JP3367592B2 - Automatic gain adjustment device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic gain control device in audio equipment such as a loudspeaker device and an auxiliary listening device used by a hearing-impaired person. 2. Description of the Related Art In a TV conference apparatus and a telephone conference apparatus, a loudspeaking call is often used because of the necessity of a conversation between a plurality of persons. At this time, the sound pickup microphone is placed on a desk or the like, and the distance from the speaker's mouth is often not uniform. Then, the input level to the microphone is not constant depending on the speaker, which contributes to the difficulty of listening on the listening side. Also, if you are wearing hearing aids, even hearing impaired people who do not have much difficulty in one-on-one conversations say that the distance to the speaker is far, such as a conference or lecture, etc. Many people complain that they do not understand. One of the reasons may be a decrease in the input level to the hearing aid. Generally, as the distance between a sound source and a sound pickup position increases, the input sound pressure level decreases. The degree of the drop depends on the directivity of the sound source and the state of the reverberation in the room. For example, in a free sound field (virtual space having no boundaries reflecting sound such as walls, floors, ceilings, etc.), the sound source is point-sourced (360 degrees). Assuming that the sound source emits sound equally in any direction (a virtual sound source), the sound pressure level is attenuated by 6 dB when the distance between the sound source and the listening point is doubled. In an actual room, not only the direct sound but also the reflected sound at the boundary is added, so that it does not attenuate much, but when it is separated by 5 m, the sound pressure level at the position of 1 m is attenuated by about 10 dB. In order to keep the input sound pressure level constant, there has conventionally been a device called a compressor, for example. This is a device that can compress the dynamic range of the input signal to increase the gain for relatively low level inputs. However, this device measures the input level in a short time of several msec to several tens of msec, and changes the compression ratio at a high speed according to the measured value. Therefore, the main purpose of use is to suppress sudden excessive input, and it is not suitable for compensating for a decrease in sound pressure due to distance attenuation having a relatively large level width. [0005] Further, in an ordinary acoustic space, noise from an air conditioner or the like often exists in addition to a target sound. If you try to keep the target sound at a certain sound pressure level,
Because the sound fluctuates depending on the size of the target sound, it becomes very difficult to hear. [0006] In order to make the target sound easier to hear, there is a means such as "automatic gain controller (Japanese Patent Application No. 8-125597)" proposed by the present inventors. FIG. 8 schematically shows this conventional means. After the amount of stationary noise input from the input terminal 101 is automatically measured by the stationary noise discriminating circuit 102 and suppressed by the stationary noise suppressing circuit 103, the effective value is calculated before the gain is calculated by the gain calculating circuit 105. The averaging circuit 104 calculates the average effective value of the input sound,
By multiplying the calculated gain by the input sound in which the stationary noise is suppressed by the multiplication circuit 106, the input sound is output from the output terminal 107 with an average constant volume without distortion. [0007] Although the above-mentioned problem is supposedly solved by this method, non-stationary noises other than the stationary noise (for example, the sound of turning over a paper, the sound of closing a door, etc.) are also constant. There is a problem that it is amplified to make the volume. These unsteady sounds cannot always be "noise", but in a loudspeaker, sounds other than voice are unnecessary sounds, i.e., noise, except for special uses. Noise is harsh. SUMMARY OF THE INVENTION It is an object of the present invention to suppress steady background noise and to amplify only the intended sound without amplifying unsteady noise in audio equipment such as a loudspeaker device or an auxiliary hearing device used by a hearing-impaired person. It is intended to be able to listen at a high volume and without distortion. SUMMARY OF THE INVENTION An automatic gain adjusting device according to the present invention solves the above-mentioned problems, and can receive only a target sound (in this case, an audio signal) at an optimum volume. Linearly predicts the frame signal for each frame,
Extract the linear residual signal and calculate its autocorrelation function
To the obtained based on the maximum value (CM), and stationary noise identification means for identifying a stationary noise, inputting said frame signal
And is not regarded as stationary noise by the stationary noise discriminating means.
Frames are output without suppression to reduce steady noise.
Frames are stored in advance as stationary noise.
To the average of the power spectrum for the past number of frames
Suppresses constant noise by using the value based on it as a subtraction factor
A stationary noise suppression means to and outputs, the stationary noise suppression hand
The effective value of the execution signal of the output signal from the stage and the power spec
And the maximum value of the autocorrelation function (C
M), the effective value is greater than the threshold value.
