JP3936819B2 - Microphone system - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a microphone system, and more particularly to a microphone system that outputs a speaker voice signal with an improved SN ratio by performing adaptive signal processing using output signals of two microphones.
[0002]
[Prior art]
The current speech recognition system has reached a technical level that can achieve a recognition rate of about 95% when a signal-to-noise ratio (S: voice / N: noise) of 15 dB or more is secured. However, when the S / N ratio is reduced due to noise present in the surrounding area, the recognition rate is rapidly lowered. Fig. 8 shows the relationship between SN ratio and recognition performance for several types of microphones (omnidirectional, unidirectional, narrow directivity, AMNOR (Adaptive Microphone-array for Noise Reduction)). Ratios and recognition rates are generally contained within a band showing the S-characteristic 100. As is apparent from FIG. 8, the recognition rate rapidly decreases due to a decrease in the SN ratio, and decreases to about 50% in an environment where the SN ratio is 0 dB.
[0003]
For this reason, the above-mentioned deterioration in recognition performance is unavoidable in automobile interiors where noise generated by automobiles (engine noise, road noise, pattern noise, wind noise, etc.) is present, and a voice recognition system is mounted on the vehicle. It is one of the big problems.
In view of the circumstances described above, various methods have been proposed for receiving sound with a high S / N ratio while reducing the influence of surrounding noise, and a high S / N ratio using a plurality of microphones and digital signal processing has been proposed. A sound receiving system is an example. The simplest configuration of such a high S / N ratio receiving system is a system using two microphones as shown in FIG. 9, but other advanced systems such as Griffith-Jim type array and AMNOR are also used. A system has been proposed.
[0004]
9, 1 and 2 first, second microphone, 3 is an adaptive signal processing unit, the output signal x ₂ microphones 2 with the error signal e is input is input as a reference signal, the error signal e Adaptive signal processing is performed based on the LMS (Least Mean Square) algorithm so that the power of the signal becomes minimum. In the adaptive signal processing unit 3, 3a is an LMS calculation unit, and 3b is an adaptive filter having, for example, an FIR type digital filter configuration. The LMS calculation unit 3a determines the coefficient of the adaptive filter 3b so that the power of the error signal e is minimized by adaptive signal processing.
[0005]
Reference numeral 4 denotes a target response setting unit that receives a noise signal output from the microphone 1 as a target signal, and is used to accurately approximate the inverse characteristics of the acoustic system. When the signal delay time that is half the tap length of the adaptive filter 3b is d, the target response setting unit 4 has a delay characteristic of the time d and has a flat characteristic (gain 1 characteristic) in the audio frequency band. That is, the target response setting unit 4 has a flat frequency characteristic with a gain of 1 as shown in FIG. 10A, and has an impulse response characteristic with a delay time d as shown in FIG. 10B. . This target response setting unit 4 can be realized by setting the coefficient corresponding to the delay time d of the FIR type digital filter to 1 and setting other coefficients to 0.
A subtracting unit 5 subtracts the output signal of the adaptive filter 3b from the target response output from the target response setting unit 4, and outputs an error signal e.
[0006]
At the time of non-speech recognition, only noise is input to the microphones 1 and 2, and the adaptive signal processing unit 3 determines the filter coefficient W by adaptive signal processing so that the error signal e, that is, the power of the noise output is minimized. On the other hand, at the time of speech recognition, the adaptive signal processing unit 3 does not update the filter coefficient, sets the filter coefficient W determined at the time of non-speech recognition to the adaptive filter 3b, and outputs a speech signal.
The ideal performance originally required for the system shown in FIG. 9 is to output only the speech signal Xs (z) (noise output is 0) as an output signal during speech recognition. That is, regarding the noise output En (z),
En (z) ＝ Xn ₁ (z) z ^-d −Xn ₂ (z) W (z) (1)
It is to determine a parameter W (coefficient of the adaptive filter 3b) W that can be adjusted so that the average of {En (z)} ² takes the minimum value. However, Xn ₁ (z) and Xn ₂ (z) are noise signals included in the output signals of the microphones 1 and 2.
[0007]
[Problems to be solved by the invention]
Since there are many noise sources in automobiles, the coherence of automobile interior noise picked up by the microphones 1 and 2 tends to decrease as the microphones 1 and 2 are moved away. For this reason, as the two microphones 1 and 2 are moved away from each other, there is a problem that the condition of the expression (1) becomes difficult to be satisfied, and the microphones 1 and 2 need to be arranged as close as possible.
