JPH09311696A - Automatic gain control device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To audibly receive objective sound alone at a proper sound volume without any distortion even if the dimension of voice of a speaker or the distance from the speaker to a microphone fluctuates. SOLUTION: After an input signal is processed by a noise suppressing means 100 so that background noise is suppressed, a threshold value serving as a boundary between compression and expansion is decided from an effective value of residual noise after noise suppression for a frame in which noise is determined. Subsequently, for finding a compression ratio in a certain frame, smoothing of effective values of frames including this frame for the last several seconds is carried out. The compression ratio is calculated from the smoothed effective value and multiplied by a necessary gain provided from the threshold value and a target mean effective value so as to be recomposed and outputted to each input frame.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic gain control device for audio equipment such as a loudspeaker device or an auxiliary listening device used by a hearing-impaired person.

[0002]

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 a hearing aid, even hearing impaired people who are not so inconvenient in one-on-one conversations may say that the distance to the speaker is long, such as a conference or a 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, the input sound pressure level decreases as the distance between the sound source and the sound collecting position increases. How much is reduced depends on the directivity of the sound source and the state of reverberation of the room. For example, the sound source is a point sound source (360) in a free sound field (a virtual space where there is no boundary such as walls, floors, and ceilings that reflects sound). Assuming a virtual sound source that emits sound evenly in any direction),
The sound pressure level is 6d when the distance between the sound source and the listening point doubles.
B is attenuated. In the actual room, not only the direct sound but also the reflected sound at the boundary is added, so it is not attenuated so much, but if it is 5 m away, the sound pressure level at the position of 1 m becomes 10
It attenuates by about dB.

In order to keep the input sound pressure level constant, there has conventionally been a device called a compressor, for example. It is a device capable of compressing the dynamic range of an input signal and increasing the gain for a relatively low level input.
However, this device observes the input level every hour and changes the hourly compression rate according to the value. Here, every hour is
It is the "minimum unit of time interval" of the system, which corresponds to "every sample" in digital terms, and is continuous "every moment" in analog. A large compression ratio is required to compensate for the sound pressure drop due to the distance attenuation having a relatively large level width, and as a result, the target voice is distorted and the clarity is deteriorated.

On the other hand, the hearing-impaired person adjusts the gain with the volume controller of the hearing aid. However, adjusting the gain every time the input sound pressure level changes (for example, the speaker changes) is very cumbersome, and some people may not operate the volume controller except when it is noisy and unbearable (for example, in a subway train). Many. As a result, "the voice of a person with a small voice and a person speaking far away is hard to hear".

Further, in the normal acoustic space, background noise is often present in addition to the intended sound. If an attempt is made to keep the target sound at a constant sound pressure level, the background noise level will vary depending on the size of the target sound, making it very difficult to hear.

In order to prevent this, it is possible to suppress only background noise before the gain control. As a conventional method therefor, there is a method such as "noise suppression device (Japanese Patent Application No. 8-14874)" proposed by the present inventor. This will be briefly described with reference to FIG.

In FIG. 5, reference numeral 200 denotes a noise suppressor, 2
Reference numeral 01 indicates the auditory weighting side, and 202 indicates the loss control side. 21 is an input signal terminal, 22 is a frequency analysis circuit, 23
Is a linear prediction analysis circuit, 24 is an autocorrelation analysis circuit, 25 is a maximum value selection circuit, 26 is a voice / non-voice discrimination circuit,
This output controls ON / OFF of switches 27A and 27B described later.

Reference numeral 28 is a noise spectrum characteristic calculation and storage circuit, in which auditory weighting is performed. 29 is a subtracting means, 30 is an inverse frequency analysis circuit, and the frequency analysis circuit 22
Performs operations in the reverse order of. The above is the auditory weighting side 20
Corresponds to 1.

Reference numeral 31 is an average noise level storage circuit, 32 is a loss control coefficient calculation circuit, 33 is an output signal calculation circuit, 34 is a calculating means, and 35 is an output signal terminal. The above corresponds to the loss control side 202.

Next, the operation will be described. The input signal is taken in from the signal input terminal 21, and the power spectrum S (f) and the phase information P (f) are obtained in the frequency analysis circuit 22 as in the conventional method. At the same time, the linear prediction analysis circuit 23 extracts a linear prediction residual signal (this is referred to as a residual signal here) from the input signal. The residual signal is sent to the autocorrelation analysis circuit 24, where the autocorrelation function (C
or [i]) is obtained. Then, the maximum value selection circuit 25 finds the peak value of the autocorrelation coefficient (which is the maximum value and is referred to as Rmax here), and the Rmax is used to determine the type of the input signal in the voice / non-voice identification circuit 26. Identify. That is, when Rmax is larger than a certain value (for example, Th), it is determined that the signal is a voice signal and below that is noise. This Rmax is often used as a feature amount that can well express the strength of the periodicity of the signal waveform. That is,
Of the input signals, most of the noise signals have random characteristics in the time or frequency domain, while most of the voice signals are voiced sounds, and the signals have periodicity. Therefore, it is effective to identify a signal section having no periodicity as noise. Of course, the voice signal contains unvoiced consonants, and accurate voice / non-voice discrimination cannot be made only by such feature amounts relating to periodicity. However, unvoiced consonants (eg, p, t, k, s, h, etc.) whose signal level is very low among various environmental noises.
It is very difficult to accurately detect (f etc.). Therefore, in the apparatus shown in FIG. 5, speech / non-speech discrimination is performed based on the idea of “discriminating a signal section that is not surely a speech signal and obtaining a long-term average spectral feature thereof”. .

In other words, it suffices to obtain "the average spectrum characteristic of a signal that is considered to be a noise signal", and by setting Rmax to a small value, a typical noise spectrum characteristic can be obtained. Is.

Now, the frequency-analyzed signal spectrum S
In (f), the switch 27A is closed only when it is identified as noise, and is stored in the noise spectrum characteristic calculation and storage circuit 28 as the noise spectrum S _ns (f). The update of the noise spectrum characteristic when the input signal is determined to be noise at time t is obtained by Expression (1).

[0014]

[Equation 1] Here, Snew (t, f) is the updated noise spectrum, Solid (f) is the noise spectrum before the update, St
(F) shows the noise spectrum when the input signal is identified as noise. Further, β is an average weighting coefficient.

W (f) as shown in equation (2) is used for noise suppression processing.

[0016]

[Equation 2] This W (f) has a function of reducing the above-mentioned "audibility" of the residual noise as much as possible, and the effect is further enhanced by using the equation (3). That is, W
Replacing f in (f) with i as a frequency point,

[0017]

(Equation 3) Is represented by Here, fc is a value corresponding to the frequency band of the input signal, B and K are weighting coefficients, and the larger this value, the larger the suppression amount. This auditory weighting coefficient naturally has a similar effect not only in the characteristic shown in the equation (3) but also in a pseudo noise average characteristic, and is not limited to the equation (3). Furthermore, the weighting factor B
Although K and K may be fixed to a value that is a device, the efficiency of noise suppression can be further increased by sequentially changing them adaptively according to the type and magnitude of noise.

By the above processing, the average spectrum of noise superimposed on the input signal is removed, and a new spectrum S'is obtained.
(F) is obtained from the subtracting means 29. This and the previously analyzed phase information P (f) are used for processing in the inverse frequency analysis circuit 30, and the signal waveform is obtained by returning from the frequency domain to the time domain. Since the frequency component of noise is suppressed in this signal waveform, only the voice signal remains.

[0019]

Although the conventional noise suppression method shown in FIG. 5 is extremely effective in suppressing background noise, when the loudness of the speaker or the distance from the microphone changes. Had a problem that it was difficult to respond appropriately.

It is an object of the present invention to deal with a change in loudness of a speaker's voice and a change in distance to a microphone in an audio device such as a loudspeaker communication device or an auxiliary hearing device used by a hearing-impaired person, and a steady darkness. It is to suppress noise so that only the target voice can be heard at an appropriate volume and without distortion.

[0021]

To achieve the above object, the present invention measures the level and background power spectrum of background noise, which is background noise in an input signal, and determines the background noise based on the average power spectrum of background noise. Noise suppression processing means for suppressing noise, threshold value calculating means for updating compression and expansion threshold values based on the background noise level, effective value calculating means for calculating an effective value of an input signal, and the input The effective value smoothing means for smoothing the effective value of the signal, the compression ratio calculating means for calculating the compression ratio from the smoothed effective value, and the gain necessary to obtain the average effective value of the target output It has a gain calculating means for calculating, and a multiplying means for multiplying the input signal by the compression ratio calculated by the compression ratio calculating means and the gain calculated by the gain calculating means and outputting the multiplied signal.

