CN100589183C - Digital auto gain control method and device - Google Patents

Abstract

This invention discloses one digital automatic gains control method and its device, which comprises the following steps: dividing the sound signals by fix interval into several frames; processing active sound test on each sound signal and dividing them into two types of sound frame and noise frame; according to current frame and next frame of sound types to processing slide on current sound samples gains between frames; processing frame slide on current each sound sample.

Description

Digital auto gain control method and device

Technical field

The present invention relates to the disposal route and the device of voice signal, relate in particular to a kind of gain is controlled to voice method and device.

Background technology

Encoding for voice signal, after the digital processing such as enhancing, generally can certain damage arranged voice signal.We can compensate this damage in Applied Digital automatic gain control (AGC).Digital AGC algorithm fundamental purpose is that the digital signal that receives is carried out certain processing, makes its amplitude relatively stable, thereby increases the subjective acoustical quality of sound.

The structure of common AGC mainly contains two kinds: feedback arrangement and do not have feedback arrangement.The feedback system method of adjustment is: the amplitude of the output signal after relatively the previous moment gain is adjusted and the amplitude R of our expectation with the variation of ride gain, if output then increases gain less than R, then reduce gain on the contrary.The cardinal principle of no feedback arrangement is: extract the amplitude envelops of input signal, set the gain that agc algorithm is controlled current sampling point according to this envelope.

For the digital AGC method of feedback arrangement, do not use the information of current input voice sampling point that gain is adjusted.This process be non-linear, the time feedback system that become, that depend on input.Total system is difficult to actual analysis, and the variation that gains between the stability and the size of gain, two adjacent spots is not easy control.Under the situation that sampling point varies in size, can cause the difference of gain adjustment simultaneously.The structure of not having feedback has been utilized the information of current input voice sampling point, voice signal is divided into several frames according to fixing duration, every frame comprises some sampling points, each sampling point has amplitude separately respectively, the control method of existing automatic gain is: 1, calculate each sampling point amplitude average absolute of present frame, with this mean value the average energy value as present frame; 2, with the energy value of the current broadcast sound volume correspondence of receiving end divided by this average energy value, obtain an enlargement factor, with the yield value of this enlargement factor as present frame, 3, this yield value multiply by the amplitude and the output of each sampling point of present frame, thereby reaches the purpose that the energy value of present frame is amplified to the energy value of current broadcast sound volume correspondence; Every frame of voice signal is all carried out the aforesaid operations process, thereby realizes the integral body of voice signal is amplified, and the energy value of each frame correspondence equals the energy value of current broadcast sound volume correspondence respectively in the voice signal behind feasible the amplification; Wherein, because the amplitude of sampling point has nothing in common with each other in each frame, therefore, the yield value that each frame calculated just has nothing in common with each other.Though the prior art can be by amplifying the purpose that reaches unified speech volume to voice signal, but, also there is following shortcoming: because the yield value of consecutive frame has nothing in common with each other, and, the average energy value of the energy value of first or last sampling point and speech frame is inequality in the speech frame, every frame adopts identical yield value to amplify, the joining place that can cause adjacent speech frame, the yield value cataclysm appears in amplitude sudden change place, thereby causing voice, instantaneous great changes will take place, make voice discontinuous, influence the subjective acoustical quality of voice.

Summary of the invention

Defective and deficiency at prior art exists the invention provides a kind of joining place at adjacent speech frame, can prevent that the digital auto gain control method and the device of yield value cataclysm from appearring in amplitude sudden change place.

For achieving the above object, the present invention by the following technical solutions: digital auto gain control method of the present invention comprises the steps:

Steps A is divided into some frames with voice signal by fixing duration;

Step B carries out movable voice to each frame voice signal and detects, and it is divided into two types of speech frame or noise frame;

Step C obtains the yield value of each voice sampling point of present frame, and in obtaining the process of yield value, according to the type of present frame, the interframe smoothing processing is carried out in the gain of the voice sampling point of present frame.

Among the described step C, in the process of the yield value of each voice sampling point that obtains present frame, further,, smoothing processing in the frame is carried out in the gain when the voice sampling point of pre-treatment according to the type of present frame.

In described digital auto gain control method, step B is specially: if present frame comprises voice signal information, then present frame is a speech frame, otherwise is noise frame.

In described digital auto gain control method, step C comprises:

Step C11 carries out envelope detected to present frame, obtains the envelope information of present frame;

Step C12 according to the type and the envelope information of present frame voice signal, adjusts the gain of present frame voice sampling point.

In described digital auto gain control method, step C comprises:

Step C21 carries out envelope detected to present frame, obtains the envelope information of present frame;

Step C22 according to the voice signal and the envelope information of the former frame of present frame and present frame, adjusts the gain of present frame voice sampling point.

In described digital auto gain control method, step C11, C21 comprise the steps:

Step 1 is chosen a voice sampling point in the present frame as the current speech sampling point, the range value x (n) of current speech sampling point is asked thoroughly deserve | x (n) |;

Step 2 is passed through formula

Make the variation of envelope-tracking input signal; α is a predetermined constant,

M is used for carrying out the level and smooth window length of envelope;

Step 3 at current sampling point with count forward in the scope of c (n) of N voice sampling points, is chosen the c (n) of maximum, i.e. x _Max(n)=max[c (n-N) ..., c (n)];

Step 4 according to the maximal value of choosing, calculates the envelope value of current speech sampling point:

e _max(n)＝αx _max(n)+(1-α)e _max(n-1)

Step 5 at the current speech sampling point with count forward in the scope of M voice sampling points, is averaged after the envelope value addition to all voice sampling points

In described digital auto gain control method, step C12 is specially:

If present frame is a noise frame, the gain of present frame is made as 1, promptly for all n, g (n)=1 does not amplify each voice sampling point of present frame yet and not to dwindle;

If present frame is a speech frame, the yield value of each voice sampling point of present frame adjusted to seamlessly transit yield value g (n),

$g\left(n\right)={\left(\frac{\mathrm{threshold}}{e_{\mathrm{ave}}\left(n\right)}\right)}^{\mathrm{\θ}}$

θ represent one with the relevant constant of adjusting of effect, generally satisfy 0＜θ≤1, threshold represents predefined threshold value.

