CN112116914B - Sound processing method and system based on variable step length LMS algorithm - Google Patents
Sound processing method and system based on variable step length LMS algorithm Download PDFInfo
- Publication number
- CN112116914B CN112116914B CN202010766086.4A CN202010766086A CN112116914B CN 112116914 B CN112116914 B CN 112116914B CN 202010766086 A CN202010766086 A CN 202010766086A CN 112116914 B CN112116914 B CN 112116914B
- Authority
- CN
- China
- Prior art keywords
- value
- gain
- input signal
- sound
- moment
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 43
- 238000003672 processing method Methods 0.000 title claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 52
- 238000004364 calculation method Methods 0.000 claims abstract description 49
- 238000000034 method Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 7
- 238000009499 grossing Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 5
- 150000001875 compounds Chemical class 0.000 claims description 4
- 230000008859 change Effects 0.000 description 10
- 230000005236 sound signal Effects 0.000 description 6
- 230000003044 adaptive effect Effects 0.000 description 4
- 230000007704 transition Effects 0.000 description 4
- 230000009467 reduction Effects 0.000 description 3
- 108010076504 Protein Sorting Signals Proteins 0.000 description 2
- 230000003321 amplification Effects 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000009795 derivation Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
- H03H21/00—Adaptive networks
Landscapes
- Control Of Amplification And Gain Control (AREA)
Abstract
The invention discloses a sound processing method and a sound processing system based on a variable step length LMS algorithm, wherein the sound processing method comprises the following steps: s1, calculating a gain value g (n) of the current moment, wherein the calculation formula of the gain value g (n) is as follows: g (n) = g (n-1) + u (n-1) × e (n-1) × P x (n-1); wherein, P x (n-1) is the power value of the sound input signal at the previous moment, u (n-1) is the step factor of the previous moment, e (n-1) is the error value of the previous moment, g (n-1) is the gain value of the previous moment, and n is the serial number of the sound input signal at the current moment; s2, performing gain processing on the sound input signal x (n) at the current moment according to the gain value obtained by calculation to obtain an output signal y (n), wherein the gain value is calculated as y (n) = g (n) × x (n).
Description
Technical Field
The invention relates to a sound signal processing technology, in particular to a voice signal processing method and system based on an adaptive algorithm.
Background
In the process of acquiring and outputting the sound signal, signal processing is generally required. One of the most common processing means is to perform gain control processing on the audio signal to achieve balanced output of the audio signal. At present, most AGC algorithms are used and are based on one algorithm of energy comparison, the maximum amplitude of a signal in a nearest time slice is searched, and if the searched current frame power value is larger than the existing maximum power value, the maximum power value is updated; according to the searched maximum power value, comparing with a preset expected value to obtain an expected gain; calculating the actual gain of the current frame by using a first-order recursive smoothing filter according to the expected gain of the current frame and the actual gain of the previous frame; and finally, multiplying all the sampling points of the current frame by the gain value to obtain the output signal of the current frame. Although the AGC algorithm used at present can prevent the overload phenomenon of the speech signal, linear distortion is generated when the speech is suddenly changed, because the algorithm does not consider the situation that the power variation range of the speech signal in each frame may be wide, and the same processing mode for the upper and lower flow signals and the linear area signal increases the processing error.
In addition, adaptive LMS, NMLS, RLS and other algorithms are also gradually applied in the sound signal processing, however, there are also problems that the error generated by linear distortion is large when the voice suddenly changes, the voice signal adjusting speed is slow, even the delay is delayed, so that the hearing comfort of the listener is not ideal.
Disclosure of Invention
The present invention aims to solve the above problems, and provide an improved sound processing method and system based on a variable step LMS algorithm, which can effectively avoid a large error caused by linear distortion when a voice suddenly changes, so that the voice signal adjustment speed is fast, the voice can be kept in a certain amplitude range, and the hearing comfort of a listener is improved.