In the present invention, the input sound signal is divided into frame units, and a voice signal, a non-voice signal,
That is, it is determined whether it is noise. Here, a frame is a voice section cut out at a certain fixed time interval, and is basically the minimum unit for frequency analysis and calculation of an effective value. First, the fast Fourier transform FFT is performed for each frame.
A frequency analysis circuit such as (Fast Fourier Transform) is used to transform the signal into the frequency domain, and the power spectrum of the signal determined to be noise is stored in an appropriate number of frames, and the average value and average noise power spectrum are calculated. The average noise power spectrum is appropriately weighted and subtracted from the power spectrum of the signal of each frame converted into the frequency domain to obtain IFFT (Inverse Fast Fourier Transfor
Inverse frequency analysis circuit such as m) restores the time domain signal. Next, the threshold value that is the boundary between compression and expansion is determined from the effective value of a few seconds of the noise-suppressed signal (residual noise) of the frame determined to be noise. Next, in order to calculate the compression ratio of a certain frame, the effective value of the past several seconds including the frame is used to smooth the effective value. Calculate the compression ratio from the smoothed effective value and time constant,
Each input frame is multiplied with the required gain obtained from the threshold value and the target average effective value, recombined, and output.

[0023]

1 is a block diagram of an embodiment of the present invention. Reference numeral 1 denotes an input terminal, and the signal input from this is a signal for each frame divided by an appropriate time. First, 2 is a non-speech discrimination circuit which determines whether or not it is noise. As such a method, for example, there is a method utilizing the maximum value of the autocorrelation of the linear prediction residual signal, which focuses on the randomness of noise, as disclosed in Japanese Patent Application No. 8-14874 of FIG. That is, the linear prediction analysis circuit 23, the autocorrelation analysis circuit 24, and the maximum value selection circuit 2 of FIG.
5, the non-voice discrimination circuit 2 judges whether the result is non-voice.

3A and 3B are switches, 4 is a frequency analysis circuit such as an FFT, and 5 is an average noise spectrum calculation circuit, which operates during the period when the non-voice identification circuit 2 identifies non-voice. 6 is a subtraction circuit, 7 is an inverse frequency analysis circuit,
8 is an effective value calculating circuit, 9 is an effective value smoothing circuit, 10 is a compression ratio calculating circuit, 11 is a compression ratio smoothing circuit, and 12 is an average noise for calculating an average noise effective value from the output of the effective value calculating circuit 8. Effective value calculation circuit, 13 is a threshold value calculation circuit, 14
Is a gain calculation circuit, 15A and 15B are multiplication circuits, and 16 is an output end.

Next, the operation of the embodiment shown in FIG. 1 will be described.

The input signal from the input terminal 1 is sent to the frequency analysis circuit 4 in parallel with being applied to the non-speech discrimination circuit 2 and converted into the frequency domain. The power spectrum S (f) of the frame determined to be noise by the non-voice identification circuit 2 is sent to the average noise spectrum calculation circuit 5 by the action of the switch 3A and stored therein. An appropriate number of frames are stored in the past, the average value thereof is calculated, and the average value is used as the noise average power spectrum S _ns (f). This is multiplied by an appropriate weighting function W (f) as described below, and the subtraction circuit 6 subtracts it from the power spectrum S (f). Power spectrum S'of the signal from which noise is subtracted
Based on (f) and the phase information P (f) of the original signal, the inverse frequency analysis circuit 7 restores the time domain signal. The processing up to this point is the processing of the noise suppression processing unit 100.

FIG. 2 is an explanatory diagram of the weighting function W (f). The weighting function W (f) can be expressed by equation (4).

[0028]

(Equation 4) As shown in FIG. 2, W (f) increases the amount to be subtracted as the noise power spectrum increases. By doing so, the unerased portion in the low range where the noise power is large and the overdraw in the high range where the power is small are reduced.

Of the effective value rms of the input signal calculated by the effective value calculating circuit 8, the effective value of the past L frame determined to be noise is stored in the average noise effective value calculating circuit 12. This is considered to be the effective value of the residual noise left unerased by the noise suppression processing means 100. When the effective value Ns (N) newly determined as (residual) noise is stored, the oldest value Ns (NL) is deleted. Then, the average value <Ns (N)> thereof is calculated, and the threshold value calculation circuit 13 calculates <Ns (N)>.
s (N)> is multiplied by a constant of 1 or more to obtain the threshold value thd. This action is performed by switching the switches 3A and 3B only when the non-voice identification circuit 2 newly determines that the input signal is noise.

On the other hand, the effective value rms calculated by the effective value calculating circuit 8 is stored in the effective value smoothing circuit 9 continuously for the past M frames. When storing a new effective value rms (N), the oldest value rms (NM) is erased. This stored effective value sequence (rms (N−M + 1), ...,
The smoothed effective value <rms (N)> is obtained from rms (N-1), rms (N)) by the following procedure.

1 During utterance of voice: stored effective value sequence (rms (N−M + 1), ..., Rms (N−1),
From rms (N)), the values above the threshold value thd are averaged.

2 Speech rising portion: <rms from smoothed effective value <rms (N-1)> of the previous frame
When (N)> is extremely large, it is determined that the rising part of the voice or sudden sound is generated, and rms (N) is directly set to <rms
(N)>.

3 No voice: When K (<M) frames are continuously input for thd or less, it is determined that there is no voice,
Average over the rms columns of all K frames.

Using the smoothed effective value <rms (N)>,
The compression ratio calculation circuit 10 calculates the compression ratio p (N) according to the equation (5).

[0035]

(Equation 5) Further, in order to suppress a sudden change in the compression ratio p particularly at the rise and fall of voice, the compression ratio smoothing circuit 11
Smoothing is performed as in equation (6).

[0036]

(Equation 6) <P (N-1)> is the smoothing compression ratio of the previous frame,
C0 + C1 = 1.0, and the larger C1, the smoother the change. In the gain calculation circuit 14, assuming that the target average effective value of the output voice is d1, the gain G is calculated by the expression (7).

[0037]

(Equation 7) The multiplying circuits 15A and 15B multiply <p> and G by the input signal, and output the output signal from the output terminal 16.

FIG. 3 shows the input / output relationship in logarithmic logarithm. All inputs above the threshold value thd are target values d1
Is compressed and amplified, and below thd is expanded and attenuated.

FIG. 4 shows an example of the processing result obtained by the present invention. FIG. 4 (a) shows the original waveform, in which the latter signal is attenuated by 15 dB from the former signal, and stationary air conditioning noise is further attenuated by 10 dB and added to the latter signal. FIG. 4B shows the waveform of the processing result. The latter voice is amplified to almost the same level as the former voice, while the background noise is suppressed. FIG. 4C shows the total gain (<p>
× G) is a level display of time variation. From this figure, it can be seen that there is no small fluctuation and no large distortion occurs in the voice.

In the above-described embodiment of the present invention, each part of the block diagram is shown as a "circuit". However, since this may be executed by a program, in general,
Expressed as "means".

[0041]

As described above, according to the present invention, since the gain control is performed after the noise is suppressed, it is possible to set only the target sound to a constant sound pressure level. In addition, by automatically setting the threshold value and smoothing the effective value for calculating the compression ratio, an appropriate value can be quickly and smoothly adjusted according to the loudness of the speaker's voice and the distance from the microphone. Gain can be obtained. As a result, the distortion of the sound that occurs in the method of compressing the dynamic range using the conventional compressor does not occur, and clear sound without distortion can always be heard at a constant average sound pressure level.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is an explanatory diagram of a weighting function used in the present invention.

FIG. 3 is a diagram showing an input / output relationship diagram of the present invention.

FIG. 4 is a diagram showing a processing example of the present invention.

FIG. 5 is a block diagram showing a previously proposed noise suppression device.

[Explanation of symbols]

 1 Input Terminal 2 Non-Voice Discrimination Circuit 3A Switch 3B Switch 4 Frequency Analysis Circuit 5 Average Noise Spectrum Calculation Circuit 6 Subtraction Circuit 7 Inverse Frequency Analysis Circuit 8 Effective Value Calculation Circuit 9 Effective Value Smoothing Circuit 10 Compression Ratio Calculation Circuit 11 Compression Ratio Smoothing Circuit 12 average noise effective value calculation circuit 13 threshold value calculation circuit 14 gain calculation circuit 15A multiplication circuit 15B multiplication circuit 16 output terminal 100 noise suppression processing means

Claims (1)

[Claims] 1. An automatic gain adjustment device in an audio device such as a loudspeaker communication device or an auxiliary hearing device used by a hearing-impaired person, wherein the level of background noise that is background noise in an input signal and an average power spectrum are measured. Noise suppression processing means for suppressing background noise based on the average power spectrum of background noise; threshold value calculating means for updating compression and expansion threshold values based on the background noise level; An effective value calculating means for calculating a value, an effective value smoothing means for smoothing the effective value of the input signal, a compression ratio calculating means for calculating a compression ratio from the smoothed effective value, and a target output Gain calculation means for calculating the gain required to obtain the average effective value of, and the input signal multiplied by the compression ratio calculated by the compression ratio calculation means and the gain calculated by the gain calculation means, Signal Automatic gain control apparatus characterized by comprising: a multiplying means for, the.