However, if the two microphones 1 and 2 are arranged as close as possible, there is a high possibility that almost the same sound and noise will be incident on the two microphones, and the adaptive filter coefficient W is determined so as to satisfy equation (1). If the noise is erased, even the sound is erased. On the other hand, if the adaptive filter coefficient W is determined so that the voice is not distorted, the noise hardly disappears and the SN ratio is hardly improved.
[0008]
Such a problem is not unique to the system shown in FIG. 9, but occurs in a similar manner even when a more advanced high S / N ratio receiving system such as a Griffith-Jim type array or AMNOR is adopted.
From the above, an object of the present invention is to improve the S / N ratio of an audio signal in the microphone system of FIG. 9 using two microphones.
[0009]
[Means for Solving the Problems]
According to the present invention, the above-described problem is achieved by the first and second microphones, the target signal output unit that receives the signal output from the first microphone and outputs the target signal, and the output from the second microphone. An adaptive signal processing unit for performing adaptive signal processing by inputting the received signal as a reference signal, and an error signal generating unit for outputting a difference between the output signal of the adaptive filter constituting the adaptive signal processing unit and the target signal In the in-vehicle microphone system, the first and second microphones are disposed close to each other, the first microphone is disposed on the ceiling of the cabin directly above the speaker's face, and the second microphone is disposed on the first microphone. by spaced apart from the position of the first microphone in the vehicle compartment ceiling 1~5cm about occipital side, the first microphone has a large speaker speech than the second microphone, and wherein The noise other than the speaker voice in the car with respect to the speaker voice is reduced, the speaker voice of the second microphone is smaller than that of the first microphone, and the voice other than the speaker voice in the car with respect to the speaker voice The adaptive signal processing unit is configured to reduce the power of the error signal output from the error signal generation unit by adaptive signal processing during non-speech recognition when only noise is input to the first and second microphones. The coefficient of the adaptive filter is determined so as to be the minimum, and when the noise and the voice signal are input to the first and second microphones, the coefficient of the adaptive filter is not updated and the error signal is converted into the voice signal. Is achieved by outputting as
In this way, the noises Xn ₁ (z) and Xn ₂ (z) included in the output signals of the two microphones can be made substantially equal, while the audio signal Xs ₁ (z) included in the output signals of the two microphones. , Xs ₂ (z) can be made different. Therefore, even if the adaptive filter coefficient W is determined so that the root mean square value of En (z) at the time of noise signal input is minimized, the audio output Es (z) in equation (2) does not become zero, The signal-to-noise ratio of the signal can be improved.
[0010]
As a specific arrangement example of the microphone, one microphone is arranged right above the speaker's face, and the other microphone is arranged about 1 to 5 cm away from the position of the one microphone on the back of the head. In this way, due to the human sound radiation characteristics, one microphone picks up the sound with the highest possible signal-to-noise ratio, even though the microphones are located relatively close, and the other microphone has the lowest possible The sound can be picked up by SN ratio.
[0011]
DETAILED DESCRIPTION OF THE INVENTION
(A) Configuration of Microphone System FIG. 1 is a configuration diagram of a microphone system according to the present invention, and the same components as those in the system of FIG. In the figure, 10 is a speaker, for example, a car driver, and 11 and 12 are first and second microphones. The first microphone 11 is arranged on the ceiling directly above the speaker's face, and the second microphone 12 is arranged on the ceiling on the back of the head about 1 to 5 cm from the first microphone position.
3 is adaptive signal processing unit, the output signal x ₂ microphones 2 with the error signal e is input is input as a reference signal, the adaptive signal processing based on the LMS algorithm so that the power becomes a minimum error signal e Do. In the adaptive signal processing unit 3, 3a is an LMS calculation unit, and 3b is an adaptive filter having an FIR digital filter configuration. The LMS calculation unit 3a determines the coefficient of the adaptive filter 3b so that the power of the error signal e is minimized by adaptive signal processing. The adaptive signal processing unit 3 determines the filter coefficient W of the adaptive filter 3b by adaptive signal processing only during non-speech recognition, does not update the filter coefficient during speech recognition, and uses the filter coefficient W determined during non-speech recognition. Set to adaptive filter 3b.
Reference numeral 4 denotes a target response setting unit which receives a signal output from the microphone 11 as a target signal, has a delay characteristic of time d, and has a flat characteristic (a characteristic of gain 1) in the audio frequency band. . A subtracting unit 5 subtracts the output signal of the adaptive filter 3b from the target response output from the target response setting unit 4, and outputs an error signal e.