Is smaller than the normal noise frame after suppression,
When the effective value is equal to or greater than the threshold value, the maximum value of the autocorrelation function is obtained.
The slope of the large value (CM) and the power spectrum S (f)
If any of them are smaller than the threshold,
The frame is determined to be a noise frame, and the effective value is the threshold value.
As described above, the maximum value (CM) of the autocorrelation function and the power
Each of the slopes of the vector S (f) is larger than the threshold.
When the input signal is a voice frame, the voice identification means for determining whether the input signal is a voice,
Frames determined to be audio in the audio identification flag storage circuit
Number is greater than the number of frames determined to be nonstationary noise
When receiving a voice, the voice is
The average effective value of the determined frame is obtained, and the sound identification
Undefined from the number of frames determined to be audio in the lag memory circuit
Non-stationary noise with a large number of frames determined to be ordinary noise
Of the predetermined number of past frames determined to be stationary noise.
The threshold of the effective value based on the average of the effective value is less than 1
The value obtained by multiplying the number of frames is the average effective value.
Continuously, the stationary noise after the suppression or the non-stationary noise.
If there is no audio with continuous input, the past is determined to be stationary noise
RMS threshold based on the average of the RMS of a given number of frames
The value obtained by multiplying the constant value by a constant smaller than 1 is the average effective value
By a voice effective value averaging means for calculating an average effective value of the audio, the average effective value is a threshold to the effective value
Is greater than the average effective threshold value.
Divided by the average effective value is equal to or less than the threshold value of the effective value.
If below, the average RMS is divided by the RMS threshold
Calculate and smooth the compression ratio, and set the target value as the effective value.
Gain calculation means for calculating a required gain by dividing the threshold value to obtain a gain; and the gain and the smoothing compression.
Multiplying means for multiplying the input signal by the ratio . FIG. 1 is a block diagram showing a basic configuration of an embodiment of the present invention . Steady noise identification circuit 10
2 to calculate the autocorrelation function of the input signal and calculate its maximum value (C
M) is calculated, and based on this, the stationary noise is identified, and the stationary noise suppressing circuit 103 suppresses the stationary noise based on the information. In the effective value calculating circuit 11, the effective value of the signal (rm
s). In voice identification circuit 108 And, peripheral
The slope of the power spectrum S (f), which is the slope of the wave number characteristic
(Ss) (detailed in [0017] to [0019])
Me. The CM, rms, and ss, which are feature quantities, and their
Threshold (thd (CM), thd (rms), thd
(Ss)) to determine the signal type. Voice run value average
Circuit 109 is in the middle of a voice call or unsteady
In the case of noise or no sound
The average effective value is calculated ([0024] to [0028]
Detailed). In the gain calculation circuit 105, the average effective value is
Use to calculate the compression ratio and calculate the gain. Multiplication circuit 10
In step 6, by multiplying the noise-suppressed audio signal by the gain , stationary noise and non-stationary noise are also suppressed, and only the voice can be made to have a constant volume. FIG. 2 is a block diagram showing the configuration of an embodiment of the present invention. In FIG. 2, 1 is an input terminal, 2 is a frequency analysis circuit, 3 is a linear prediction analysis circuit, 4 is an autocorrelation analysis circuit, 5 is a maximum value detection circuit, and 6 is a stationary noise detection circuit. 10A and 10B are turned on and off. Reference numeral 7 denotes an average noise power spectrum calculation circuit, which performs visual weighting. 8 is a subtraction circuit,
Reference numeral 9 denotes an inverse frequency analysis circuit that performs an operation in a reverse order to that of the frequency analysis circuit 2. 10A and 10B are switches, 11 is an effective value calculation circuit, 12 is an effective value storage circuit, 13 is a threshold value calculation circuit, 14 is a voice identification circuit, 15 is a voice identification flag storage circuit, 16 is a gain calculation circuit, and 17A and 17A. 17B is a multiplication circuit, 18 is a sound average effective value calculation circuit, 19 is a compression ratio calculation circuit, 20 is a compression ratio smoothing circuit, and 21 is an output terminal. Next, the operation will be described. The signal input from the input terminal 1 is a signal for each frame divided at an appropriate time. First, the stationary noise detection circuit 6 determines whether it is stationary noise. For this purpose, first, the input signal is subjected to linear prediction analysis by the linear prediction analysis circuit 3, and a linear prediction residual signal (herein, referred to as a residual signal) is extracted. The autocorrelation function is calculated by the autocorrelation analysis circuit 4, and the maximum value (CM) is calculated by the maximum value detection circuit 5. Thus, the magnitude of the periodicity of the input signal can be detected. In general, the nature of stationary noise such as air conditioner noise often has low periodicity. Therefore, when the CM is smaller than a certain threshold value, the stationary noise detection circuit 6 determines that the CM is the stationary noise (BN). The signal input from the input terminal 1 is sent to a frequency analysis circuit (FFT) 2 in parallel and converted into a frequency domain. The power spectrum S (f) of the frame determined to be stationary noise by the stationary noise detection circuit 6 is sent to the average noise power spectrum calculation circuit 7 by the operation of the switch 10A and stored therein. The average noise power spectrum calculation circuit 7 stores the power spectra S (f) for the past appropriate number of frames and calculates the average value.
