise, etc., which have a lower frequency than the voice command words, occur. These running noises, wind noises, and the like also become noises that include the audio frequency band and interfere with the operation of the voice recognition unit 24.

In addition to this, various other noises enter the cabin 16 from outside the passenger car, such as road construction noise and ambulance sirens. These noises from outside the passenger car 10 also become noises that include the audio frequency band and interfere with the operation of the voice recognition unit 24.

In order to eliminate the influence of such noise and perform accurate voice recognition, the voice input device 15 is equipped with a noise canceling unit 22 to which the present invention is applied.

The operating principle of the voice input device 15 will be explained below with reference to FIG.

The sounds coming from the audio source 31 and the noise source 32 are converted into electrical signals by the main microphone 13 and the reference microphone 19. The two converted electrical signals (main signal and reference signal) are applied to adder 233. Adder 2
At 33, the reference signal is subtracted from the main signal.

Voice commands for the driver's seat occupant or passenger's seat occupant are in cabin 1.
6 is converted into an electrical signal by the main microphone 13 and the reference microphone 19. At this time, the noise inside the vehicle compartment 16 is captured by the main microphone 13 and the reference microphone 19 at approximately the same level. However, since the main microphone 13 is located closer to the driver's seat occupant or the passenger's seat occupant than the reference microphone 19, the main microphone 13 receives the occupant's voice commands at a higher level than the reference microphone 19. It is captured by

Therefore, in principle, when the reference signal is subtracted from the main signal, the noise within the passenger compartment 16 is canceled out and is completely eliminated.

However, in reality, even if the reference signal is subtracted from the main signal, only a small amount of noise within the vehicle interior 16 is eliminated. This is because there is a time delay between the electrical signal converted by the main microphone 13 and the electrical signal converted by the reference microphone 19, and the influence of reflected waves generated when sound waves are reflected within the vehicle interior 16. This is because the noise within the room 16 is only slightly offset.

Therefore, in the device of this embodiment, the adaptive filter 232 is arranged in the passage of the reference input, and the filter coefficients of the adaptive filter 232 are controlled so that the noise component of the output added by the adder 233 becomes zero.

Therefore, the time delay between the electrical signal converted by the main microphone 13 and the electrical signal converted by the reference microphone 19, and the influence of reflected waves generated when sound waves are reflected within the vehicle interior 16 are eliminated. The noise inside the vehicle compartment 16 is completely eliminated.

The electric circuit of the voice input device 15 will be described below with reference to FIG. The voice input device 15 includes a first voice input circuit 21 , a second voice input circuit 22 , a noise cancellation unit 23 , a voice recognition unit 24 , and a monitor speaker 25 .

The first audio input circuit 21 further includes a bandpass filter 2.
11, an amplifier 212, and an A/D converter 213. The main microphone 13 is connected to the first audio input circuit 21 . The electrical signal converted by the microphone 13 is input to a bandpass filter 211 . The bandpass filter 211 extracts only the frequency band that includes the voice command word from the input electrical signal. The electrical signal that has passed through the bandpass filter 211 is amplified by an amplifier 212 and then input to an A/D converter 213 where it is quantized.

(Similarly, the second audio input circuit 22 further includes a bandpass filter 221, an amplifier 222, an A/D converter 2
23. A reference microphone 19 is connected to the second audio input circuit 22 . The electrical signal converted by reference microphone 19 is input to bandpass filter 221 . The bandpass filter 221 has exactly the same frequency characteristics as the bandpass filter 211 of the first audio input circuit 21. The electrical signal that has passed through the bandpass filter 221 is amplified by the amplifier 222, and then the A/D
The signal is input to converter 223 and quantized.

By the way, the A/D converters 213 and 223 may generate electrical noise called "folding waves" in the process of quantizing analog signals. A/D converter 213.
223 generates a "return wave", the noise canceling unit 23, which will be described later, will not operate correctly. This "folding wave" occurs when an electrical signal having a high frequency equal to or higher than the sampling frequency A is included in the analog signal. Therefore, in the device of this embodiment, bandpass filters 211 and 221 are used to remove electrical signals having a frequency higher than the frequency band of the voice command word, and the A/D
The converters 213 and 223 are designed to prevent "return waves" from occurring.

