JP3086539B2 - Wind noise suppression circuit in sound pickup 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 wind noise suppressing circuit in a sound pickup device provided in a video camera or the like.

[0002]

2. Description of the Related Art In recording with a video camera, there is a problem that sound quality is degraded by wind noise at the time of outdoor photographing. Therefore, research on a method of suppressing wind noise has been conventionally conducted.

[0003] By the way, wind noise and speech generally have different frequency spectra. As shown in FIG.
While there is no large difference between the level in the high frequency band and the level in the low frequency band at the boundary of 500 Hz, most of the wind noise is a low frequency component and the high frequency component is slight.

Therefore, a wind noise suppression circuit has been proposed which compares the level of the high frequency component of the microphone output with the level of the low frequency component and automatically controls the degree of suppression of the low frequency component based on the result. (JP-A-64-20798 (H0
4R3 / 00)).

The circuit is a microphone as shown in FIG.
The output of (1) is supplied to a high-frequency level detection circuit (4) through a high-pass filter (2) and a low-pass filter
The signal is supplied to the low-level detection circuit (5) via (3).

The output of the high-pass filter (2) is supplied to one input terminal of an adder (8), and the output of the low-pass filter (3) is supplied through an attenuator (7) to the other input terminal of the adder (8). And the addition result of the adder (8) is output to the subsequent circuit via the output terminal (9).

The level A _{H of the} high frequency component obtained from the high frequency level detection circuit (4) and the level A _L of the low frequency component obtained from the low frequency level detection circuit (5) are sent to a judgment circuit (60). Judging circuit (60), when the representative value obtained by dividing the level A _L of the low-frequency component at the level A _H of the high-frequency component is below a predetermined threshold, the microphone output is determined that the audio signal, the threshold representative value of a predetermined Is exceeded, the microphone output is determined as wind noise.

The determination signal S is supplied as a control signal to an attenuator (7). When the microphone output is determined to be a voice signal, the attenuator (7) does not attenuate the output of the low-pass filter (3). Then, when it is determined that the noise is wind noise, the output of the low-pass filter (3) is attenuated and output to the adder (8).

[0009]

The frequency spectrum of the voice and the frequency spectrum of the wind noise shown in FIG. 5 are not universal, and in the case of voice, they vary from person to person.
The state of the wind noise varies depending on the wind speed and the wind direction with respect to the microphone.

However, in the above-mentioned conventional wind noise suppression circuit, the level A of the low frequency component of the microphone output is low.
Since _L and based on only one representative value obtained by dividing the level A _H of the high-frequency component is determined of the voice signal and the wind noise is performed, always those representative value corresponds to the distinction between voice signals and wind noise Absent. For example, even if the representative values are the same, the ratio of the level of the sound and the level of the wind noise included in the level A _L of the low frequency component is not constant, and even though the level of the sound is higher than the wind noise, If the low frequency component is attenuated, the sound quality will be degraded.

Of course, the threshold value in the decision circuit (60) is a predetermined constant value, and the attenuator (7)
Is ON-OFF controlled, and if the setting of the threshold value is inappropriate, there is a problem that the sound quality is further deteriorated.

It is an object of the present invention to provide a wind noise suppressing circuit in which appropriate level suppression processing is performed on wind noise so that high-quality sound pickup is always realized.

[0013]

A first wind noise suppressing circuit according to the present invention comprises a level A of a high frequency component of a microphone output.
_H and the level A _L of the low frequency component as input signals, the evaluation level W of the wind noise as a reference for the attenuation process for the low frequency component
Is determined by fuzzy inference. The control circuit (6) comprises rule storage means and fuzzy inference processing means. The rule storage unit, a plurality of fuzzy control rules are stored, as a variable of the antecedent part of the control rules, with the level A _L level A _H and the low frequency component of the high frequency components of at least the microphone output is defined The wind noise evaluation level W is defined as a variable of the consequent part. Further, the fuzzy inference processing means execute at least the basis of the input level A _L level A _H and the low frequency component of a high frequency component, the fuzzy inference in accordance with a plurality of fuzzy control rules stored in the rule storage means Then, the evaluation level W of the wind noise is determined.

[0014] The second wind noise suppression circuit according to the present invention, the input signal level A _L level A _H and the low frequency component of the high-frequency component of the microphone output, the neural network to an output signal of the evaluation level W of the wind noise (63) to control circuit (6)
And attenuates the low-frequency component of the output signal of the microphone according to the evaluation level W of the wind noise output from the control circuit (6).