In described digital auto gain control method, step C22 is specially:

If the former frame of present frame and present frame is noise frame, the gain of present frame is made as 1, present frame is not amplified yet do not dwindle;

If former frame is a noise frame, present frame is a speech frame, noise frame is to the speech frame transition, this moment, the yield value with first voice sampling point of present frame was taken as 1, the yield value of last voice sampling point is for seamlessly transitting yield value, the yield value of all the other each voice sampling points by 1 linear increment to seamlessly transitting yield value;

If former frame is a speech frame, present frame is a noise frame, speech frame is to the transition of noise frame, be taken as the yield value of first voice sampling point of present frame and seamlessly transit yield value this moment, the yield value of last voice sampling point is 1, and the yield value of all the other each voice sampling points is by seamlessly transitting yield value linear decrease to 1;

If former frame and present frame are speech frame, the yield value of each voice sampling point of present frame all is taken as and seamlessly transits yield value.

In described digital auto gain control method, the described yield value g (n) that seamlessly transits,

$g\left(n\right)={\left(\frac{\mathrm{threshold}}{e_{\mathrm{ave}}\left(n\right)}\right)}^{\mathrm{\θ}}---\left(5\right)$

θ represent one with the relevant constant of adjusting of effect, generally satisfy 0＜θ≤1, threshold represents predefined threshold value.

In described digital auto gain control method, smoothing processing is specially in the described frame: the gain of each voice sampling point of present frame is adjusted into level and smooth yield value g in the frame _Ave(n):

$g_{\mathrm{ave}}\left(n\right)=\frac{1}{M}\underset{i=0}{\overset{M}{\mathrm{\Σ}}}g(n-i).$

In described digital auto gain control method, if level and smooth yield value is greater than predetermined threshold in the frame of current sampling point, then level and smooth yield value is made as predetermined threshold in the frame.

In described digital auto gain control method, also comprise step D: the input voice range value of N voice sampling point multiply by the interior level and smooth yield value of frame and obtains exporting voice before the current speech sampling point.

A kind of automatic gain control equipment comprises:

Voice signal is cut apart module, is used for voice signal is divided into some frames by fixing duration;

It is characterized in that also comprising:

The movable voice detection module is cut apart each frame after the module segmentation with described voice signal, divides into two types of speech frame or noise frame;

Interframe smoothing processing module is obtained the yield value of each voice sampling point of present frame, and in obtaining the process of yield value, according to the type of present frame, the interframe smoothing processing is carried out in the gain of the voice sampling point of present frame.

In described automatic gain control equipment, also comprise: smoothing processing module in the frame, according to the type of present frame, smoothing processing in the frame is carried out in the gain when the voice sampling point of pre-treatment.

In described automatic gain control equipment, described interframe smoothing processing module comprises:

The envelope detected module is carried out envelope detected to present frame, obtains the envelope information of present frame;

Gain regulation module according to the type and the envelope information of present frame voice signal, is adjusted the gain of present frame voice sampling point.

In described automatic gain control equipment, described gain regulation module according to the voice signal and the envelope information of the former frame of present frame and present frame, is adjusted the gain of present frame voice sampling point.

In described automatic gain control equipment, described envelope detected module comprises:

Range value is asked form unit, is used for the range value x (n) of present frame current speech sampling point is asked absolute value, obtains | x (n) |;

Envelope smoothing processing unit, according to | x (n) |, utilize formula

Make the variation of envelope-tracking input speech signal;

Maximal value is chosen the unit, at the current speech sampling point with count forward in the scope of c (n) of N voice sampling points, chooses the range value c (n) of maximum, i.e. x _Max(n)=max[c (n-N) ..., c (n)];

The envelope value computing unit, the amplitude peak value c (n) according to choosing utilizes formula e _Max(n)=α x _Max(n)+(1-α) e _Max(n-1) envelope value of calculating current speech sampling point;

At the current speech sampling point with count forward in the scope of M voice sampling points, average after the envelope value addition to all voice sampling points in envelope amplitude smoothing processing unit

In described automatic gain control equipment, described gain regulation module comprises:

Seamlessly transit the yield value computing unit, according to formula

Calculating seamlessly transits yield value.

Control module is adjusted in gain, adjusts the gain of present frame voice sampling point.

In described automatic gain control equipment, described gain adjustment control module gains to present frame voice sampling point and carries out following adjustment:

If present frame is a noise frame, the gain of present frame is made as 1, promptly for all n, g (n)=1 does not amplify each voice sampling point of present frame yet and not to dwindle;

If present frame is a speech frame, the yield value of each voice sampling point of present frame is adjusted to seamlessly transit yield value.

In described automatic gain control equipment, described gain adjustment control module gains to present frame voice sampling point and carries out following adjustment:

If the former frame of present frame and present frame is noise frame, the gain of present frame is made as 1, present frame is not amplified yet do not dwindle;

If former frame is a noise frame, present frame is a speech frame, noise frame is to the speech frame transition, this moment, the yield value with first voice sampling point of present frame was taken as 1, the yield value of last voice sampling point is for seamlessly transitting yield value, the yield value of all the other each voice sampling points by 1 linear increment to seamlessly transitting yield value;

If former frame is a speech frame, present frame is a noise frame, speech frame is to the transition of noise frame, be taken as the yield value of first voice sampling point of present frame and seamlessly transit yield value this moment, the yield value of last voice sampling point is 1, and the yield value of all the other each voice sampling points is by seamlessly transitting yield value linear decrease to 1;

If former frame and present frame are speech frame, the yield value of each voice sampling point of present frame all is taken as and seamlessly transits yield value.

In described automatic gain control equipment, the smoothing processing module comprises in the described frame:

Level and smooth yield value computing unit in the frame is according to formula

Calculate level and smooth yield value in the frame;

Gain adjusting unit in the frame is adjusted into level and smooth yield value g in the frame with the gain of each voice sampling point of present frame _Ave(n).