One of the purposes of the present invention is to provide a sound processing method based on a variable step length LMS algorithm, which comprises the following steps:
s1, calculating a gain value g (n) at the current moment, wherein the calculation formula of the gain value g (n) is as follows:
g(n)＝g(n-1)+u(n-1)*e(n-1)*P x (n-1)；
wherein ,Px (n-1) is the power value of the sound input signal at the previous moment, u (n-1) is the step factor of the previous moment, e (n-1) is the error value of the previous moment, g (n-1) is the gain value of the previous moment, n is the serial number of the sound input signal at the current moment, n is a positive integer, and n is more than or equal to 2;
s2, performing gain processing on the sound input signal x (n) at the current moment according to the gain value obtained by calculation to obtain an output signal y (n),
y(n)＝g(n)*x(n)。
preferably, the method further comprises the following steps:
s3, calculating the power value P of the sound input signal at the current moment x (n), a step size factor u (n) at the current time, and an error value e (n) at the current time;
and S4, iteratively calculating a gain value g (n + 1) at the next moment, wherein the calculation formula is g (n + 1) = g (n) + u (n) × e (n) × P x (n)；
And S5, performing gain processing on the voice input signal x (n + 1) at the next time according to the gain value g (n + 1) to obtain a voice output signal y (n + 1) at the next time, wherein the calculation formula is y (n + 1) = g (n + 1) × x (n + 1).
Preferably, the power value P of the voice input signal at the current moment x The calculation formula of (n) is:
P x (n)＝λP x (n-1)+(1-λ)x 2 (n)，
wherein λ is a smoothing factor, x 2 And (n) is the square of the sound input signal x (n) at the current moment.
Preferably, the step factor u (n) at the current time is calculated by:
wherein ,
or ,
in the above-mentioned formula, the compound has the following formula,
a (n-1) is the square value of the gradient at the previous moment, the value range of beta is more than 0 and less than 1, the value range of theta is more than 0 and less than 1 y (n-1) is the power value of the sound output signal at the previous moment, P 2 x (n) is the power value P of the voice input signal at the current moment x The square of (n).
Has good effectOptionally, the sound output signal power P y The calculation formula of (n) is:
P y (n)＝g(n)*g(n)*P x (n)。
preferably, the error value e (n) = P at the current time exp -P y (n)，P exp To a desired value, P exp The value range is more than 0 and less than P exp ＜1。
The invention also discloses a sound processing system based on the variable step LMS algorithm, which comprises:
a signal input unit for acquiring a sound input signal;
a gain control processing unit for calculating the gain value of the voice input signal at the current time and calculating the output signal according to the calculated gain value,
a signal output unit which performs gain processing on the voice input signal according to the calculated output signal and outputs the voice signal after the gain processing;
the signal parameter calculation unit comprises a power calculation module for calculating the power value of the sound input signal at the current moment, an error calculation module for calculating the error value between the expected power value of the sound output signal and the power value of the sound output signal after gain processing, and a step length calculation module for calculating the step length factor at the current moment;
the storage unit is used for storing the calculation process data and providing the gain control processing unit for calling;
the formula of the gain control processing unit for calculating the gain value is as follows:
g(n)＝g(n-1)+u(n-1)*e(n-1)*P x (n-1)
wherein ,Px (n-1) is the power value of the sound input signal at the previous moment, u (n-1) is the step factor at the previous moment, e (n-1) is the error value at the previous moment, g (n-1) is the gain value at the previous moment, n is the serial number of the sound input signal at the current moment, n is a positive integer, and n is more than or equal to 2;
the output signal is calculated as:
y (n) = g (n) × x (n), where x (n) is the sound input signal at the current time.
Compared with the prior art, the remarkable progress of the invention is at least reflected in that: the sound processing based on the improved LMS algorithm provided by the invention can reduce the amplitude of a large signal and amplify a small signal, so that the amplitude of the whole speech is kept relatively consistent, the transition zone is small, and the sound gain speed is high. The gain value can quickly respond according to the change amplitude of the sound signal, and when the sound amplitude changes suddenly, the gain value can also be changed quickly, so that the effect of slow amplification or amplitude reduction is effectively avoided, and the sense of discomfort of hearing is reduced.
Drawings
FIG. 1 is a schematic diagram illustrating an algorithm derivation process in a sound processing method according to an embodiment of the present invention;
FIG. 2 is a time domain comparison of an input signal and an output signal according to an embodiment of the present invention;
FIG. 3 is a graph reflecting input speech and gain curves according to an embodiment of the present invention;
fig. 4 is a flowchart illustrating a sound processing method according to an embodiment of the invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings and specific examples. It should be noted that the embodiments of the present invention are not limited to the examples provided.