[0012]
(B) Human voice radiation characteristics FIG. 2 shows human voice radiation characteristics, and (a) is a radiation characteristic chart showing the voice level at a predetermined distance from the mouth for each frequency on a horizontal plane including the mouth of the speaker 10, (B) is a radiation characteristic diagram showing the sound level at a predetermined distance from the mouth for each frequency on a vertical plane including the mouth of the speaker 10. In the figure, A is a characteristic of 125 Hz to 250 Hz, B is a characteristic of 500 Hz to 700 Hz, C is a characteristic of 1400 Hz to 2000 Hz, and D is a characteristic of 4000 Hz to 5600 Hz. As is clear from this radiation characteristic diagram, human-generated speech is radiated most strongly in the direction of the speaker's front, and the power of the sound radiated upward, downward, and in the left-right direction is higher than that in the direction of the speaker's front. small.
Therefore, as shown in FIG. 1, the first microphone 11 is arranged on the ceiling directly above the speaker's face, and the second microphone 12 is arranged on the ceiling on the back of the head about 1 to 5 cm from the first microphone position. (1) While the power of noise received by the two microphones 11 and 12 can be made substantially the same, (2) the sound power received by the two microphones 11 and 12 can be made different. That is, the noises Xn ₁ (z) and Xn ₂ (z) included in the output signals of the two microphones 11 and 12 can be made substantially equal, and the audio signal Xs ₁ (z) included in the output signals of the two microphones 11 and 12 ), Xs ₂ (z) can be made different, and [Xn ₁ (z) / Xn ₂ (z)] ≠ [Xs ₁ (z) / Xs ₂ (z)].
[0013]
(C) At the time of non-speech recognition in which only noise is input to the operating microphones 11 and 12, the adaptive signal processing unit 3 performs the following equation by adaptive signal processing:
En (z) ＝ Xn ₁ (z) z ^-d −Xn ₂ (z) W (z) (1)
, The filter coefficient W of the adaptive filter 3b is determined so that the average value of {En (z)} ² is minimized.
On the other hand, at the time of speech recognition, the adaptive signal processing unit 3 does not update the filter coefficient, sets the filter coefficient W determined at the time of non-speech recognition to the adaptive filter 3b, and outputs a speech signal. In this case, the audio signal Xs ₁ (z) included in the outputs of the microphones 11 and 12,
Xs ₂ (z) is different and [Xn ₁ (z) / Xn ₂ (z)] ≠ [Xs ₁ (z) / Xs ₂ (z)].
Es (z) ＝ Xs ₁ (z) z ^-d −Xs ₂ (z) W (z) (2)
The audio output Es (z) obtained by the above does not become the minimum value (unlike noise, it is not so small).
From the above, even if the adaptive filter coefficient W is determined so that the power of the noise output En (z) in equation (1) is 0, the audio output Es (z) in equation (2) is not as small as noise. In other words, the S / N ratio of the audio signal can be improved.
[0014]
In summary, as shown in FIG. 1, by arranging the microphones 11 and 12 at a relatively short distance, the reduction in coherence between the noises output by the two microphones is reduced. Further, as shown in FIG. Considering the human sound radiation characteristics, one microphone 11 picks up sound with as high an S / N ratio as possible even though the microphones 11 and 12 are arranged at a relatively short distance, and the other microphone 12 Pick up the audio with the lowest possible signal-to-noise ratio. As a result, even if the adaptive filter coefficient W is determined so that the noise output becomes zero, the audio output does not become small like the noise, and the SN ratio of the audio signal can be improved.
[0015]
(D) Examination of microphone position and S / N ratio improvement amount The sound uttered by humans and radiated into the space from the radiation characteristics of FIG. 2 is particularly greatly attenuated on the back of the head, and the level becomes smaller than the sound radiated to the front. I understand that. Therefore, as shown in FIG. 1, the microphone system of the present invention is basically installed from near the human head to the back of the head. Thus, the first and second microphones 11 and 12 are installed in this way. By installing it, the SN ratio can be greatly improved.
FIG. 3 is an explanatory diagram of the position of the pair microphone, and FIG. 4 is a diagram showing the SN ratio improvement amount at each pair microphone position of FIG. As shown in FIG. 3, pair microphones 11 and 12 with a spacing of 3 cm are installed at a plurality of positions {circle around (1)}, {circle over (2)}, {circle around (3)}, and each SN ratio improvement amount (how much the SN ratio is improved). ) With a 1500cc passenger car (sedan), the results shown in FIG. 4 were obtained. From FIG. 4, when the pair microphones 11 and 12 are installed at the position (1), that is, “one microphone is installed almost directly above the face of the speaker 10, and the other microphone is separated a little from the back of the head. It can be seen that the amount of improvement in the S / N ratio is the highest when it is arranged on the side.