In described automatic gain control equipment, also comprise: the voice output module, according to the input voice range value of N voice sampling point before the current speech sampling point, multiply by the interior level and smooth yield value of frame and obtain exporting voice, and export this voice.

Digital auto gain control method of the present invention is divided into two types of speech frame or noise frame with each frame voice signal; Joining place between dissimilar speech frames carries out the interframe smoothing processing.For the transition of noise frame to speech frame, gain the present invention of each voice sampling point is made as 1 in the noise frame, the gain of first voice sampling point is 1 in the speech frame, the gain of last voice sampling point is for seamlessly transitting yield value, the yield value of each middle voice sampling point by 1 linear increment to seamlessly transitting yield value.Prevented from like this to the joining place of speech frame transition the gain cataclysm to take place in noise frame.For the transition of speech frame to noise frame, the yield value of first voice sampling point is for seamlessly transitting yield value in the speech frame, and the yield value of last voice sampling point is 1, and the yield value of each middle voice sampling point is by seamlessly transitting yield value linear attenuation to 1.Prevented from like this to the joining place of noise frame transition the gain cataclysm to take place at speech frame.For two continuous speech frames, the yield value of each voice sampling point of present frame is adjusted into and seamlessly transits yield value.Prevented from like this to the joining place of speech frame transition the gain cataclysm to take place at speech frame.

Digital auto gain control method of the present invention not only can prevent the yield value cataclysm at the joining place of adjacent speech frame, and can in speech frame between adjacent voice sampling point, prevent the yield value cataclysm.With the gain of each voice sampling point in the present frame, adjust to level and smooth yield value in the frame by seamlessly transitting yield value.Level and smooth yield value is that the yield value that seamlessly transits to several continuous speech sampling points is averaged and obtains in the frame.Make gain transitions smooth between adjacent voice sampling point like this.

Description of drawings

Fig. 1 is the process flow diagram of auto gain control method embodiment 1 of the present invention;

Fig. 2 is the process flow diagram of auto gain control method embodiment 2 of the present invention;

Fig. 3 is the block scheme of automatic gain control equipment of the present invention;

Fig. 4 is the block diagram that seamlessly transits the yield value computing unit.

Embodiment

In order to solve the joining place of adjacent speech frame, the problem of yield value cataclysm appears in amplitude sudden change place, and auto gain control method of the present invention adopts the interframe antialiasing.Below in conjunction with the drawings and specific embodiments the present invention is described in further detail:

Embodiment 1:

With reference to accompanying drawing 1, the present embodiment digital auto gain control method comprises:

Step 1, voice signal that will be to be imported carry out the branch frame to be handled.

Be about to voice signal to be imported and be divided into some frames according to fixing duration (as 20ms), every frame comprises several voice sampling points, and each voice sampling point has amplitude separately respectively.

Step 2, movable voice detection module (VAD detection module) detects the voice signal of each frame input, divides into two types of speech frame and noise frame.

Select a frame voice signal as present frame.Present frame detects through the movable voice detection module, obtains the information vad_flag whether present frame comprises voice signal, if comprise voice signal vad_flag=1, otherwise vad_flag=0.The main effect of movable voice detection module is a noise frame of distinguishing speech frame and no voice, and purpose is: only speech frame is adjusted when gain is adjusted below, noise frame is not amplified yet do not dwindle.Noise when avoiding when gain is adjusted, amplifying no voice.

Step 3 can be carried out simultaneously with step 2, and present frame is carried out envelope detected, obtains the envelope information of present frame.

Envelope detected comprises preprocessing part and envelope detected part.

Described preprocessing part concrete steps are as follows:

Step 1 is chosen a voice sampling point in the present frame as current sampling point, the range value x (n) of current sampling point is asked thoroughly deserve | x (n) | and, wherein n is the sequence number of voice sampling point.

Step 2, in order to make envelope can follow the tracks of the variation of input signal, incite somebody to action | x (n) | do smoothing processing:

$c\left(n\right)=\max \left[\right|x\left(n\right)|,\frac{\left|x\left(n\right)\right|-\mathrm{\β}{e}_{\mathrm{ave}}(n-1)}{1-\mathrm{\β}}]---\left(1\right)$

Step 3 is that several forward N the voice sampling points of starting point are in the scope of terminal point with current sampling point, chooses the maximal value of the corresponding c of voice sampling point (n)

x _max(n)＝max[c(n-N)，…，c(n)] (2)；

Wherein, the span of N is 10～40, can value be 20 in the present embodiment, is several forward 20 the voice sampling points of starting point with current sampling point promptly.Obviously, when handling top N-1 voice sampling point of present frame, in N several forward the voice sampling points, be the voice sampling point of former frame with some voice sampling point.

Be the envelope detected part below:

Step 4 is according to the amplitude peak value x that chooses _Max(n) and the envelope value e of the previous sampling point of current sampling point _Max(n-1), calculate the envelope value of current sampling point:

e _max(n)＝αx _max(n)+(1-α)e _max(n-1) (3)

In order to obtain level and smooth envelope amplitude, through following smoothing processing:, average after the envelope value addition to all voice sampling points at current sampling point with forward in the scope of several M voice sampling points:

$e_{\mathrm{ave}}\left(n\right)=\frac{1}{M}\underset{i=0}{\overset{M}{\mathrm{\Σ}}}{e}_{\max }(n-i)---\left(4\right)$

α is a predetermined constant in the above formula 1～4, can adjust as requested; General, at x _Max(n)＞e _Max(n-1) the α time=0.2, otherwise α=0.01.

$\mathrm{\β}=\frac{1}{M}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}{(1-\mathrm{\α})}^{N+1-i};$

M is used for carrying out the level and smooth window length of envelope, for integer and generally greater than 10, can get M=20 in the present embodiment.

Step 4 is determined the sampling point gain according to the type and the envelope information of present frame voice signal.

In this step, at different classification, we adopt diverse ways to handle.Overall thought is:

0 expression present frame is a noise frame, the gain of present frame can be made as 1, and promptly for all n, g (n)=1 does not amplify present frame yet and not dwindle.