The embodiment of the invention provides a sound processing method based on a variable step length LMS algorithm, which comprises the following steps:
s1, calculating a gain value g (n) at the current moment, wherein the calculation formula of the gain value g (n) is as follows:
g(n)＝g(n-1)+u(n-1)*e(n-1)*P x (n-1)；
wherein ,Px (n-1) is the power value of the sound input signal at the previous moment, u (n-1) is the step factor of the previous moment, e (n-1) is the error value of the previous moment, g (n-1) is the gain value of the previous moment, n is the serial number of the sound input signal at the current moment, n is a positive integer, and n is more than or equal to 2;
in this embodiment, the gain value g (n) is used as the iteration weight of the variable step length LMS algorithm, and the magnitude of g (n) is the value P of the power of the voice input signal passing through the previous moment x (n-1), step size factoru (n-1), the error value e (n-1), and the gain value g (n-1).
S2, calculating output signal y (n)
According to the gain value obtained by calculation, the sound input signal x (n) at the current moment is subjected to gain processing to obtain an output signal y (n), namely
y(n)＝g(n)*x(n)。
It is understood that the digital audio output signal sequence y (n) processed by the automatic gain control is generated as a product of the input signal sequence x (n) and the gain parameter sequence g (n).
It will be appreciated that the iterative calculation of gain values from the second acoustic input signal is performed with the first acoustic input signal as an initial value and the gain processing performed on the corresponding acoustic input signal. The initial value setting includes: initial gain value g (1) =1, step factor u (1) =0, error value e (1) = P exp -P x (1) Value of power P x (1)＝x 2 (1) And calculating a gain value of the second sound input signal according to the initial value.
Preferably, after the signal output is completed once, the processing of the input signal at the next time is performed, and the method specifically includes the steps of:
iteratively calculating the gain value g (n + 1) at the next time, wherein the calculation formula is g (n + 1) = g (n) + u (n) × e (n) × P x (n)；
The audio input signal x (n + 1) at the next time is subjected to gain processing based on the gain value g (n + 1) to obtain the audio output signal y (n + 1) at the next time, and the calculation formula is y (n + 1) = g (n + 1) × x (n + 1).
In some embodiments, the power value P of the sound input signal at the current moment x The calculation formula of (n) is:
P x (n)＝λP x (n-1)+(1-λ)x 2 (n)，
the lambda is a smoothing factor, the influence of the previous moment on the current moment is determined by the lambda value, the larger the lambda value is, the larger the influence of the previous moment is, the more gradual the power change is, and otherwise, the larger the power change is. Preferably, λ is recommended to be 0.9.x is the number of 2 (n) is the current timeThe square of the sound input signal x (n) is plotted.
It should be noted that the value of the error value e (n) at the current time is the desired power value P exp With the power value P of the output signal y (n) calculated after each iteration y The difference between (n), i.e. e (n) = P exp -P y (n)，P exp The value range is more than 0 and less than P exp < 1, a value of 0.2 is suggested. In some embodiments, the sound output signal power calculated after each iteration is P y (n) the calculation formula is:
P y (n)＝g(n)*g(n)*P x (n)。
it is understood that, taking the processing of the speech signal as an example, in the processing of the speech signal, in order to keep the amplitude of the speech signal relatively consistent, the adjustment change of the speech signal needs a gain value for adjustment. For example, when the voice signal is small, the gain value is large, and when the voice signal is large, the gain value is small. In the embodiment of the invention, the adjustment of the gain value is iteratively adjusted through a variable-step LMS algorithm (least mean square), the input and output energy value of the signal is used as the input and output signals of the variable-step LMS algorithm, and the mechanism of the variable-step LMS algorithm is to perform adaptive iteration on the output signal which is continuously close to the expected value.
