JP3296024B2 - Loudspeaker automatic adjustment device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a loudspeaker automatic adjusting apparatus for automatically adjusting the level and frequency characteristics of an audio signal loudspeaked in a sound field using a speaker, for example, by using fuzzy inference.

[0002]

2. Description of the Related Art Conventionally, an automatic loudspeaker sound adjusting apparatus detects noise in a sound field where an audio signal is loudspeaked, for example, and determines the level of the audio signal loudspeaked in the sound field in accordance with the detection result of the noise. There is something to adjust.

For example, Japanese Patent Publication Nos. 4-78045 and 4-7804
No. 6 and 4-78047, a microphone is arranged in a sound field, and a loud sound loudspeaked from a speaker is removed from an audio signal detected by the microphone using a low-pass filter. An apparatus for separating only a noise signal and adjusting the level of an audio signal supplied to a speaker in proportion to the level of the noise signal is disclosed.

[0004]

However, in general, in order to construct a good loudspeaker sound field, a skilled operator adjusts the sound system in accordance with the expertise and experience of the operator. In this case, the adjustment involves not only adjusting the level of the audio signal supplied to the speaker, but also adjusting various parameters such as equalization. This is because simply adjusting the level of the audio signal is not enough to make the loudspeaker sound satisfactorily audible.

However, in the above-described automatic adjusting device, the level of the audio signal is merely adjusted according to the level of the noise level, and the adjustment of the level is basically performed with a large noise level. Sometimes, when the level of the audio signal is increased, and when the noise level is low, simple linear control is performed to decrease the level of the audio signal, so that there is a problem that the loudspeaker cannot be heard well. In addition, when the noise level is high, the level of the audio signal is also high, so that there is a problem that the loudness of the sound field as a whole may be further increased.

[0006]

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and has a signal source of an audio signal and a loudspeaker for loudspeaking the audio signal from the signal source into a sound field. Noise detecting means for detecting noise in the sound field, frequency analyzing means for frequency-analyzing the audio signal, and controlling the level of the audio signal based on a level control signal to provide the sound signal to the loudspeaker.
And feeding to the level control means, the frequency characteristic of the audio signal is controlled based on the frequency characteristic control signal, the expansion
Frequency characteristic control means for supplying to the voice means, and control means for generating the level control signal and the frequency characteristic control signal based on a noise detection signal of the noise detection means and a frequency analysis result from the frequency analysis means. Have
Things. The noise detecting means is provided for detecting the sound field.
Audio signal detecting means for detecting an audio signal;
E signal is supplied and is equal to the transfer function in the above sound field
Sound field transfer function equalizing means having a transfer function;
Signal of the sound field transfer function equalizing means.
Calculating means for calculating a difference from the output signal.
You. The control means may generate the level control signal and the frequency characteristic control signal based on fuzzy inference. Further, the noise detecting means may include a level detecting means for detecting a level of an output signal of the calculating means. Further, the noise detection means may include noise characteristic detection means for detecting noise characteristics different from the noise level. Further, the frequency characteristic control means may be configured to control the signal level of the frequency band of the audio signal designated based on the band designation signal, based on the frequency band level control signal, The control means may generate a band designation signal and a frequency band level control signal.

The present invention also provides a signal source for an audio signal.
And the above audio signal from this signal source to the sound field
And a noise detecting means for detecting noise in the sound field.
Noise detection means and frequency analysis of the audio signal
Frequency analysis means, volume and sound quality of the audio signal
Adjustment means for adjusting the noise, and noise detection by the noise detection means.
Outgoing signal and the frequency analysis result from the frequency analysis means
Control means for controlling the adjusting means based on the
ing. The frequency analysis means converts the audio signal
Input to the audio signal and the predetermined teacher signal.
A neural network that supplies a value representing the similarity to the control means
Network. The above control means
The noise detection signal from the noise detection means and the neural network
Based on the similarity from the network
A fuzzy inference unit for controlling the adjusting means.
You. The noise detecting means is a sound signal in the sound field.
Audio signal detecting means for detecting
Transfer function provided and equal to the transfer function in the sound field
Sound field transfer function equalizing means having
The output signal of the stage and the output signal of the sound field transfer function equalizing means
Calculation means for calculating the difference from
You.

[0008]

According to the present invention, the audio signal from the signal source is loudspeaked into the sound field by the loudspeaker. At this time, noise in the sound field is detected by the noise detection means.
On the other hand, the frequency analysis of the audio signal is performed by the frequency analysis means. The noise detection signal of the noise detection unit and the frequency analysis result from the frequency analysis unit are input to the control unit, and the control unit generates a level control signal and the frequency characteristic control signal based on these input values. The level control signal is input to the level control means, and the level control means controls the level of the audio signal. Further, the frequency characteristic control signal is input to the frequency characteristic control means, and the frequency characteristic control means controls the frequency characteristic of the audio signal. The noise detecting means converts the audio signal into a sound field transfer function.
The sound field transfer function is processed by the equalizing means.
Calculate the difference between the output signal and the sound field detection signal.
The loudspeaker was affected by the transfer function of the sound field
Remove the audio signal and superimpose it on the audio signal.
Only the noise signal obtained can be detected. When fuzzy inference is used for the control means, control according to human senses can be performed instead of simple linear control. In addition,
According to Ming, neural networks are used as frequency analysis means.
The frequency analysis result is unique because
And only one transmission path to the control means is required.

[0009]

1 and 2 show a first embodiment. As shown in FIG. 1, the present embodiment has an audio signal source 2,
The audio signal from the signal source 2 is supplied to an amplifier 8 via frequency characteristic control means such as an equalizer 4 and level adjustment means such as a variable attenuator 6. The audio signal amplified by the amplifier 8 is loudspeaked to a sound field 10 from a loudspeaker provided in a sound field 10 such as a hall. In addition, in FIG.
2 shows only one unit, but a plurality of units may be provided.

The output signal of the variable attenuator 6 is supplied to a frequency analysis means 14. The frequency analysis means 14 has a fast Fourier transform (FFT) analyzer 16 by means of which, for example, the audio signal is spread over three decades.
1/3 octave analysis is performed to obtain each level value in 29 frequency bands. These level values represent the spectrum of the audio signal. This spectrum is input to the neural network 18 in which the male and female voices have been learned, and the degree to which the audio signal is male / female is determined.

For example, when an announcement is louder, the clarity at the listening point differs depending on the voice quality of the speaker, and the female voice generally sounds clearer than the male voice. However, even in the case of a male voice, if the level in an appropriate frequency band is increased by the equalizer 4, the clarity of the male voice is considerably improved. Therefore, the frequency analysis means 14 is provided to determine whether or not to increase the level in an appropriate frequency band, and to determine how much to increase the level.

Note that the neural network 18
If the input spectrum input from the FT analyzer 16 is a male voice, 1 has been learned as a teacher signal if the input spectrum is female, and a value of -1 has been learned as a teacher signal. Output.

The sound field 10 is provided with an audio signal detecting means, for example, a microphone 20, and picks up an audio signal loudspeaked from a speaker 12 and noise existing in the sound field 10. In addition, 22 and 22 shown in FIG. 1 represent noise sources.

The output signal of the variable attenuator 6 is supplied to an FIR filter 26 via a delay circuit 24.
The delay time Δt of the delay circuit 24 is adjusted to a time required for the audio signal loudspeaked from the speaker 12 to reach the microphone 20. The filter coefficient of the FIR filter 26 is set so that the transfer function of the FIR filter 26 is equal to the transfer function from the speaker 12 to the microphone 20 in the sound field 10. Therefore, the output of the FIR filter 26 is almost equal to the audio signal collected by the microphone 20.

[0015] The output of the FIR filter 26, the arithmetic hand
The output of the adder 28 is equal to the noise waveform in the sound field 10 because it is subtracted from the output signal of the microphone 20 in a stage, for example in the adder 28, and
The output of stage 34 is the level of the noise signal in sound field 10. Thus, the microphone 20, the delay circuit 24, and the F
IR filter 26, adder 28 and level detecting means 34
Constitutes a noise detecting means. The reason why the noise detecting means is provided is to adjust the level of the audio signal according to the noise.

As the filter coefficients of the FIR filter 26, for example, when the sound field 10 is in a sufficiently quiet state, the speaker 12 and the microphone 20 are measured using a measuring instrument.
Can be used. Alternatively, an adaptive filter is used instead of the FIR filter 26, and also in a state where the sound field 10 is sufficiently quiet, the adder 28 obtains the difference between the microphone 20 and the output of the adaptive filter, and this difference is minimized. The coefficients of the adaptive filter may be adjusted to be as follows.

The output signal of the level detection means 34 , ie, the level of the noise signal in the sound field 10, and the output signal of the neural network 18, ie, the signal indicating the degree of male / female voice of the audio signal, are controlled by control means, for example, a fuzzy inference unit. 30.

The fuzzy inference unit 30 controls the variable attenuator 6 and the equalizer 4 based on the two inputs. The control on the variable attenuator 6 changes the amount of attenuation, and the control on the equalizer 4 fixes a region to be equalized to a specific frequency band, for example, 1 to 3 kHz, preferably 2.5 kHz band. How many dB this band is raised or lowered. That is, the amount of equalization is adjusted.

That is, the noise level and the degree of male / female voice are used as the antecedent variables, and the attenuation of the variable attenuator 6 and the level of the 2.5 kHz frequency band of the equalizer 4 are used as the consequent variables. ing.

As shown in FIG. 2A, the noise level as the antecedent variable includes seven fuzzy sets LB to H
B. The membership function for LB (the smaller the noise level is the largest) has a membership of 1 when the noise level is 30 dB (A) or less, and the membership is reduced to 0 as the noise level approaches 40 dB (A). It gradually decreases toward it.

The membership function of the membership function for LM (the noise level is intermediate when the noise level is small) is 0 when the noise level is 30 dB (A) or less, and the membership level increases as the noise level approaches 40 dB (A). The degree of membership gradually increases toward 1, and the degree of membership gradually decreases toward 0 as the noise level approaches 50 dB (A).

The membership function for LS (the smaller the noise level is, the smaller the noise level is)
Below 0 dB (A), the degree of belonging is 0, and the degree of belonging gradually increases toward 1 as the noise level goes to 50 dB (A), and goes toward 0 as the noise level goes to 60 dB (A). Decrease gradually.

The membership functions for M (intermediate noise level), HS (high noise level is low), HM (high noise level is medium), and HB (high noise level is high) are also shown in FIG. 2 (a)
Are defined in the same manner as described above.

The degree of male / female voice as an antecedent variable (output signal of the neural network 18) is composed of a fuzzy set of male and female, and the membership function for male is as shown in FIG. When the output signal of the neural network 18 is 0.5 or more, the degree of membership is 1
And the output signal of the neural network 18 is-
As the value gradually decreases toward 0.5, the degree of belonging also gradually decreases toward 0. The membership function for women is such that the membership is 1 when the output signal of the neural network 18 is -0.5 or less, and gradually decreases as the output signal of the neural network 18 gradually increases toward 0.5. is there.

The attenuation of the variable attenuator 6 as a consequent variable is composed of five fuzzy sets LS to HB, and their membership functions are used to determine the noise level as shown in FIG. They correspond to membership functions of fuzzy sets LS to HB, respectively.

The equalizer 4 is also used as a consequent variable.
The 2.5 kHz band level is composed of fuzzy sets 0 to 6 as shown in FIG. The membership function corresponding to the fuzzy set 0 has a membership of 0 when the level of the equalizer 4 is -1 dB or less, and the membership increases gradually toward 1 as the level of the equalizer 4 gradually increases toward 0 dB. And the level of the equalizer 4 is 1d
As the number gradually decreases toward B, the degree of belonging gradually decreases toward zero.

The membership function corresponding to the fuzzy set 1 has a membership of 0 when the level of the equalizer 4 is equal to or less than 0 dB, and increases the membership toward 1 as the level of the equalizer 4 gradually increases toward 1 dB. As the level of the equalizer 4 gradually decreases toward 2 dB, the degree of membership gradually decreases toward zero.

Hereinafter, the membership functions for the fuzzy sets 2 to 6 are also determined as shown in FIG.

Based on each such membership function,
The fuzzy inference unit 30 performs inference based on rules as shown in Table 1.

[Table 1]

That is, for example, in rule 1, the noise is LB
In the case of a male voice, the attenuation of the variable attenuator 6 is set to LS, and the level of the equalizer 4 is set to fuzzy set 0. This is called "IF noise = LB & sex = man
THEN attenuation level = LS & equalizer level = 0 "
It is described.

When the other rules 2 to 14 are described according to this description, the respective rules are as follows. Rule 2 IF Noise = LB & Sex = Female THEN Attenuation Level =
LS & equalizer level = 0. Rule 3 IF Noise = LM & Sex = Male THEN Attenuation Level =
M & equalizer level = 1. Rule 4 IF Noise = LM & Sex = Female THEN Attenuation Level =
M & equalizer level = 0. Rule 5 IF noise = LS & sex = male THEN attenuation level =
M & equalizer level = 2. Rule 6 IF Noise = LS & Sex = Female THEN Attenuation Level =
M & equalizer level = 1. Rule 7 IF Noise = M & Sex = Male THEN Attenuation Level = H
S & Equalizer level = 4. Rule 8 IF Noise = M & Sex = Female THEN Attenuation Level = H
S & Equalizer level = 2. Rule 9 IF Noise = HS & Sex = Male THEN Attenuation Level =
HS & equalizer level = 5. Rule 10 IF Noise = HS & Sex = Female THEN Attenuation Level =
HS & equalizer level = 3. Rule 11 IF noise = HM & sex = male THEN attenuation level =
HM & equalizer level = 6. Rule 12 IF Noise = HM & Sex = Female THEN Attenuation Level =
HM & equalizer level = 4. Rule 13 IF Noise = HB & Sex = Male THEN Attenuation Level =
HB & equalizer level = 6. Rule 14 IF Noise = HB & Sex = Female THEN Attenuation Level =
HB & equalizer level = 5.

According to these rules, the fuzzy inference unit 30 makes an inference. For example, if inference is performed by the max-min centroid method, the following is performed. For example, FIG.
As shown by the symbol a in FIG.
It is assumed that the output signal of the neural network 18 is approximately 0.3 as shown in FIG.

In this case, as apparent from FIG. 2A, the noise level crosses the membership function for LS and M. Therefore, as is clear from Table 1, LS,
Rules related to M are rules 5 to 8. Here, the membership degrees of the membership functions of M and LS with respect to about 58 dB (A) are about 0.8 and about 0.2, respectively, as shown in FIG. The membership degrees of the male membership function and the female membership function with respect to the output signal 0.3 of the neural network 18 are 0.78 and 0.22 as shown in FIG.

Therefore, rule 5 (IF noise = LS &
Sex = Male THEN Attenuation level = M & Equalizer level = 2. ), The && (logical product) of the degree of belonging of the noise level of the antecedent part to LS of 0.2 and the degree of belonging of male to gender of 0.78 is 0.2. Considering the attenuation level in the consequent part, the membership function of M is truncated according to the above-mentioned 0.2, which is a region indicated by hatching in the lower left direction in FIG.

Rule 6 (IF noise = LS & sex = female THEN attenuation level = M & equalizer level =
One. ), The degree of belonging of the noise level of the antecedent part to LS of 0.2 and the degree of belonging of the female to sex 0.22 are also 0.2, and the attenuation level of the consequent part is 0 .2 indicates the region indicated by hatching in the lower left direction, with the membership function of M truncated.

Rule 7 (IF noise = M & property =
Male THEN attenuation level = HS & equalizer level =
4. ), 0.75 is obtained when & of the noise level M of the antecedent part to M of 0.8 and the male degree of sex to sex 0.78 is obtained. The attenuation level of the consequent part is a region indicated by a satin pattern obtained by truncating the HS membership function by 0.78.

Rule 8 (IF noise = M & property =
Female THEN Attenuation level = HS & equalizer level =
2. ), The degree of belonging of the antecedent part to M of the noise level of 0.8 and the degree of belonging of the female to gender of 0.22 are 0.22. The attenuation level of the consequent part is a region indicated by a vertical line obtained by truncating the HS membership function according to 0.22.

Then, the logical sum of each of the obtained regions is obtained, the position of the center of gravity of the logical sum region is obtained, and the variable attenuation is set so that the amount of attenuation of the variable attenuator 6 becomes equal to the amount of attenuation corresponding to the position of the center of gravity. The controller 6 is controlled. That is, defuzzification is performed by the centroid method.

In the control of the equalizer 4, the rule 5
As a result, & of the degree of belonging of the noise level of the antecedent part to LS of 0.2 and the degree of belonging of male to gender of 0.78 is 0.2.
When the membership function of the fuzzy set 2 in (d) is truncated, the region becomes a region indicated by hatching in the lower left direction.

Further, if the " of 0.2 of the noise level of the antecedent part to LS and 0.22 of the female to gender are found by rule 6, " 0.2 " When the membership function of the fuzzy set 2 of the partial equalizer 4 is truncated by the above-mentioned 0.2, the region becomes a region indicated by hatching in the lower left.

According to rule 7, & of the degree of belonging of noise level M of the antecedent part to M of 0.8 and the degree of belonging of male to gender of 0.78 is 0.78. The membership function of fuzzy set 4 in the consequent is
When the head is truncated by 8, the area becomes a matte area.

According to Rule 8, & of the degree of belonging of the antecedent part of noise level to M of 0.8 and the degree of belonging of female to sex of 0.22 is 0.22. The membership function of the fuzzy set 2 of the consequent part is described in the above-mentioned “0.
When the head is truncated by 22, the region is indicated by a vertical line.

Then, the logical sum of each of the obtained regions is obtained, the position of the center of gravity of the logical OR region is obtained, and the equalizer 4 is controlled so that the level output of the equalizer 4 corresponding to the position of the center of gravity is obtained.

As described above, the levels of the variable attenuator 6 and the frequency equalizer 4 are sequentially adjusted. In particular, not only the signal level but also the frequency characteristics are adjusted. The clarity of the announcement can be improved without increasing the size. Further, since the control is not a simple linear control but a fuzzy control, it can be controlled to a state close to human sense by appropriately setting rules and membership functions of the fuzzy control unit.

FIG. 3 shows a second embodiment. In this embodiment, in addition to the level of the noise signal by the adder 28, the fuzzy inference unit 30 calculates an index representing the frequency component of the noise signal and / or the quality of the ambient noise such as the temporal change of the noise signal. The output of the index calculator 32 is input to the fuzzy inference unit 30. The output of the variable attenuator 6 is also input to the fuzzy inference unit 30.

The fuzzy inference unit 30 controls the frequency band in which equalization is performed by the equalizer 4 and the amount of equalization in that band, in addition to adjusting the amount of attenuation of the variable attenuator 6. Of course, since the number of data in the antecedent part and the number of controlled objects in the consequent part are increasing, it is necessary to use a different rule from that of the first embodiment.

In this embodiment, since the band in which the equalization is performed is also controlled, the clarity can be further improved in accordance with the state of the noise and the state of the audio signal.

In each of the above embodiments, the fuzzy inference unit is used as the control means. However, the control unit may perform a normal linear control. The noise detecting means includes a microphone 20, a delay circuit 24, an FIR filter 26, an adder 2
Is constituted by 8 and level detector 34 may detect the noise level by other configurations.

[0049]

As described above, according to the present invention, not only the control of the audio signal level according to the noise level, but also the frequency characteristic of the audio signal to be amplified in the sound field according to the frequency characteristic of the audio signal. , The clarity of the audio signal that is amplified in the sound field is improved. In addition, noise detection is based on audio signal detection means and sound field.
Since it is performed using transfer function equalization means,
Noise can be detected. Since fuzzy inference is used for control, it is possible to perform control in accordance with human senses more than using linear control. Further, in the control of the frequency characteristics, in addition to the adjustment of the level of a specific frequency band, the clarity is further improved when the band for controlling the frequency characteristics is changed according to noise or an audio signal. be able to. In addition, since control based on noise is also performed according to noise characteristics different from the noise level , clarity can be further improved. In this way, by setting rules and membership functions to effectively control not only the level of the audio signal but also the frequency characteristics, clarity can be achieved without unnecessarily increasing the loudness of the entire sound field. Since it can be improved, a comfortable sound space can be configured. Neural networks are used for frequency analysis.
Analysis results are represented by unique values,
Only one transmission path of the analysis result to the control means is required.

[Brief description of the drawings]

FIG. 1 is a block diagram of a first embodiment of a loudspeaker automatic adjusting apparatus according to the present invention.

FIG. 2 is a diagram showing each membership function of fuzzy inference used in the first embodiment.

FIG. 3 is a block diagram of the second embodiment.

[Explanation of symbols]

 Reference Signs List 2 audio signal source 4 equalizer (frequency characteristic control means) 6 variable attenuator (level adjustment means) 10 sound field 12 speaker (loudspeaker means) 14 frequency analysis means 20 microphone (noise detection means) 24 delay circuit (noise detection means) 26 FIR filter (noise detection means) 28 Adder (noise detection means) 30 Fuzzy inference unit (control means) 32 Index calculation unit (noise characteristic detection means)

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H03G 3/20-11/08

Claims (8)

(57) [Claims] 1. A signal source of an audio signal, a loudspeaker for loudspeaking the audio signal from the signal source into a sound field, a noise detector for detecting noise in the sound field, and a frequency analysis of the audio signal Frequency analysis means, controlling the level of the audio signal based on a level control signal, level control means to supply to the loudspeaker means, controlling the frequency characteristics of the audio signal based on the frequency characteristic control signal , Frequency characteristic control means for supplying to the loudspeaker means ; control means for generating the level control signal and the frequency characteristic control signal based on a noise detection signal of the noise detection means and a frequency analysis result from the frequency analysis means; Wherein the noise detection means detects an audio signal in the sound field.
And the audio signal is supplied and transmitted in the sound field.
Sound field transfer function equalization means having a transfer function equal to the function
And an output signal of the audio signal detecting means, and the sound field transfer function
A calculating means for calculating a difference between the output signal of the equalizing means, amplified sound automatic adjustment device provided. 2. The loudspeaker automatic control apparatus according to claim 1, wherein said control means generates said level control signal and said frequency characteristic control signal based on fuzzy inference. . 3. The automatic loudspeaker adjustment apparatus according to claim 1, wherein said noise detecting means is provided by said arithmetic means.
An automatic loudspeaker adjustment device comprising level detection means for detecting the level of an output signal . 4. The automatic loudspeaker sound adjusting apparatus according to claim 3, wherein said noise detecting means is different from a noise level.
A noise characteristic detecting means for detecting a characteristic of the loudspeaker. 5. The automatic loudspeaker adjustment apparatus according to claim 1, wherein said frequency characteristic control means adjusts a signal level of a frequency band of said audio signal designated based on a band designation signal. An automatic loudspeaker adjusting apparatus configured to be controlled based on a band level control signal, wherein the control means generates the band designation signal and the frequency band level control signal. 6. A signal source of an audio signal, a loudspeaker for loudspeaking the audio signal from the signal source into a sound field, a noise detector for detecting noise in the sound field, and a frequency analysis of the audio signal. Frequency analysis means, adjustment means for adjusting the volume and sound quality of the audio signal, control means for controlling the adjustment means based on a noise detection signal of the noise detection means and a frequency analysis result from the frequency analysis means, Wherein the frequency analysis means is provided with a neural network which inputs the audio signal and supplies a value representing the similarity of the audio signal to a predetermined teacher signal to the control means. . 7. The automatic loudspeaker adjustment apparatus according to claim 6, wherein said control means controls said adjusting means based on a noise detection signal from said noise detection means and a value representing similarity from said neural network. A loudspeaker automatic adjustment device having a fuzzy inference unit to control. 8. The automatic loud sound adjusting apparatus according to claim 7, wherein said noise detecting means is a sound signal detecting means for detecting a sound signal in said sound field, and said audio signal is supplied and transmitted in said sound field. Sound field transfer function equalizing means having a transfer function equal to the function, and calculating means for calculating a difference between an output signal of the audio signal detecting means and an output signal of the sound field transfer function equalizing means. Loudspeaker automatic adjustment device.