1 expression present frame is a speech frame, and need carry out the interframe smoothing processing this moment to the yield value of each sampling point.

Interframe gain-smoothing processing procedure in the time of will describing present frame below in detail and be speech frame.

Steps A 1 is by the yield value g (n) of each voice sampling point in formula (5) the calculating present frame.

$g\left(n\right)={\left(\frac{\mathrm{threshold}}{e_{\mathrm{ave}}\left(n\right)}\right)}^{\mathrm{\θ}}---\left(5\right)$

θ represent one with the relevant constant of adjusting of effect, generally satisfy 0＜θ≤1, best value is 0.5, θ is big more, and the speech energy that obtains is smooth more.The envelope of limiting case θ=1 o'clock output voice will be held constant at threshold.Threshold represents predefined threshold value, and general value is 16384, and with reduced, and the big more multiple that dwindles of amplitude is also big more greater than the voice of threshold; Voice less than threshold will be exaggerated, and the more for a short time multiple that is exaggerated of the amplitude of voice sampling point is big more.

Step 5 in order to prevent in the present frame that the yield value cataclysm takes place between adjacent voice sampling point, is carried out smoothing processing in the frame to the gain of each voice sampling point of present frame below.Promptly the yield value g (n) that calculates according to steps A 2 calculates level and smooth yield value g in the frame _Ave(n):

$g_{\mathrm{ave}}\left(n\right)=\frac{1}{M}\underset{i=0}{\overset{M}{\mathrm{\Σ}}}g(n-i)---\left(6\right);$

In addition, excessive in order to prevent that ground unrest is exaggerated in step 4, also can set a maximum multiple MAX_GAIN who allows amplification, MAX_GAIN is an empirical constant, is MAX_GAIN=3 as best value.If the gain of current sampling point is made as MAX_GAIN greater than the gain of the so current sampling point of MAX_GAIN.

Step 6 utilizes the yield value that obtains to multiply by the current speech sampling point input voice range value of N voice sampling point before, obtains final digital AGC adjustment output voice afterwards.

Be formulated as: y (n)=g _Ave(n) * x (n-N).

The method that embodiment 1 provides by the envelope detected step, can detect the envelope of voice signal accurately and make its level and smooth variation.And, when obtaining yield value, by smoothing processing, prevent that yield value from interframe and frame cataclysm taking place to yield value according to envelope value.Simultaneously, present embodiment has adopted different disposal routes at noise frame and speech frame, can prevent that noise frame is exaggerated.Therefore adopt the digital gain control method of present embodiment, can obtain the voice signal that voice link up and do not have the good quality of instantaneous cataclysm phenomenon.

The present invention provides the device of realizing this method simultaneously.With reference to accompanying drawing 3, automatic gain control equipment of the present invention comprises:

Voice signal is cut apart module, is used for voice signal is divided into some frames by fixing duration;

The movable voice detection module is cut apart each frame after the module segmentation with described voice signal, divides into two types of speech frame or noise frame;

Interframe smoothing processing module is obtained the yield value of each voice sampling point of present frame, and in obtaining the process of yield value, according to the type of present frame, the interframe smoothing processing is carried out in the gain of the voice sampling point of present frame.

Smoothing processing module in the frame according to the type of present frame, is carried out smoothing processing in the frame to the gain when the voice sampling point of pre-treatment.

The voice output module according to the input voice range value of N voice sampling point before the current speech sampling point, multiply by the interior level and smooth yield value of frame and obtains exporting voice, and export this voice.

Described interframe smoothing processing module comprises:

The envelope detected module is carried out envelope detected to present frame, obtains the envelope information of present frame;

Gain regulation module according to the type and the envelope information of present frame voice signal, is adjusted the gain of present frame voice sampling point.

With reference to accompanying drawing 4, described envelope detected module comprises:

Range value is asked form unit, is used for the range value x (n) of present frame current speech sampling point is asked absolute value, obtains | x (n) |;

Envelope smoothing processing unit, according to | x (n) |, utilize formula Make the variation of envelope-tracking input speech signal;

Maximal value is chosen the unit, at the current speech sampling point with count forward in the scope of c (n) of N voice sampling points, chooses the range value c (n) of maximum, i.e. x _Max(n)=max[c (n-N) ..., c (n)];

The envelope value computing unit, the amplitude peak value c (n) according to choosing utilizes formula e _Max(n)=α x _Max(n)+(1-α) e _Max(n-1) envelope value of calculating current speech sampling point;

At the current speech sampling point with count forward in the scope of M voice sampling points, average after the envelope value addition to all voice sampling points in envelope amplitude smoothing processing unit

Described gain regulation module comprises:

Seamlessly transit the yield value computing unit, according to formula

Calculating seamlessly transits yield value.

Control module is adjusted in gain, adjusts the gain of present frame voice sampling point.

Gain adjustment control module gains to present frame voice sampling point and carries out following adjustment:

If present frame is a noise frame, the gain of present frame is made as 1, promptly for all n, g (n)=1 does not amplify each voice sampling point of present frame yet and not to dwindle;

If present frame is a speech frame, the yield value of each voice sampling point of present frame is adjusted to seamlessly transit yield value.

The smoothing processing module comprises in the described frame:

Level and smooth yield value computing unit in the frame is according to formula

Calculate level and smooth yield value in the frame;

Gain adjusting unit in the frame is adjusted into level and smooth yield value g in the frame with the gain of each voice sampling point of present frame _Ave(n).

Embodiment 2:

With reference to accompanying drawing 2, the present embodiment digital auto gain control method comprises:

Step 100 (not shown) is carried out the branch frame with voice signal to be imported and is handled.

Be about to voice signal to be imported and be divided into some frames according to fixing duration (as 20ms), every frame comprises several voice sampling points, and each voice sampling point has amplitude separately respectively.

Step 101, movable voice detection module (VAD detection module) detects the voice signal of each frame input, divides into two types of speech frame and noise frame, and the type of each frame voice signal is preserved.