It can be further understood that the core idea of the LMS algorithm is to use the square error instead of the mean square error to calculate the gradient value of the current time point, the error signal is equivalent to a paraboloid in the two-dimensional graph, the lowest point is the minimum error point, the output signal is the expected signal at this time, the gradient value of each point on the paraboloid represents the magnitude of the error signal, the gradient amplitude is large in the initial stage, and is small and tends to 0 in the steady state. If the step value of the LMS algorithm with fixed step length is set to be too small, the signal convergence speed is too low, iteration is needed for many times to reach the minimum value point, if the step length is too large, the signal convergence speed is very high, but when the step length is close to an expected signal, an error signal always swings left and right at the bottom of a bowl on the performance surface due to the fact that the step length is too large, and a steady-state error is large. The fixed step LMS algorithm cannot equalize between the steady state error and the convergence speed. The variable-step LMS algorithm adopted in the embodiment of the invention can carry out iterative computation according to the current gradient value, and when the current iteration gradients are larger, the dithering is caused due to the overlarge step length, and the step length is reduced to reduce the dithering; when the gradient of the current rounds of iteration is small, the step length is increased to accelerate the iteration. Therefore, balance between the steady-state error and the convergence rate is achieved, and the calculation efficiency is improved.
It should be noted that, in a sudden change speech signal with a large duration, it is critical to quickly adjust the gain value of the signal to avoid a large transition band of speech output. To achieve fast adjustment of the gain value, in some embodiments, the step size factor u (n) at the current time is changed in real time according to the sampling point, and the calculation formula is:
wherein ,μ0 The suggested value is 0.1 for the initial step length; the setting of e is to prevent the case that the denominator is 0, e can be set to be a very small positive number, and as an embodiment, the value of e is 10 -6 . The maximum value of the step length is mu according to the calculation formula of the step length factor 0 I e, the error goes to 0 as the desired value is approached. Correspondingly, the gain value calculation formula shows that the updated value of the gain also tends to 0, the gain is accelerated to be infinitely small, the iteration change is gradually reduced, and finally tends to 0, so that the optimal weight value is reached.
It can be understood that, in the scheme of this embodiment, in the current speech scene, when the gain change is aggressive and the gradient value is large, the step size is controlled to be relatively reduced to increase the stability, so as to avoid the occurrence of jitter due to an excessively large step size. And when the output signal is gradually close to the expected signal, the step length is properly increased, so that the convergence speed is increased. The problem that in a traditional mode of fixing the step length, when an error performance surface tends to be in a stable state, the step length is updated slowly along with the reduction of error change, and more iteration times are needed before the error performance surface approaches to a stable state but an optimal point is not reached is solved.
To improve gain stability, maintain volume consistency after speech gain, and algorithm on processed soundWhen the quantity is suddenly changed, the tracking performance can be faster. In the embodiment, the step length is adjusted by using historical gradient information, the aggressiveness of gain change is measured by introducing a historical gradient square value, and the aggressiveness is increased when the value is larger. In some embodiments of the present invention, the,
wherein ,
is composed of
The square of the square,
for error performance surface gradient values, it can be expressed as:
in order to record gradient information for a longer period of time, the value of beta is suggested to be 0.9. The LMS algorithm in this embodiment uses the square e of the instantaneous error 2 (n) instead of the mean square error, the gradient of the instantaneous error signal squared is used as the estimate of the mean square error function gradient. The adaptation step size will get smaller and smaller over time if only the square root of the square of all historical gradients is used as the denominator. Therefore, in this embodiment, the average value of the past short time gradient is estimated by using the moving average, the gradient value can be smoothed, if a local discontinuity point occurs, the value of the square value a (n) of the gradient at the current time depends on the gradient value a (n-1) at the previous time with a larger weight, the step cannot be suddenly changed, and the problem that the speed of the self-adaptive step decaying along with the time is too aggressive can be avoided. In this embodiment, since a (n) measures the average of the gradient of the previous period including the current time, the step size of the current point can be adjusted and controlled by using a (n) as the empirical information.
Considering that the historical gradient mean value inaccurate phenomenon of estimation may be generated in the early stage of iteration, the problem is solved by introducing bias correction as a preferred embodiment. The formula after adding the deviation correction is as follows:
where 0 < θ < 1, θ is suggested to be 0.1, n is the serial number of the sound input signal at the current time, and is a positive integer, for example, if the sound input signal at the current time is the 3 rd sound input signal, n =3.