Furthermore, a weighting function W (f) for minimizing the “audibility” of the residual noise previously proposed in Japanese Patent Application No. 8-125597 is multiplied by a noise average power spectrum S
ns (f) (for the weighting function W (f),
This will be explained later). It is subtracted by the subtraction circuit 8 into S
Subtract from (f). From the power spectrum S '(f) of the signal from which noise has been subtracted and the phase P (f) of the original signal, the signal is returned to a time domain signal by an inverse frequency analysis circuit (IFFT) 9.
The processing up to this point is the stationary noise suppression processing. The speech discrimination circuit 14 determines whether speech (SP), non-stationary noise (NSN), or stationary noise (B
N ′). Note that the stationary noise before noise suppression is BN, and the stationary noise (residual noise) after noise suppression is B
N ′. FIG. 3 is a block diagram showing details of one embodiment of the voice identification circuit 14. As shown in FIG. In this figure, 14A is a power spectrum tilt calculation circuit, 14B is a signal type determination circuit, 1
Reference numeral 4C denotes a parameter threshold value storage unit which stores thd (CM),
thd (SS) is stored. In the speech identification circuit 14, in addition to the two values calculated in the previous stage, that is, the effective value rms calculated by the effective value calculation circuit 11 and the maximum value CM of the linear prediction residual, the power is calculated from the power spectrum S (f). The slope (dB / Oct.) Of the frequency characteristic calculated by the spectrum tilt calculation circuit 14A is used. An example of the method will be described below. First, an average of the power spectrum S (f) for each one-third octave is obtained, and the level (dB) of the value is obtained.
Is calculated. The slope b of the regression line of this value is calculated as follows. ## EQU1 ## Here, x is N = 1, 2,..., N where y is the total number of divisions per one-third octave, and y is N powers averaged for one-third octave. The spectrum level. Sxy is the sum of the products of the residuals of x and y, and Sx is x
Is the residual sum of squares. Since b is a slope for each third octave, if it is multiplied by three, the slope for each octave (dB / dB)
Oct. ) Is obtained. For audio, this slope is usually
Since it becomes a negative value, the sign is inverted and the value is set to SS. That is, The characteristic amounts of the above signals, rms, CM, and SS, are input to the signal type determination circuit 14B, and the signal type is determined as follows. 1 Stationary noise: When the effective value rms is smaller than the threshold value thd (rms), the stationary noise (B
N '). 2 Non-stationary noise: When the effective value rms is equal to or larger than the threshold value thd (rms) and one of CM and SS is smaller than each of the threshold values thd (CM) and thd (SS), the non-stationary noise (NSN). 3. Voice: If any of the feature values exceeds the respective threshold value, it is determined that the voice is voice (SP). Referring again to FIG. 2, the signal type (SP / NSN / BN ') identified by the audio identification circuit 14 is stored in the audio identification flag storage circuit 15 continuously for the past M frames. When storing a new signal type, the oldest value is deleted. The effective value r calculated by the effective value calculation circuit 11
Of the ms, the effective value of the past L frame determined as the stationary noise by the stationary noise detection circuit 6 is stored in the threshold value calculation circuit 13. This is considered to be the effective value of the residual noise remaining after the stationary noise suppression processing. When storing an effective value N _S (N) newly determined as (residual) noise, the oldest value N _S (N−L) is deleted. Then, the average value <N _S (N)> is calculated and multiplied by one or more constants to obtain a threshold value thd (rms). This operation is performed by switching the switch 10B only when the stationary noise detection circuit 6 newly determines that the input signal is noise.
The value of (rms) is updated. On the other hand, the effective value rms calculated by the effective value calculation circuit 11 is stored in the effective value storage circuit 12 continuously for the past M frames. When storing a new effective value, the oldest value is deleted. The stored effective value sequence (rms
(N−M + 1),..., Rms (N−1), rms
(N)), in the voice average effective value calculation circuit 18,
An average effective value <rms (N)> is obtained by the following procedure. This will be described with reference to FIG. (A) During utterance of voice: stored effective value sequence (rms (N−M + 1),..., Rms (N−1),
rms (N)), the voice identification flag storage circuit 15
Averages the values determined as voice (SP) in <rms
(N)>. (B) Non-stationary noise: However, if the number of frames determined as non-stationary noise (NSN) is larger than the number of frames determined as speech, the threshold thd
A value obtained by multiplying (rms) by an appropriate constant a smaller than 1 is defined as <rms (N)>. (C) When there is no voice: input other than voice continuously for K (<M) frames, that is, threshold thd (rms)
Hereinafter, when the input of the stationary noise or the non-stationary noise continues, it is determined that there is no voice, and a value obtained by multiplying the threshold value thd (rms) by an appropriate constant a smaller than 1 is set as <rms (N)>. Using the average effective value <rms (N)>, the compression ratio calculation circuit 19 calculates the compression ratio p (N) as shown in equation (3). (Equation 3) [Equation 4] <P (N-1)> is the smoothing compression ratio of the immediately preceding frame, which is C0 + C1 = 1.0. The larger the value of C1, the smoother the change. The gain calculating circuit 16 calculates the gain G by equation (5), where d1 is the target average effective value of the output voice. ## EQU5 ## FIG. 5 shows the relationship between input and output in a double logarithmic manner. All inputs above the threshold thd (rms) are compressed and amplified to the target value dl, and the threshold thd (rm)
s) The following is elongated and attenuated. Since the speech average effective value calculation circuit 18 calculates all inputs other than speech below the threshold value, non-stationary noise other than speech is also attenuated. Here, the correspondence between the parts in FIG. 1 and FIG. 101: 1 102: 3,4,5,6 103: 2,7,8,9,10A 105: 16,19,20 106: 17A, 17B 107: 21 108: 14,15 109: 11 12, 18 FIG. 6 is an example of the processing result obtained by the present invention. (A) is an input signal, and in a certain conference room, a male voice (distance between microphone and speaker: 50 cm) and a female voice (id: 3)
m) is picked up under the air conditioning noise about 20 dB lower than the male voice,
In the meantime, several types of non-stationary noise are added.
(B) is a processing waveform according to the conventional method (Japanese Patent Application No. 8-125597), in which steady air-conditioning noise is suppressed and the sound is almost at the same level, but non-stationary noise is amplified similarly to the sound. ing. (C) is an example of the processing result obtained by the present invention, and it can be seen that non-stationary noise other than speech is sufficiently suppressed. Next, the above-mentioned weighting function W (f) will be supplementarily described. FIG. 7 is an explanatory diagram of the weighting function W (f). The weighting function W (f) can be expressed by equation (6). (Equation 6) As shown in FIG. 7, the larger the noise power spectrum of W (f) is, the larger the amount to be subtracted is. By doing so, it is possible to reduce the residual noise in the low band where the noise power is large and the overdrawing in the high band where the power is small. Β is an average weighting coefficient, S
_ns (f) indicates the noise average power spectrum. In the above embodiment of the present invention, each part of the block diagram is shown as a "circuit". However, since this can be realized by software or the like, it is generally used as "means". It is what is expressed. As described above, according to the present invention, first, stationary noise is suppressed, and gain control is performed after discriminating between speech and non-stationary noise. ,
It is possible to set only a target sound to a constant sound pressure level. As a result, it is possible to realize a comfortable listening system in which unpleasant paper-turning sounds, door sounds, and the like are suppressed together with stationary background noise, and only voice can be always heard at a constant volume.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing a basic configuration of the present invention. FIG. 2 is a block diagram showing one embodiment of the present invention. FIG. 3 is a block diagram showing details of a voice identification circuit in the embodiment of FIG. 2; FIG. 4 is a diagram for explaining the operation of the voice average effective value calculation circuit in the embodiment of FIG. 2; FIG. 5 is an input / output relationship diagram of the embodiment of FIG. 2; FIG. 6 is a waveform chart showing one processing example according to the present invention. FIG. 7 is an explanatory diagram of a weighting function used in the embodiment of FIG. FIG. 8 is a block diagram showing a schematic configuration of an automatic gain adjustment device proposed earlier. [Description of Signs] 1 Input terminal 2 Frequency analysis circuit 3 Linear prediction analysis circuit 4 Autocorrelation analysis circuit 5 Maximum value detection circuit 6 Stationary noise detection circuit 7 Average noise power spectrum calculation circuit 8 Subtraction circuit 9 Inverse frequency analysis circuits 10A, 10B Switch 11 Effective value calculating circuit 12 Effective value storing circuit 13 Threshold calculating circuit 14 Voice discriminating circuit 14A Power spectrum tilt calculating circuit 14B Signal type determining circuit 14C Parameter threshold storing unit 15 Voice discriminating flag storing circuit 16 Gain calculating circuit 17 Multiplication circuit 18 Voice average effective value calculation circuit 19 Compression ratio calculation circuit 20 Compression ratio smoothing circuit 21 Output terminal 101 Input terminal 102 Stationary noise discrimination circuit 103 Stationary noise suppression circuit 104 Effective value averaging circuit 105 Gain calculation circuit 106 Multiplication circuit 107 Output terminal 108 Audio identification circuit 109 Audio effective value averaging circuit

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ identification code FIG10L 9/14 D (58) Investigated field (Int.Cl. ⁷ , DB name) H03G 3/32 G10L 11/02 H04B 1 / 16 H04R 25/00

Claims (1)

(57) [Claims] [Claim 1] Linear prediction of a frame signal for each frame
Take out the linear residual signal and calculate its autocorrelation function
A stationary noise identifying means for identifying stationary noise based on the maximum value (CM) obtained by calculation ;
Input, and is regarded as stationary noise by the stationary noise discriminating means.
For frames that do not exist, output without suppression
Is stored in advance as stationary noise
Of the power spectrum for the number of past frames
The constant noise is reduced by using the value based on the average as the subtraction factor.
A stationary noise suppression means that suppresses and outputs, the stationary noise suppression
The effective value and the power value of the execution signal of the output signal from the
The slope of the vector S (f) and the maximum value of the autocorrelation function
(CM), the effective value is the threshold
When the value is smaller than the value, it is determined that the frame is a stationary noise frame after suppression.
And the effective value is equal to or greater than the threshold value, and the autocorrelation function
Between the maximum value (CM) and the slope of the power spectrum S (f)
When one of them is smaller than the threshold value,
It is determined that the frame is a normal noise frame, and the effective value is the threshold value.
The maximum value (CM) of the autocorrelation function and the power
Each of the slopes of the spectrum S (f) exceeds the threshold
When the voice signal is large, a voice identification means for determining whether the input signal is a voice by determining that the input signal is a voice , and a frame determined as a voice in the voice identification flag storage circuit.
The number of frames is determined by the number of frames determined to be non-stationary noise.
When receiving a large number of voices, the sound from the specified number in the past to the present
The average effective value of the frame determined to be voice is determined, and
From the number of frames determined to be audio in the separate flag storage circuit
Unsteady noise with a large number of frames determined to be unsteady noise
In the case of sound, a predetermined number of past frames determined to be stationary noise
RMS threshold based on the average of RMS
The value obtained by multiplying the constant by the constant
Continuous noise or non-stationary noise
If there is no sound with continuous sound input, it is determined to be stationary noise
RMS value based on the average of the RMS values of a predetermined number of past frames
The value obtained by multiplying the threshold by a constant smaller than 1 is the average effective value.
A sound effective value averaging means for calculating an average effective value of the sound, wherein the average effective value is a threshold of the effective value.
If the threshold is greater than the average value,
Divide by RMS, and the average RMS is the threshold of the RMS
If it is less than or equal to the value, the average RMS
Calculates the compression ratio by smoothing, and sets the target value to the effective value.
Gain calculation means for calculating a required gain by dividing by the threshold of
Multiplying means for multiplying an input signal by a compression ratio .