Select a frame voice signal as present frame.Present frame detects through the movable voice detection module, obtains the information vad_flag whether present frame comprises voice signal, if comprise voice signal vad_flag=1, otherwise vad_flag=0.The main effect of movable voice detection module is a noise frame of distinguishing speech frame and no voice, and purpose is: only speech frame is adjusted when gain is adjusted below, noise frame is not amplified yet do not dwindle.Noise when avoiding when gain is adjusted, amplifying no voice.

Step 102 can be carried out simultaneously with step 101, and present frame is carried out envelope detected, obtains the envelope information of present frame.

Envelope detected comprises preprocessing part and envelope detected part.

Described preprocessing part concrete steps are as follows:

Step 1 is chosen a voice sampling point in the present frame as current sampling point, the range value x (n) of current sampling point is asked thoroughly deserve | x (n) |; Wherein n is the sequence number of voice sampling point.

Step 2, in order to make envelope can follow the tracks of the variation of input signal, incite somebody to action | x (n) | do smoothing processing:

$c\left(n\right)=\max \left[\right|x\left(n\right)|,\frac{\left|x\left(n\right)\right|-\mathrm{\β}{e}_{\mathrm{ave}}(n-1)}{1-\mathrm{\β}}]---\left(1\right)$

Step 3 is that several forward N the voice sampling points of starting point are in the scope of terminal point with current sampling point, chooses the maximal value x of the corresponding c of voice sampling point (n) _Max(n)=max[c (n-N) ..., c (n)] and (2);

Wherein, the span of N is 10～40, can value be 20 in the present embodiment, is several forward 20 the voice sampling points of starting point with current sampling point promptly.Obviously, when handling top N-1 voice sampling point of present frame, in N several forward the voice sampling points, be the voice sampling point of former frame with some voice sampling point.

Be the envelope detected part below:

Step 4 is according to the amplitude peak value x that chooses _Max(n) and the envelope value e of the previous sampling point of current sampling point _Max(n-1),

Calculate the envelope value of current sampling point:

e _max(n)＝αx _max(n)+(1-α)e _max(n-1) (3)

In order to obtain level and smooth envelope amplitude, through following smoothing processing:, average after the envelope value addition to all voice sampling points at current sampling point with forward in the scope of several M voice sampling points:

$e_{\mathrm{ave}}\left(n\right)=\frac{1}{M}\underset{i=0}{\overset{M}{\mathrm{\Σ}}}{e}_{\max }(n-i)---\left(4\right)$

α is a predetermined constant in the above formula 1～4, can adjust as requested; General, at x _Max(n)＞e _Max(n-1) the α time=0.2, otherwise α=0.01.

$\mathrm{\β}=\frac{1}{M}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}{(1-\mathrm{\α})}^{N+1-i};$

M is used for carrying out the level and smooth window length of envelope, for integer and generally greater than 10, can get M=20 in the present embodiment.

Step 103, the type of the former frame voice signal of the present frame that taking-up step 101 is preserved.

In the present embodiment, can adopt the method for time-delay 1 frame, the type of output former frame voice signal.

Step 104 is utilized the type of present frame and its former frame voice signal, and present frame is divided into 4 classes.

To import voice according to the vad_flag of the vad_flag of former frame and present frame and be divided into four classes, and represent that with xy wherein x will represent the vad_flag of former frame, y represents the vad_flag of present frame.Four classes are respectively 00,01, and 10,11.

Step 105 is determined the sampling point gain according to different classification and envelope information.

In this step, at different classification, we adopt diverse ways to handle.Overall thought is:

00 former frame that is expressed as present frame and present frame is noise frame, the gain of present frame can be made as 1, present frame is not amplified yet do not dwindle.

01 expression former frame is a noise frame, and present frame is a speech frame, and noise frame is to the speech frame transition, and the gain linearity that can make each sampling point of present frame this moment is from 1 gain that is incremented to last sampling point of present frame.

10 expression former frame are speech frame, and present frame is a noise frame, and at this moment speech frame can make the gain of each sampling point of present frame decay to 1 gradually to the transition of noise frame.

11 expression former frame and present frame are speech frame, need carry out smoothing processing to the yield value of each sampling point.

To describe the gain-smoothing processing procedure of this four classes situation below in detail.

(1), when being 00, the former frame that is expressed as present frame and present frame is noise frame, as previously mentioned, the gain of present frame is made as 1, promptly for all n, and g (n)=1.

(2), when being 01, former frame is a noise frame, present frame is a speech frame, this moment, the yield value with first voice sampling point of present frame was taken as 1, the yield value of last voice sampling point is for seamlessly transitting yield value, the yield value of all the other each voice sampling points by 1 linear increment to seamlessly transitting yield value.Its processing procedure is specific as follows:

Steps A 1 is by the yield value g (n) of last voice sampling point in formula (5) the calculating present frame.

$g\left(n\right)={\left(\frac{\mathrm{threshold}}{e_{\mathrm{ave}}\left(n\right)}\right)}^{\mathrm{\θ}}---\left(5\right)$

θ represent one with the relevant constant of adjusting of effect, generally satisfy 0＜θ≤1, best value is 0.5, θ is big more, and the speech energy that obtains is smooth more.The envelope of limiting case θ=1 o'clock output voice will be held constant at threshold.Threshold represents predefined threshold value, and general value is 16384, and with reduced, and the big more multiple that dwindles of amplitude is also big more greater than the voice of threshold; Voice less than threshold will be exaggerated, and the more for a short time multiple that is exaggerated of the amplitude of voice sampling point is big more.

Steps A 2, get the yield value that calculates last sampling point of the present frame that obtains in the steps A 1 and seamlessly transit yield value as it, the yield value of getting first sampling point of present frame is 1, yield value to other sampling point in the present frame carries out the yield value that linearization process obtains each sampling point in order, and makes the seamlessly transit yield value of yield value from 1 linear increment to a last sampling point of each sampling point in the present frame.

The process of described linearization process can be the arithmetic progression that order increases progressively for the yield value that makes each voice sampling point in the present frame.