Referring to fig. 1, which is a flowchart illustrating algorithm derivation in the sound processing method according to the embodiment of the present invention, as shown in the figure, the gain value g (n) is obtained by calculating the gain adaptive adjustment algorithm (i.e. the calculation formula of g (n) in step S1), which is obtained by calculating four paths of input data, wherein z-1 represents delaying the input signal, i.e. taking the signal parameter of the previous time point, i.e. taking the power value P of the sound input signal of the previous time point x And (n-1), the step size factor u (n-1) of the previous moment, the error value e (n-1) of the previous moment and the gain value g (n-1) of the previous moment are used as calculation parameters to calculate the gain value g (n) of the current moment. Step-size dynamic adjustment algorithm (i.e. formula)
) Then the data is passed through two paths (i.e., e (n) and P) x (n)) are calculated. And finally, multiplying the gain value g (n) obtained by calculation at the current moment by the sound input signal x (n) at the current moment to obtain an output signal y (n).
In order to verify the remarkable progress of the invention, a section of voice signals with different amplitudes are input, and the voice gain processing is carried out on the voice signals by the voice processing method based on the variable step LMS algorithm provided by the invention, and referring to fig. 2 showing a comparison graph of the input signals and the output signals, it can be seen that after the adjustment processing is carried out by the method of the invention, the voice signals with larger amplitude can be reduced, and the voice signals with smaller amplitude can be amplified, so that the amplitudes of the whole section of voice keep relatively consistent, the transition band is small, and the voice gain speed is higher.
Reference is made to FIG. 3, which shows a reflection diagram of the input speech and gain curves, with a sampling frequency of 16kHz, selectedNo. 2.3X 10 5 3.3X 10 th 5 And (3) a small-amplitude voice segment range of the sampling points at each position, wherein the upper graph is the original voice of the segment, and the lower graph is the gain value processed by the gain scheme. The overall amplitude of the original speech signal is small, but the relative amplitude variation is large. After the sound processing method based on the variable step length LMS algorithm is carried out, the gain value can quickly react according to the change amplitude of the voice, and the gain value can also be quickly changed when the voice amplitude changes suddenly. Therefore, the voice gain range of the voice processing method based on the variable step LMS algorithm is large, and the convergence rate of the new variable step LMS causes that the gain speed is changed quickly and the transition section from small voice to large voice is narrower when a section of jump voice is processed. Avoid the slow amplification or amplitude reduction effect and reduce the uncomfortable feeling of the auditory sense.
The embodiment of the invention also provides a sound processing system based on the variable step length LMS algorithm, which comprises:
the signal input unit is used for acquiring a sound input signal at the current moment;
a gain control processing unit for calculating the gain value of the voice input signal at the current time and calculating the output signal according to the calculated gain value,
a signal output unit which performs gain processing on the voice input signal according to the calculated output signal and outputs a voice signal after the gain processing;
the signal parameter calculation unit comprises a power calculation module for calculating the power value of the sound input signal at the current moment, an error calculation module for calculating the error value between the expected power value of the sound output signal and the power value of the sound output signal after gain processing, and a step length calculation module for calculating the step length factor at the current moment;
the storage unit is used for storing the calculation process data and providing the gain control processing unit for calling;
the formula of the gain control processing unit for calculating the gain value is as follows:
g(n)＝g(n-1)+u(n-1)*e(n-1)*P x (n-1)
wherein ,Px (n-1) is the sound of the previous momentThe method comprises the steps that the power value of an input signal, u (n-1) is a step size factor of a previous moment, e (n-1) is an error value of the previous moment, g (n-1) is a gain value of the previous moment, n is a serial number of a sound input signal of the current moment, the value is a positive integer, and n is larger than or equal to 2;
the output signal is calculated as:
y (n) = g (n) × x (n), where x (n) is the sound input signal at the current time.
It will be appreciated that the gain values are calculated iteratively starting from the second acoustic input signal, with the first acoustic input signal as an initial value. The initial value setting includes: initial gain value g (1) =1, step factor u (1) =0, error value e (1) = P exp -P x (1) Value of power P x (1)＝x 2 (1) And calculating a gain value of the second sound input signal according to the initial value.
Preferably, the power value P of the sound input signal at the current moment x The calculation formula of (n) is:
P x (n)＝λP x (n-1)+(1-λ)x 2 (n)，
wherein, lambda is a smoothing factor, the suggested value is 0.9 2 And (n) is the square of the sound input signal x (n) at the current moment.
Preferably, the step factor u (n) at the current time is calculated by:
wherein ,
or ,
in the above-mentioned formula, the compound of formula,
a (n-1) is the gradient value of the previous moment, the value range of beta is more than 0 and less than 1, and the suggested value is0.9, theta is greater than 0 and less than 1, the suggested value is 0.1 y (n-1) is the power value of the sound output signal at the previous moment, P 2 x (n) is the power value P of the voice input signal at the current moment x The square of (n).