Apart from that, noises such as the engine sound and running sound of the passenger car 10 have a lower frequency than the voice command words. Therefore, in the device of this embodiment, these noises having low frequencies are removed by the band pass filters 211 and 221 without being sent to the noise canceling unit 23.

Therefore, the amount of signal processing by adaptive filter 232 is reduced.

The binary codes quantized by the A/D converters 213 and 223 are sent to the noise cancellation unit 23. The noise cancellation unit 23 includes a delay device 231
, an adaptive filter 232 , and an adder 233 .

The delay device 231 temporally delays the binary code output from the first audio input circuit 21 and ensures the causality of the adaptive filter 232.

The adaptive filter 232 monitors the output signal of the adder 233 and adjusts the phase, time delay, amplitude direction, magnitude, etc. of the output signal from the second audio input circuit 22. As a result of this adjustment, it is estimated that there is a time delay between the electrical signal converted by the main microphone 13 and the electrical signal converted by the reference microphone 19, and the influence of reflected waves generated when sound waves are reflected inside the vehicle interior 16. The output signal of the second audio input circuit 22 is corrected to be as similar to the noise signal contained in the output signal of the first audio input circuit 21 as possible. The adaptive filter 232 used in the device of this embodiment is as follows:
A learning identification method is used to change the filter characteristics so that the root mean square error of the output is minimized.

Adder 233 subtracts the output signal of adaptive filter 232 from the output signal of delay device 231. In the adder 233, when the output signal of the adaptive filter 232 is subtracted from the output signal of the delay device 231, the noise in the passenger compartment 16 is canceled out.
will be deleted.

The output signal of adder 233 is applied to speech recognition unit 24 . The speech recognition unit 24 is a device capable of recognizing limited languages of unspecified speakers.

The voice recognition unit 24 has pre-registered voice patterns such as "on", "off", and "douzachu".
The next stage navigation unit 14 is controlled while handshaking with the speaker using the monitor speaker 25.

If the noise inside the vehicle interior 16 and the voice command word are applied to the voice recognition unit 24 at the same time, it may become impossible to recognize the voice command word. However, in the device of this embodiment, the noise inside the vehicle compartment 16 is removed by the noise canceling unit 23.
, and only the voice command word is applied to the voice recognition unit 24. Therefore, the spoken command word can be recognized more reliably.

Note that in the embodiment device described above, the main microphone 1
3 is placed near the driver's seat, the voice command words generated by the speaker are mainly input to the first voice input circuit 21. Further, since the reference microphone 19 is arranged behind the rear seat 20, almost no voice command words are input to the second voice input circuit 22. Therefore, it is very unlikely that the voice command word will be erased or weakened by the signal processing of the digital signal processing circuit 26.

Furthermore, since the device of this embodiment eliminates noise using the adaptive filter 232, the noise generated by the noise source can be eliminated regardless of the directivity of the microphone 13, 19 or the direction of arrival of the noise.

〔Effect of the invention〕

According to the present invention, a voice command word can be extracted from noise generated by a device at a high S/N ratio without degrading the function of the device.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a perspective view of a vehicle compartment to which a preferred embodiment of the present invention is applied. FIG. 2 is a block diagram illustrating the noise canceling principle of a preferred embodiment of the present invention. FIG. 3 is a block diagram depicting the electrical configuration of a first preferred embodiment of the present invention. 10... Passenger car, 11... Driver's seat, 12... Passenger seat. 13...Main microphone (first microphone). 14...Navigation device, 15...Voice input device. 16... Vehicle interior (predetermined space). 17R, 17L...Speaker. 18...Car audio equipment. 19...Reference microphone (second microphone). 20... Rear seat, 21... First audio input circuit. 22...Second audio input circuit. 23... Noise cancellation unit (noise cancellation unit) 9 24... Voice recognition unit. 25...Monitor speaker. 211... Bandpass filter (first bandpass filter). 221... Bandpass filter (second bandpass filter).

Claims (1)

[Claims] (1) a first microphone that is arranged in a predetermined space and converts a speaker's voice command into an electrical signal; a second microphone that converts noise propagating in the predetermined space into an electrical signal; a first bandpass filter electrically connected to the first microphone; a second bandpass filter electrically connected to the second microphone; and the first and second bandpass filters. and a noise canceling unit electrically connected to the noise canceling unit that includes an adaptive filter and adaptively subtracts the output signal of the second bandpass filter from the output signal of the first bandpass filter.