This step is the joining place to adjacent speech frame, carries out the interframe smoothing processing.

Steps A 3 in order to prevent in the present frame that the yield value cataclysm takes place between adjacent voice sampling point, is carried out smoothing processing in the frame to the gain of each voice sampling point of present frame below.Promptly the yield value g (n) that calculates according to steps A 2 calculates level and smooth yield value g in the frame _Ave(n):

$g_{\mathrm{ave}}\left(n\right)=\frac{1}{M}\underset{i=0}{\overset{M}{\mathrm{\Σ}}}g(n-i)---\left(6\right);$

(3) when being 10, be that former frame is a speech frame, present frame is a noise frame, be taken as the yield value of first voice sampling point of present frame and seamlessly transit yield value this moment, the yield value of last voice sampling point is 1, and the yield value of all the other each voice sampling points is by seamlessly transitting yield value linear decrease to 1.

Its processing procedure is roughly the same with the situation that is categorized as at 01 o'clock, and same execution in step A1 utilizes formula 5 to try to achieve the yield value of first sampling point of present frame; Carrying out A3 carries out smoothly the gain of trying to achieve, when difference is execution in step A2, be that yield value with first voice sampling point of present frame is taken as first sampling point yield value of present frame that steps A 1 is calculated gained, and the yield value of last voice sampling point is taken as 1, the arithmetic progression that successively decreases is in order then got the yield value of other voice sampling point.

Also should be noted that in this step the yield value of last voice sampling point of desirable former frame as seamlessly transitting yield value, then with the yield value of each sampling point in present frame linear decrease to 1 in order.So just need not to have carried out the calculating of steps A 1 again.

(4) when being 11, promptly former frame and present frame are speech frame, and the yield value of each sampling point of present frame all should be taken as and seamlessly transit yield value at this moment.

Its processing procedure is with the difference that is categorized as at 01 o'clock, utilizes formula 5 to try to achieve the yield value of each sampling point at execution in step A1; Promptly obtained the yield value that seamlessly transits of each voice sampling point behind the execution A3.And do not need to carry out A2.

In addition, excessive in order to prevent that ground unrest is exaggerated in step 105, also can set a maximum multiple MAX_GAIN who allows amplification, MAX_GAIN is an empirical constant, is MAX_GAIN=3 as best value.If the gain of current sampling point is made as MAX_GAIN greater than the gain of the so current sampling point of MAX_GAIN.

Step 106 utilizes the yield value that obtains to multiply by the current speech sampling point input voice range value of N voice sampling point before, obtains final digital AGC adjustment output voice afterwards.

Be formulated as: y (n)=g _Ave(n) * x (n-N).

In the present embodiment, introduce additional noise in order noise not to be adjusted, adding and voice activity detection (vad) module are in order to classify to the input voice in system architecture, and be divided into 4 classes in conjunction with the characteristic of former frame voice and handle respectively, make the adjustment of gain more level and smooth, the voice that finally obtain do not have audible distortion, and make the relative size of input voice concern the low voice segments of raising energy under the condition that remains unchanged, and reduce the too high voice segments of energy.

In a word, through above-mentioned processing, the digital AGC algorithm that present embodiment provides is handled the signal of high s/n ratio, can effectively adjust small-signal and excessive signal, make it in reasonable range, and whole adjustment process is not introduced additional voice audible distortion.

The present invention provides the device of realizing this method simultaneously.With reference to accompanying drawing 3, automatic gain control equipment of the present invention comprises:

Voice signal is cut apart module, is used for voice signal is divided into some frames by fixing duration;

The movable voice detection module is cut apart each frame after the module segmentation with described voice signal, divides into two types of speech frame or noise frame;

Interframe smoothing processing module is obtained the yield value of each voice sampling point of present frame, and in obtaining the process of yield value, according to the type of present frame, the interframe smoothing processing is carried out in the gain of the voice sampling point of present frame.

Smoothing processing module in the frame according to the type of present frame, is carried out smoothing processing in the frame to the gain when the voice sampling point of pre-treatment.

The voice output module according to the input voice range value of N voice sampling point before the current speech sampling point, multiply by the interior level and smooth yield value of frame and obtains exporting voice, and export this voice.

Described interframe smoothing processing module comprises:

The envelope detected module is carried out envelope detected to present frame, obtains the envelope information of present frame;

Described gain regulation module according to the voice signal and the envelope information of the former frame of present frame and present frame, is adjusted the gain of present frame voice sampling point.

With reference to accompanying drawing 4, described envelope detected module comprises:

Range value is asked form unit, is used for the range value x (n) of present frame current speech sampling point is asked absolute value, obtains | x (n) |;

Envelope smoothing processing unit, according to | x (n) |, utilize formula

Make the variation of envelope-tracking input speech signal;

Maximal value is chosen the unit, at the current speech sampling point with count forward in the scope of c (n) of N voice sampling points, chooses the range value c (n) of maximum, i.e. x _Max(n)=max[c (n-N) ..., c (n)];

The envelope value computing unit, the amplitude peak value c (n) according to choosing utilizes formula e _Max(n)=α x _Max(n)+(1-α) e _Max(n-1) envelope value of calculating current speech sampling point;

At the current speech sampling point with count forward in the scope of M voice sampling points, average after the envelope value addition to all voice sampling points in envelope amplitude smoothing processing unit

Described gain regulation module comprises:

Seamlessly transit the yield value computing unit, according to formula Calculating seamlessly transits yield value.

Control module is adjusted in gain, adjusts the gain of present frame voice sampling point.