The value of the error value e (n) of the sampling point at the current moment is the expected power value P exp With the power value P of the output signal y (n) calculated after the iterative gain y The difference between (n), i.e. e (n) = P exp -P y (n)，P exp The value range is more than 0 and less than P exp If < 1, the value is suggested to be 0.2. In some embodiments, the sound output signal power calculated after each iteration of the gain is P y (n) the calculation formula is:
P y (n)＝g(n)*g(n)*P x (n)。
referring to fig. 4, in the processing flow of the initial sampling point of the variable step LMS algorithm-based sound processing system according to the embodiment of the present invention, after parameter initialization is completed, VAD detection (mute detection) is performed to determine whether an input is a mute section, if the input is a mute section, a gain value is set to 1, otherwise, gain calculation processing is performed, after the gain value is determined, an output signal y (n) is calculated, and amplitude limiting processing is performed on the input signal according to the calculated y (n) to output a signal. It can be understood that when the signal processing needs to be continued, the power value P of the sound input signal at the current moment needs to be calculated x (n), a step factor u (n), and an error value e (n) are iteratively calculated to obtain a gain value g (n + 1) at the next time, which is calculated by the formula g (n + 1) = g (n) + u (n) × e (n) × P x (n); further, the audio input signal x (n + 1) at the next time is subjected to gain processing based on the gain value g (n + 1) to obtain the audio output signal y (n + 1) at the next time, and the calculation formula is y (n + 1) = g (n + 1) × x (n + 1).
The embodiments of the present invention have been described in detail, but the embodiments are only examples, and the present invention is not limited to the embodiments described above. Any equivalent modifications and substitutions to those skilled in the art are also within the scope of the present invention. Accordingly, equivalent changes and modifications made without departing from the spirit and scope of the present invention should be covered by the present invention.
Claims (10)
1. A sound processing method based on a variable step LMS algorithm is characterized by comprising the following steps:
s1, iteratively calculating a gain value g (n) at the current moment, wherein the calculation formula of the gain value g (n) is as follows:
g(n)＝g(n-1)+u(n-1)*e(n-1)*P x (n-1)；
wherein ,Px (n-1) is the power value of the sound input signal at the previous moment, u (n-1) is the step factor at the previous moment, e (n-1) is the error value at the previous moment, g (n-1) is the gain value at the previous moment, n is the serial number of the sound input signal at the current moment, n is a positive integer, and n is more than or equal to 2;
s2, performing gain processing on the sound input signal x (n) at the current moment according to the gain value obtained by calculation to obtain an output signal y (n):
y(n)＝g(n)*x(n)。
2. the sound processing method based on the variable step LMS algorithm according to claim 1, further comprising the steps of:
s3, calculating the power value P of the sound input signal at the current moment x (n), a step size factor u (n) at the current time, and an error value e (n) at the current time;
and S4, iteratively calculating a gain value g (n + 1) at the next moment, wherein the calculation formula is g (n + 1) = g (n) + u (n) × e (n) × P x (n)；
S5, gain processing is performed on the audio input signal x (n + 1) at the next time based on the gain value g (n + 1) to obtain the audio output signal y (n + 1) at the next time, and the calculation formula is y (n + 1) = g (n + 1) × x (n + 1).
3. Sound processing method based on the step-size-variable LMS algorithm as claimed in claim 2, characterized in that the power value P of the sound input signal at the current time is x The formula for (n) is calculated as:
P x (n)＝λP x (n-1)+(1-λ)x 2 (n)，
wherein, lambda is a smoothing factor, the value range of lambda is more than 0 and less than 1 2 And (n) is the square of the sound input signal x (n) at the current moment.
4. A sound processing method based on a variable step LMS algorithm according to claim 2, characterized in that: the step size factor u (n) at the current time is calculated as:
wherein ,u0 For the initial step size, e can be set to a small positive number,
or ,
in the above-mentioned formula, the compound has the following formula,
a (n-1) is the square value of the gradient at the previous moment, the value range of beta is more than 0 and less than 1, and the value range of theta is more than 0 and less than 1.