[0015]

In the first wind noise suppression circuit, the level A _H of the high frequency component and the level A _{L of the} low frequency component of the microphone output are based on the characteristic points of the frequency spectrum of the audio signal and the frequency spectrum of the wind noise. A plurality of fuzzy control rules for deriving an evaluation level W of wind noise from are defined. As a result, the wind noise evaluation level W is determined to be an appropriate value by fuzzy inference using the high frequency component level A _H and the low frequency component level A _L as input signals. Attenuation processing is performed on the low frequency components of the signal.

In the second wind noise suppression circuit, an appropriate teacher signal is given to the neural network (63) in advance, and the learning corresponding to the frequency spectrum of various voice signals and the frequency spectrum of wind noise is performed. Is performed.
That is, the input of the neural network (63), is the input level A _L level A _H and the low frequency component of the high-frequency components of the microphone output, the output of the neural network (63), corresponding to the level An appropriate wind noise evaluation level W is input. As a result, the connection weight and the threshold value for each neuron constituting the neural network (63) are determined. In the actual wind noise suppression operation, the wind noise evaluation level W is calculated to an appropriate value by knowledge processing using the high frequency component level A _H and the low frequency component level _AL as input signals, and the evaluation level is calculated. In accordance with W, the low-frequency component of the output signal of the microphone is subjected to an attenuation process.

[0017]

According to the wind noise suppression circuit in the sound pickup apparatus according to the present invention, an appropriate evaluation level W can be obtained by fuzzy inference or neuro processing reflecting the characteristics of the frequency spectrum of the voice signal and the frequency spectrum of the wind noise. Since the low noise component is attenuated based on the calculated evaluation level W, the wind noise can be effectively suppressed without deteriorating the sound quality.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a first configuration example in which fuzzy inference is introduced for calculating the evaluation level W and a second configuration example in which neuro processing is introduced will be described with reference to the drawings. Note that the same components as those of the conventional circuit shown in FIG. 4 are denoted by the same reference numerals, and redundant description will be omitted.

First configuration example As shown in FIG. 1, the level A _{H of the} high frequency component obtained from the high frequency level detection circuit (4) and the level A _L of the low frequency component obtained from the low frequency level detection circuit (5) are: Fuzzy inference processing unit (6
1) and a control circuit (6) including a fuzzy rule storage unit (62)
Supplied to

The fuzzy rule storage unit (62) stores the following 2
One fuzzy control rule is stored. Rule 1: If the level A _{L of the} low frequency component is small,
The wind noise evaluation level W is set small. Rule 2: If the level A _L of the low frequency component is large and the level A _{H of the} high frequency component is small, the evaluation level W of the wind noise is set large.

Since the level of the low frequency component of the wind noise always increases, the level A _L of the low frequency component of the microphone output is obtained.
Is smaller, it can be determined that the microphone output contains almost no wind noise. Therefore, the evaluation level W of the wind noise is set small according to the rule 1.

Conversely, if the level A _L of the low frequency component is large, it cannot be determined whether this is due to a voice signal or to wind noise. The magnitude of the level A _H of the high frequency component is determined.
When the level A _H is small, a large source of level A _L of the low frequency component, mainly determines to be due to wind noise, it is to set a large evaluation level W of the wind noise.

In the above rules, the magnitude of the level is expressed by a membership function shown in FIG. That is, FIG. 2 (a), antecedent membership function of rule 1 mu and (level A _L of the low frequency component is small), and represents the consequent part membership function mu (evaluation level W is small) . FIG. 2B shows two membership functions μ of the antecedent part (the level A _L of the low-frequency component is large and the level A _{H of the} high-frequency component is small) and the membership function μ ( (The evaluation level W is large).

The fuzzy inference processing unit (61) in FIG. 1 executes inference processing at the evaluation level W in accordance with the above-described rule, performs defuzzification by the well-known MIN-MAX centroid method, and performs evaluation. The level W is calculated.

The wind noise evaluation level W calculated in this way is supplied to the attenuator (7).
(7) is variably set to an attenuation rate corresponding to the magnitude of the evaluation level W. As a result, a signal of high sound quality with sufficiently suppressed wind noise is obtained at the output terminal (9).

In the above wind noise suppressing circuit, the control circuit
The configuration of (6) is simple, and naturally, by fuzzy control according to the characteristics of the frequency spectrum of the wind noise and the voice signal,
Since the attenuation rate of the attenuator (7) is finely adjusted, a stable wind noise suppressing operation can be realized.

Second Configuration Example As shown in FIG. 3, the control circuit (6) has two input terminals (6
4) A neural network (63) having (65) and one output terminal (66) is provided.