Gain adjustment control module gains to present frame voice sampling point and carries out following adjustment:

If the former frame of present frame and present frame is noise frame, the gain of present frame is made as 1, present frame is not amplified yet do not dwindle;

If former frame is a noise frame, present frame is a speech frame, noise frame is to the speech frame transition, this moment, the yield value with first voice sampling point of present frame was taken as 1, the yield value of last voice sampling point is for seamlessly transitting yield value, the yield value of all the other each voice sampling points by 1 linear increment to seamlessly transitting yield value;

If former frame is a speech frame, present frame is a noise frame, speech frame is to the transition of noise frame, be taken as the yield value of first voice sampling point of present frame and seamlessly transit yield value this moment, the yield value of last voice sampling point is 1, and the yield value of all the other each voice sampling points is by seamlessly transitting yield value linear decrease to 1;

If former frame and present frame are speech frame, the yield value of each voice sampling point of present frame all is taken as and seamlessly transits yield value.

The smoothing processing module comprises in the described frame:

Level and smooth yield value computing unit in the frame is according to formula

Calculate level and smooth yield value in the frame; Gain adjusting unit in the frame is adjusted into level and smooth yield value g in the frame with the gain of each voice sampling point of present frame _Ave(n).

Claims (17)

1, a kind of digital auto gain control method comprises:

Steps A is divided into some frames with voice signal by fixing duration;

It is characterized in that also comprising:

Step B carries out movable voice to each frame voice signal and detects, and it is divided into two types of speech frame or noise frame;

Step C obtains the yield value of each voice sampling point of present frame, and in obtaining the process of yield value, according to the type of present frame, the interframe smoothing processing is carried out in the gain of the voice sampling point of present frame;

Described step C comprises:

Present frame is carried out envelope detected, obtain the envelope information of present frame;

According to the type and the envelope information of present frame voice signal, adjust the gain of present frame voice sampling point;

Perhaps

According to the voice signal and the envelope information of the former frame of present frame and present frame, adjust the gain of present frame voice sampling point.

2, digital auto gain control method according to claim 1 is characterized in that also comprising step D: according to the type of present frame, smoothing processing in the frame is carried out in the gain when the voice sampling point of pre-treatment.

3, digital auto gain control method according to claim 1 is characterized in that step B is specially: if present frame comprises the information of voice signal, then present frame is a speech frame, otherwise is noise frame.

4, digital auto gain control method according to claim 1 is characterized in that described present frame being carried out envelope detected, and the step that obtains the envelope information of present frame comprises:

Step 1, a voice sampling point choosing present frame be as the current speech sampling point, the range value x (n) of current speech sampling point asked thoroughly deserve | x (n) | and, n is the sequence number of voice sampling point;

Step 2 is passed through formula $c\left(n\right)=\max \left[\right|x\left(n\right)|,\frac{\left|x\left(n\right)\right|-{\mathrm{\βe}}_{\mathrm{ave}}(n-1)}{1-\mathrm{\β}}]$ Make the variation of envelope-tracking input signal; α is a predetermined constant, $\mathrm{\β}=\frac{1}{M}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}{(1-\mathrm{\α})}^{N+1-i},$ M is used for carrying out the level and smooth window length of envelope; e _Ave(n-1) be the level and smooth envelope value of the previous voice sampling point of current speech sampling point;

Step 3 at current sampling point with count forward in the scope of c (n) of N voice sampling points, is chosen the c (n) of maximum, i.e. x _Max(n)=max[c (n-N) ..., c (n)];

Step 4 according to the maximal value of choosing, calculates the envelope value of current speech sampling point:

e _max(n)＝αx _max(n)+(1-α)e _max(n-1)

Step 5 at the current speech sampling point with count forward in the scope of M voice sampling points, is averaged after the envelope value addition to all voice sampling points, calculates the level and smooth envelope value of current speech sampling point:

$e_{\mathrm{ave}}\left(n\right)=\frac{1}{M}\underset{i=0}{\overset{M}{\mathrm{\Σ}}}{e}_{\max }(n-i).$

5, digital auto gain control method according to claim 1 is characterized in that described type and envelope information according to the present frame voice signal, and the step of adjusting the gain of present frame voice sampling point is specially:

If present frame is a noise frame, the gain of present frame is made as 1, promptly for all n, g (n)=1 does not amplify each voice sampling point of present frame yet and not to dwindle;

If present frame is a speech frame, the yield value of each voice sampling point of present frame adjusted to seamlessly transit yield value g (n),

$g\left(n\right)={\left(\frac{\mathrm{threshold}}{e_{\mathrm{ave}}\left(n\right)}\right)}^{\mathrm{\θ}}$

N is the sequence number of voice sampling point, θ represent one with the relevant constant of adjusting of effect, 0＜θ≤1, threshold represents predefined threshold value, e _Ave(n) be at the current speech sampling point and count forward in the scope of M voice sampling points, the mean value after the envelope value addition of all voice sampling points.

6, digital auto gain control method according to claim 1 is characterized in that the voice signal and the envelope information of described former frame according to present frame and present frame, and the step of adjusting the gain of present frame voice sampling point is specially:

If the former frame of present frame and present frame is noise frame, the gain of present frame is made as 1, present frame is not amplified yet do not dwindle;

If former frame is a noise frame, present frame is a speech frame, this moment, the yield value with first voice sampling point of present frame was taken as 1, and the yield value of last voice sampling point is for seamlessly transitting yield value, the yield value of all the other each voice sampling points by 1 linear increment to seamlessly transitting yield value;

If former frame is a speech frame, present frame is a noise frame, be taken as the yield value of first voice sampling point of present frame and seamlessly transit yield value this moment, and the yield value of last voice sampling point is 1, and the yield value of all the other each voice sampling points is by seamlessly transitting yield value linear decrease to 1;

If former frame and present frame are speech frame, the yield value of each voice sampling point of present frame all is taken as and seamlessly transits yield value.

7, digital auto gain control method according to claim 2 is characterized in that smoothing processing is specially in the described frame: the gain of each voice sampling point of present frame is adjusted into level and smooth yield value g in the frame _Ave(n):

$g_{\mathrm{ave}}\left(n\right)=\frac{1}{M}\underset{i=0}{\overset{M}{\mathrm{\Σ}}}g(n-i)$

N is the sequence number of voice sampling point.

8, digital auto gain control method according to claim 7 is characterized in that, if level and smooth yield value is greater than predetermined threshold in the frame of current sampling point, then level and smooth yield value is made as predetermined threshold in the frame.

9, digital auto gain control method according to claim 2 is characterized in that also comprising:

The input voice range value of N voice sampling point multiply by the interior level and smooth yield value of frame and obtains exporting voice before the step e, current speech sampling point.

10, a kind of automatic gain control equipment comprises:

Voice signal is cut apart module, is used for voice signal is divided into some frames by fixing duration;

It is characterized in that also comprising:

The movable voice detection module is cut apart each frame after the module segmentation with described voice signal, divides into two types of speech frame or noise frame;

Interframe smoothing processing module is obtained the yield value of each voice sampling point of present frame, and in obtaining the process of yield value, according to the type of present frame, the interframe smoothing processing is carried out in the gain of the voice sampling point of present frame;

Described interframe smoothing processing module comprises:

The envelope detected module is carried out envelope detected to present frame, obtains the envelope information of present frame;

Gain regulation module according to the type and the envelope information of present frame voice signal, is adjusted the gain of present frame voice sampling point; Perhaps, adjust the gain of present frame voice sampling point according to the voice signal and the envelope information of the former frame of present frame and present frame.

11, automatic gain control equipment according to claim 10 is characterized in that also comprising: smoothing processing module in the frame, according to the type of present frame, smoothing processing in the frame is carried out in the gain when the voice sampling point of pre-treatment.

12, automatic gain control equipment according to claim 10 is characterized in that described envelope detected module comprises:

Range value is asked form unit, is used for the range value x (n) of present frame current speech sampling point is asked absolute value, obtains | x (n) |;

Envelope smoothing processing unit, according to | x (n) |, utilize formula $c\left(n\right)=\max \left[\right|x\left(n\right)|,\frac{\left|x\left(n\right)\right|-{\mathrm{\βe}}_{\mathrm{ave}}(n-1)}{1-\mathrm{\β}}]$ Make the variation of envelope-tracking input speech signal, n is the sequence number of voice sampling point, and α is a predetermined constant, $\mathrm{\β}=\frac{1}{M}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}{(1-\mathrm{\α})}^{N+1-i},$ M is used for carrying out the level and smooth window length of envelope; e _Ave(n-1) be the level and smooth envelope value of the previous voice sampling point of current speech sampling point;

Maximal value is chosen the unit, at the current speech sampling point with count forward in the scope of c (n) of N voice sampling points, chooses the range value c (n) of maximum, i.e. x _Max(n)=max[c (n-N) ..., c (n)];

The envelope value computing unit, the amplitude peak value c (n) according to choosing utilizes formula e _Max(n)=α x _Max(n)+(1-α) e _Max(n-1) envelope value of calculating current speech sampling point;

Envelope amplitude smoothing processing unit is at the current speech sampling point with count forward in the scope of M voice sampling points, to all voice

Average after the envelope value addition of sampling point, calculate the level and smooth envelope value of current speech sampling point:

$e_{\mathrm{ave}}\left(n\right)=\frac{1}{M}\underset{i=0}{\overset{M}{\mathrm{\Σ}}}{e}_{\max }(n-i).$

13, automatic gain control equipment according to claim 10 is characterized in that described gain regulation module comprises:

Seamlessly transit the yield value computing unit, according to formula $g\left(n\right)={\left(\frac{\mathrm{threshold}}{e_{\mathrm{ave}}\left(n\right)}\right)}^{\mathrm{\θ}}$ Calculating seamlessly transits yield value, and n is the sequence number of voice sampling point, θ represent one with the relevant constant of adjusting of effect, 0＜θ≤1, threshold represents predefined threshold value, e _Ave(n) be at the current speech sampling point and count forward in the scope of M voice sampling points, the mean value after the envelope value addition of all voice sampling points.

Control module is adjusted in gain, adjusts the gain of present frame voice sampling point.

14, automatic gain control equipment according to claim 13 is characterized in that described gain adjusts control module to the following adjustment of present frame voice sampling point gain carrying out:

If present frame is a noise frame, the gain of present frame is made as 1, promptly for all n, g (n)=1 does not amplify each voice sampling point of present frame yet and not to dwindle, and n is the sequence number of voice sampling point;

If present frame is a speech frame, the yield value of each voice sampling point of present frame is adjusted to seamlessly transit yield value.

15, automatic gain control equipment according to claim 13 is characterized in that described gain adjusts control module to the following adjustment of present frame voice sampling point gain carrying out:

If the former frame of present frame and present frame is noise frame, the gain of present frame is made as 1, present frame is not amplified yet do not dwindle;

If former frame is a noise frame, present frame is a speech frame, noise frame is to the speech frame transition, this moment, the yield value with first voice sampling point of present frame was taken as 1, the yield value of last voice sampling point is for seamlessly transitting yield value, the yield value of all the other each voice sampling points by 1 linear increment to seamlessly transitting yield value;

If former frame is a speech frame, present frame is a noise frame, speech frame is to the transition of noise frame, be taken as the yield value of first voice sampling point of present frame and seamlessly transit yield value this moment, the yield value of last voice sampling point is 1, and the yield value of all the other each voice sampling points is by seamlessly transitting yield value linear decrease to 1;

If former frame and present frame are speech frame, the yield value of each voice sampling point of present frame all is taken as and seamlessly transits yield value.

16, automatic gain control equipment according to claim 11 is characterized in that the smoothing processing module comprises in the described frame:

Level and smooth yield value computing unit in the frame is according to formula $g_{\mathrm{ave}}\left(n\right)=\frac{1}{M}\underset{i=0}{\overset{M}{\mathrm{\Σ}}}g(n-i)$ Calculate level and smooth yield value in the frame, n

Sequence number for the voice sampling point;

Gain adjusting unit in the frame is adjusted into level and smooth yield value g in the frame with the gain of each voice sampling point of present frame _Ave(n).

17, automatic gain control equipment according to claim 10, it is characterized in that also comprising: the voice output module, according to the input voice range value of N voice sampling point before the current speech sampling point, multiply by the interior level and smooth yield value of frame and obtain exporting voice, and export this voice.

Images