5. Sound processing method based on the variable-step LMS algorithm according to claim 4, characterized in that the sound output signal power calculated after the iterative gain is P y (n)，P y The calculation formula of (n) is:
P y (n)＝g(n)*g(n)*P x (n)。
6. sound processing method based on a step-size-variable LMS algorithm according to claim 4 characterized in that P exp The error value of the sampling point at the current time is e (n) which is the value of the expected valueIs a desired value of power P exp With the calculated power value P of the output signal y (n) y The difference between (n), e (n) is calculated as e (n) = P exp -P y (n)，P exp The value range is more than 0 and less than P exp ＜1。
7. A sound processing system based on a variable step LMS algorithm, comprising:
a signal input unit for acquiring a sound input signal;
the gain control processing unit is used for calculating the gain value of the sound input signal at the current moment;
the signal output unit is used for carrying out gain processing on the sound input signal at the current moment according to the gain value obtained by calculation to obtain an output signal;
the signal parameter calculation unit comprises a power calculation module for calculating the power value of the sound input signal at the current moment, an error calculation module for calculating an error value between the expected power value of the sound output signal and the power value of the sound output signal after gain processing, and a step length calculation module for calculating a step length factor at the current moment;
iteratively calculating the gain value of the next moment;
performing gain processing on the voice input signal at the next moment according to the gain value at the next moment to obtain a voice output signal at the next moment;
the storage unit is used for storing the calculation process data and providing the gain control processing unit for calling;
the formula of the gain control processing unit for calculating the gain value is as follows:
g(n)＝g(n-1)+u(n-1)*e(n-1)*P x (n-1)
wherein ,Px (n-1) is the power value of the sound input signal at the previous moment, u (n-1) is the step factor of the previous moment, e (n-1) is the error value of the previous moment, g (n-1) is the gain value of the previous moment, n is the serial number of the sound input signal at the current moment, n is a positive integer, and n is more than or equal to 2;
the output signal is calculated as:
y (n) = g (n) × x (n), where x (n) is the sound input signal at the current time.
8. Sound processing system based on the step-size-variable LMS algorithm of claim 7, characterized in that the power P of the sound input signal at the current moment x The calculation formula of (n) is: p x (n)＝λP x (n-1)+(1-λ)x 2 (n), wherein lambda is a smoothing factor, lambda ranges from 0 to 1 2 And (n) is the square of the sound input signal x (n) at the current moment.
9. A sound processing system based on a variable step LMS algorithm according to claim 8, wherein the step size factor u (n) at the current time is calculated as:
wherein ,
or ,
in the above-mentioned formula, the compound of formula,
a (n-1) is the gradient value of the previous moment, the value range of beta is more than 0 and less than 1, and the value range of theta is more than 0 and less than 1.
10. The variable step LMS algorithm based sound processing system of claim 9, wherein the value of the error e (n) of the sample point at the current time is the desired power value P exp With the calculated power value P of the output signal y (n) y (n) ofThe difference therebetween, i.e. e (n) = P exp -P y (n)，P exp The value range is more than 0 and less than P exp < 1, the power of the acoustic output signal calculated after the iterative gain is P y (n) the calculation formula is: p y (n)＝g(n)*g(n)*P x (n)。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010766086.4A CN112116914B (en) | 2020-08-03 | 2020-08-03 | Sound processing method and system based on variable step length LMS algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010766086.4A CN112116914B (en) | 2020-08-03 | 2020-08-03 | Sound processing method and system based on variable step length LMS algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112116914A CN112116914A (en) | 2020-12-22 |
| CN112116914B true CN112116914B (en) | 2022-11-25 |
Family
ID=73799598
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010766086.4A Active CN112116914B (en) | 2020-08-03 | 2020-08-03 | Sound processing method and system based on variable step length LMS algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112116914B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002026788A (en) * | 2000-07-13 | 2002-01-25 | Matsushita Electric Ind Co Ltd | Receiver |