One input terminal (64) is connected to the level A _{H of the} high frequency component obtained from the high frequency level detection circuit (4), and the other input terminal (65) is connected to the low frequency level detection circuit. The level A _{L of the} low frequency component obtained from (5) is connected.

The neural network (63) has a known multi-layer structure, and in the intermediate layer, a neural process such as weighting is sequentially performed, and finally the wind noise evaluation level W is output from the output terminal (66). Is output.

In the control circuit (6), learning using a so-called backward propagation method (back propagation method) is performed on the neural network (63) in advance.
That is, both input terminals (64) (6) of the neural network (63)
5), the level A _L level A _H and the low frequency component of the high-frequency components of the microphone output is respectively input, output terminals (66)
Has an appropriate wind noise evaluation level W according to the level.
Is input to determine the connection weights and threshold values for each neuron constituting the hidden layer.

In the actual wind noise suppressing operation, an evaluation level W of wind noise is calculated by a neuro process using the level A _H of the high frequency component and the level A _L of the low frequency component as input signals. Is supplied as a control signal to the attenuator (7).

As a result, the attenuator (7) is variably set to an attenuation rate corresponding to the magnitude of the evaluation level W, and the output terminal (9)
Thus, a high-quality signal with sufficiently suppressed wind noise can be obtained.

[0033] In the wind noise suppression circuit, since suitable evaluation level W in accordance with the combination of the level A _L level A _H and the low frequency component of the high frequency component is calculated, simply attenuator by the ratio of the two levels Control with a higher degree of knowledge can be realized as compared with a conventional circuit that performs ON-OFF control. or,
Compared to the structure in advance a table of a suitable evaluation level W in accordance with the number of combinations of level A _L level A _H and the low frequency component of the high frequency component, the circuit scale is small.

The description of the above embodiments is for the purpose of illustrating the present invention, and should not be construed as limiting the invention described in the appended claims or reducing the scope thereof. Further, the configuration of each part of the present invention is not limited to the above-described embodiment, and it is needless to say that various modifications can be made within the technical scope described in the claims. For example, the attenuator (7) can be configured to perform ON-OFF control according to the evaluation level W of wind noise.

The high-frequency level detection circuit (4) and the low-frequency level detection circuit (5) detect the peaks of the high-frequency component and the low-frequency component, respectively, and calculate and output an integrated value or an average value. Various well-known configurations, such as a configuration that performs the above, can be adopted.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a sound collection device according to the present invention.

FIG. 2 is a graph showing a membership function used for fuzzy inference.

FIG. 3 is a conceptual diagram of a neural network provided in a control circuit.

FIG. 4 is a block diagram illustrating a configuration of a conventional sound collection device.

FIG. 5 is a frequency spectrum diagram of wind noise and an audio signal.

[Explanation of symbols]

 (1) Microphone (2) High-pass filter (3) Low-pass filter (4) High-pass level detection circuit (5) Low-pass level detection circuit (6) Control circuit (7) Attenuator

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int. Cl. ⁷ , DB name) H04R 3/00 320 G05B 13/02 G10K 11/178

Claims (2)

(57) [Claims] 1. A based on the level A _L level A _H and the low frequency component of the high-frequency component of the microphone output signal, the sound pickup apparatus to output the suppressed wind noise included in the output signal of the microphone, a plurality fuzzy control rules are stored in as a variable of the antecedent part of the control rules, with the level a _L level a _H and the low frequency components of at least the high frequency component is defined as a variable of the consequent part, the wind a rule storage means for evaluation level W of the noise is defined, at least based on said input level a _L level a _H and the low frequency component of the high frequency components, a plurality of fuzzy control rules stored in the rule storage means And fuzzy inference processing means for executing a fuzzy inference according to the above formula to determine an evaluation level W of wind noise. The low frequency component of the output signal of the microphone is attenuated according to the evaluation level W of wind noise, and the rule storage means stores at least the following two fuzzy control rules. circuit. Rule 1: If the level A _{L of the} low frequency component is small,
The wind noise evaluation level W is set small. Rule 2: If the level A _L of the low frequency component is large and the level A _{H of the} high frequency component is small, the evaluation level W of the wind noise is set large. 2. Based on the level A _L level A _H and the low frequency component of the high-frequency component of the microphone output signal, in Shuon apparatus for outputting by suppressing wind noise included in the output signal of the microphone, at least level a _L input signal level a _H and the low frequency component of the high frequency components, equipped with a neural network for the signal output evaluation level W of the wind noise, the evaluation level W of the wind noise that is output from the neural network A wind noise suppression circuit characterized in that a low frequency component of an output signal of a microphone is attenuated accordingly.