[0016]
FIG. 5 is an explanatory diagram of the pair microphone interval, and FIG. 6 is a diagram showing the SN ratio improvement amount at each pair microphone interval of FIG. As shown in FIG. 5, the first microphone 11 is fixed almost right above the face of the speaker 10, and the second microphone 12 is placed 3 cm, 6 cm, 9 cm, and 12 cm apart from each other on the back of the head, so that the optimum When the distance between the microphones was examined, the result shown in FIG. 6 was obtained. From FIG. 6, it can be considered that the smaller the distance between the two microphones 11 and 12, the higher the improvement in the S / N ratio. However, in the system shown in FIG. 1, if the interval is set to 0 cm, the noise can be completely erased, but the sound is also completely erased. For this reason, it does not function as a voice receiving system. In addition, even a small microphone has its own size, so even if the microphones are completely attached to each other, the distance between the centers of the microphones cannot be smaller than about 1 cm. Therefore, the distance between the microphones is preferably about 1 to 5 cm at most, although there is a slight width depending on the difference in the vehicle type and the size of the microphone.
[0017]
FIG. 7 is an explanatory diagram of the SN ratio improvement amount for each speaker. As can be seen from FIG. 7, in the microphone system of the present invention, the variation in performance (SN ratio improvement amount) due to the difference between persons is about 1 dB, and the influence due to the difference between speakers is small.
The case where the two microphones are arranged above the speaker's head has been described above, but the two microphones are arranged at a relatively short distance, one microphone picks up the sound with the highest possible SN ratio, and the other microphone If the voice can be picked up with a signal-to-noise ratio as low as possible, the arrangement position is not limited to overhead.
The present invention has been described with reference to the embodiments. However, the present invention can be variously modified in accordance with the gist of the present invention described in the claims, and the present invention does not exclude these.
[0018]
【The invention's effect】
As described above, according to the present invention, the microphones are arranged close to each other, the SN ratio of the signal output from one microphone is increased, and the SN ratio of the signal output from the other microphone is decreased. Even if the adaptive filter coefficient is determined so that the output is minimized, the audio output does not become zero, and the SN ratio of the audio signal can be improved. That is, although the number of microphones is small, sound can be received and output with a high SN ratio.
Also, according to the present invention, one microphone is disposed on the ceiling directly above the speaker's face, and the other microphone is disposed about 1 to 5 cm away from the position of the one microphone on the back of the head, thereby Although one microphone is picked up with a signal-to-noise ratio as high as possible, one microphone can pick up sound with a signal-to-noise ratio as low as possible.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a microphone system of the present invention.
FIG. 2 is a human radiation characteristic diagram.
FIG. 3 is a diagram illustrating the position of a pair microphone.
FIG. 4 is a relationship diagram between a pair microphone position and an SN ratio improvement amount.
FIG. 5 is an explanatory diagram of a pair microphone interval;
FIG. 6 is a relationship diagram between a pair microphone interval and an SN ratio improvement amount.
FIG. 7 is an explanatory diagram of an SN ratio improvement amount for each speaker.
FIG. 8 is a relationship diagram between SN ratio and recognition rate.
FIG. 9 shows a high S / N ratio sound receiving system in the case of using two conventional microphones.
FIG. 10 is a characteristic diagram of a target response setting unit.
[Explanation of symbols]
11, 12... First and second microphones 3... Adaptive signal processing unit 3 a... LMS computing unit 3 b.

Claims (1)

The first and second microphones, a signal output from the first microphone, a target signal output unit that outputs a target signal, and a signal output from the second microphone are input as reference signals. In-vehicle microphone system comprising an adaptive signal processing unit that performs adaptive signal processing and an error signal generation unit that outputs a difference between an output signal of an adaptive filter that constitutes the adaptive signal processing unit and the target signal,
The first and second microphones are disposed close to each other, the first microphone is disposed on the ceiling of the passenger compartment and directly above the speaker's face, and the second microphone is positioned 1 from the position of the first microphone. By placing the first microphone about 5 cm away from the ceiling on the back of the passenger compartment , the first microphone has a louder speaker voice than the second microphone, and noise other than the speaker voice in the car relative to the speaker voice. The second microphone has a lower speaker voice than the first microphone, and a noise other than the speaker voice in the car corresponding to the speaker voice is increased.
The adaptive signal processing unit is adapted to minimize the power of the error signal output from the error signal generation unit by adaptive signal processing during non-speech recognition in which only noise is input to the first and second microphones. A coefficient of a filter is determined, and at the time of voice recognition in which noise and a voice signal are input to the first and second microphones, the error signal is output as a voice signal without updating the coefficient of the adaptive filter;
A microphone system characterized by this.

Images