| CN106443597A (en) * | 2016-12-06 | 2017-02-22 | 上海机电工程研究所 | Self-adaption anti-deception system and self-adaption anti-deception method for resisting active decoy jamming |
| CN107393550A (en) * | 2017-07-14 | 2017-11-24 | 深圳永顺智信息科技有限公司 | Method of speech processing and device |
| CN109643554A (en) * | 2018-11-28 | 2019-04-16 | 深圳市汇顶科技股份有限公司 | Adaptive voice Enhancement Method and electronic equipment |
| CN110677796A (en) * | 2019-03-14 | 2020-01-10 | 深圳市攀高电子有限公司 | Audio signal processing method and hearing aid |
| CN111052767A (en) * | 2017-08-28 | 2020-04-21 | 索尼公司 | Audio processing device, audio processing method, and information processing device |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8615208B2 (en) * | 2010-11-02 | 2013-12-24 | Crestcom, Inc. | Transmitter linearized in response to signal magnitude derivative parameter and method therefor |
-
2020
- 2020-08-03 CN CN202010766086.4A patent/CN112116914B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002026788A (en) * | 2000-07-13 | 2002-01-25 | Matsushita Electric Ind Co Ltd | Receiver |
| CN106443597A (en) * | 2016-12-06 | 2017-02-22 | 上海机电工程研究所 | Self-adaption anti-deception system and self-adaption anti-deception method for resisting active decoy jamming |
| CN107393550A (en) * | 2017-07-14 | 2017-11-24 | 深圳永顺智信息科技有限公司 | Method of speech processing and device |
| CN111052767A (en) * | 2017-08-28 | 2020-04-21 | 索尼公司 | Audio processing device, audio processing method, and information processing device |
| CN109643554A (en) * | 2018-11-28 | 2019-04-16 | 深圳市汇顶科技股份有限公司 | Adaptive voice Enhancement Method and electronic equipment |
| CN110677796A (en) * | 2019-03-14 | 2020-01-10 | 深圳市攀高电子有限公司 | Audio signal processing method and hearing aid |
Non-Patent Citations (2)
| Title |
|---|
| Control of DVR using Variable Step-Size Griffith"s LMS with Optimized tuning of Proportional Integrator Gains;Talada Appala Naidu;《2019 International Conference on Power Electronics Applications and Technology in Present Energy Scenario (PETPES)》;20200220;全文 * |
| 音频信号AGC及响度控制算法的研究;吴振宙;《中国优秀硕士学位论文全文数据》;20160715(第7期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112116914A (en) | 2020-12-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6490641B2 (en) | Audio signal compensation based on loudness | |
| CN107431858B (en) | Loudspeaker protection | |
| JP4177882B2 (en) | Hearing aid with adaptive feedback suppression system | |
| US8068629B2 (en) | Hearing aid and method of utilizing gain limitation in a hearing aid | |
| US9917562B2 (en) | Systems and methods for adjusting automatic gain control | |
| EP2002688B1 (en) | Hearing aid and method of estimating dynamic gain limitation in a hearing aid | |
| JP5927558B2 (en) | Howling detection device, howling suppression device, and howling detection method | |
| GB2539725A (en) | Loudspeaker protection | |
| WO1999051059A1 (en) | Apparatus and methods for combining audio compression and feedback cancellation in a hearing aid | |
| JP4016206B2 (en) | Audio signal processing apparatus and audio signal processing method | |
| AU2007220498A1 (en) | A Hearing Aid And Method of Compensation For Direct Sound in Hearing Aids | |
| US8583717B2 (en) | Signal processing circuit | |
| GB2574679A (en) | Techniques for howling detection | |
| CN108630216B (en) | MPNLMS acoustic feedback suppression method based on double-microphone model | |
| CN112116914B (en) | Sound processing method and system based on variable step length LMS algorithm | |
| JP2022519945A (en) | Noise reduction signal saturation control | |
| GB2455823A (en) | Active noise cancellation filter cut-off frequency adjusted in accordance with magnitude of filter output signal | |
| JP3296330B2 (en) | Adaptive filter, control method of adaptive filter, and storage medium storing program | |
| Belloch et al. | Efficient target-response interpolation for a graphic equalizer | |
| JP2004015125A (en) | Automatic gain control apparatus | |
| KR20230057287A (en) | Method for generating active noise reduction filter, storage medium and earphone | |
| JP2574283B2 (en) | Howling prevention device | |
| CN117998255B (en) | Adaptive equalization method, equalizer and system with dynamic range control | |
| CN112039498B (en) | Self-adaptive signal processing method and medium based on polymorphic variable step-length least mean square | |
| EP3940954B1 (en) | Systems and methods for limiter functions |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |