JP4321049B2 - Automatic gain controller - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an automatic gain control device that controls a gain when an input signal is amplified according to the level of the input signal.
[0002]
[Prior art]
In the communication field such as telephone, it is known that transmission loss varies depending on communication forms (transmission forms) such as wired and wireless, or analog communication and digital communication. Therefore, when trying to perform stable communication regardless of the magnitude of transmission loss, for example, trying to make the volume of the sound heard by the listener almost constant regardless of the volume (volume) of the speaker's voice (speech) In this case, it is necessary to control the gain when amplifying the audio signal so that the volume is constant. As described above, the automatic gain control for the audio signal is performed not only in the communication field, but also in the case where the recognition rate fluctuates depending on the volume level of the input voice as in the voice recognition system, or the volume of the alarm sound or the like is set to the ambient noise level. It is used in various fields, such as when adjusting according to.
[0003]
[Non-Patent Document 1]
Materials for the Telecommunications Terminal Equipment Examination Association “Telecom Terminal Equipment Conformity Certification Criteria / Technical Conditions” Chapter 3 Explanation of Technical Standards (Terminal Equipment Regulations, etc.) 3.2 Analog Telephone Terminal Article 14 Transmission Power p. 120
[0004]
[Problems to be solved by the invention]
By the way, the conventional automatic gain control apparatus has only one gain function F indicating the gain relationship according to the level of the input signal as shown in FIG. 27. Therefore, when the level fluctuation of the input signal is large, the output signal It is difficult to perform gain control that keeps the level of the above constant (see Non-Patent Document 1).
[0005]
The present invention has been made in view of the above circumstances, and an object of the present invention is to provide an automatic gain control apparatus capable of gain control that keeps the output signal level constant even when the level fluctuation of the input signal is large. .
[0006]
[Means for Solving the Problems]
  In order to achieve the above object, an automatic gain control device for controlling a gain when amplifying the input signal according to the level of the input signal detects a noise component included in the input signal. A noise detection unit, a short-time average level detection unit for detecting a time average of input signals in a short time of about 16 milliseconds, and a plurality of gains prepared in advance according to the magnitude of noise components detected by the noise detection unit A gain function selection unit that selects one of the gain functions from the functions, a short time average value of the input signal detected by the short time average level detection unit, and an input signal from the gain function selected by the gain function selection unit A gain calculation unit for calculating a gain when amplifyingA gain fluctuation correction unit that detects a fluctuation amount of the gain calculated by the gain calculation unit and corrects the gain in a direction to suppress the fluctuation when the detected fluctuation amount exceeds a predetermined range;The noise detection unit includes a long-time average level detection unit that detects a time average in a time obtained by multiplying the short time of the input signal by 16 times, and the gain function selection unit is detected by the long-time average level detection unit. The long-time average level detection unit estimates a noise component included in the input signal by selecting any one of a plurality of gain functions prepared in advance according to the long-time average value of the input signal. A long-term average value is detected from the value.
[0009]
  Claim2The invention of claim1In the invention, the long-time average level detection unit detects a long-time average of the input signal by averaging a plurality of short-time average values obtained by the short-time average level detection unit.
[0010]
  Claim3The invention of claim1 or 2In the invention, an input signal discriminating unit for discriminating whether or not the input signal is an audio signal is provided, and the long-time average level detecting unit determines the length of the input signal only when the input signal discriminating unit discriminates that the input signal is not an audio signal. It is characterized by detecting a time average.
[0011]
  Claim4In the invention of claim 1, in the invention of claim 1, the noise detection unit includes a non-speech section detection unit that detects a non-speech section in which an input signal does not include a speech signal, a frequency conversion unit that obtains a frequency spectrum of the input signal, A non-speech spectrum estimation unit that estimates a frequency spectrum of a non-speech signal that is not a speech signal from a frequency spectrum obtained by a frequency conversion unit in a non-speech segment detected by a non-speech segment detection unit; One of the gain functions is selected from a plurality of gain functions prepared in advance according to the frequency spectrum of the non-speech signal estimated by the non-speech spectrum estimation unit.
[0012]
  Claim5The invention of claim 1 to claim 14In any of the inventions, the short-time peak detection unit that detects the peak value of the input signal within a unit time when the short-time average level detection unit detects the short-time average value, and the short-time peak detection unit detects And a gain correction unit that corrects the gain calculated by the gain calculation unit when the peak value to be exceeded exceeds a predetermined threshold value.
[0013]
  Claim6The invention of claim 1 to claim 15In any of the inventions, an input signal discriminating unit that discriminates whether or not the input signal is an audio signal, and the gain function selecting unit changes the gain function to be selected according to the discrimination result of the input signal discriminating unit. Features.
[0014]
  Claim7The invention of claim 1 to claim 16In any of the inventions, an input signal discriminating unit for discriminating whether or not the input signal is an audio signal, and an output signal for amplifying the output signal after the input signal discriminated to be an audio signal by the input signal discriminating unit And an attenuation unit.
[0015]
  Claim8The invention of claim7In the invention, the output signal attenuating unit changes the attenuation amount of the output signal stepwise when the discrimination result in the input signal discriminating unit is switched.
[0016]
  Claim9The invention of claim6 or 7 or 8In the invention, the input signal determination unit includes an average value ratio calculation unit that calculates a ratio between the long-time average value and the short-time average value of the input signal, and an average value difference calculation that calculates a difference between the long-time average value and the short-time average value. Of the audio signal depending on whether one of the ratio calculated by the average value ratio calculation unit and the difference calculated by the average value difference calculation unit exceeds a predetermined threshold corresponding to each of the ratio. The threshold value in the other is changed based on one of the ratio calculated by the average value ratio calculation unit and the difference calculated by the average value difference calculation unit.
[0017]
  Claim10The invention of claim6 or 7 or 8In the invention, the input signal determining unit determines whether or not the input signal is an audio signal from the frequency analysis result of the input signal.
[0018]
  The invention of claim 11 is claimed in claim5In the invention ofThe gain fluctuation correction unitThe gain variation corrected by the acquisition correction unit is detected, and when the detected variation exceeds a predetermined range, the gain is corrected so as to suppress the variation.RukoAnd features.
[0019]
  Claim12The invention of claim 1 to claim 111In any of the inventions, a noise removing unit for removing a noise component from an input signal is provided, and an output signal of the noise removing unit is used as an input to the short-time average level detecting unit.
[0020]
  Claim13The invention of claim 1 to claim 112The present invention further comprises a noise component removing unit that removes noise components other than the audio signal from the previous input signal input to the noise detecting unit, and the noise component removing unit includes the audio signal in the time axis input signal. A non-speech section detector for detecting non-speech sections, a frequency converter for obtaining a frequency spectrum of an input signal by frequency conversion, and a frequency obtained by the frequency converter for a non-speech section detected by the non-speech section detector A non-speech spectrum estimator that estimates the frequency spectrum of a non-speech signal that is not a speech signal from the spectrum, and a frequency that subtracts the frequency spectrum of the non-speech signal estimated by the non-speech spectrum estimator from the frequency spectrum obtained by the frequency converter Spectral subtraction unit and frequency spectrum subtraction unit to convert time domain signal from frequency spectrum Characterized by comprising the Mel frequency inversion unit.
[0021]
  Claim14The invention of claim 1 to claim 112The noise component removing means removes a noise component other than the voice signal from the previous input signal inputted to the noise detecting section, and the noise component removing means has a frequency for obtaining a frequency spectrum of the input signal by frequency conversion. A conversion unit, a non-speech segment detection unit for detecting a non-speech segment in which the speech signal is not included in the input signal from the frequency spectrum of the input signal obtained from the frequency conversion unit, and a non-speech segment detected by the non-speech segment detection unit A non-speech spectrum estimation unit that estimates a frequency spectrum of a non-speech signal that is not a speech signal from the frequency spectrum obtained by the frequency conversion unit, and a non-speech spectrum estimation unit that estimates the frequency spectrum obtained by the frequency conversion unit. Frequency spectrum subtraction unit that subtracts the frequency spectrum of the audio signal and the frequency obtained by the frequency spectrum subtraction unit Characterized by comprising the frequency inverse conversion unit for obtaining the signal in the time axis by frequency inverse conversion from spectrum.
[0022]
  Claim15The invention of claim14In the invention, a spectrum normalization unit is provided that obtains a normalized spectrum obtained by power normalizing the frequency spectrum obtained from the frequency conversion unit and supplies the normalized spectrum to the non-speech section detection unit.
[0023]
  Claim16The invention of claim13 or 14In the present invention, the non-speech spectrum estimation unit of the noise detection unit is also used as the non-speech spectrum estimation unit of the noise component removing means.
[0024]
  Claim17The invention of claim13 or 14In the present invention, the non-speech section detecting unit of the noise detecting unit is also used as the non-speech section detecting unit of the noise component removing means.
[0025]
DETAILED DESCRIPTION OF THE INVENTION
  Hereinafter, the present invention will be described in detail with reference to the drawings. However, each embodiment described below exemplifies a case where the automatic gain control device of the present invention is applied to a system or device that handles an audio signal. However, the type of input signal to be processed is limited to the audio signal. The technical idea of the present invention can be applied not only to a system but also to all systems and devices having an amplifier for amplifying a signal. In each embodiment, an input signal converted into a digital signal is handled, but the input signal can be handled as an analog signal.Here, before describing the embodiment of the present invention, a basic example having the same basic configuration as the present invention will be described.
[0026]
  (Basic example1)
  BookBasic exampleAs shown in FIG. 1, the automatic gain control apparatus includes an input signal buffer unit 1 that temporarily stores an input signal, and a relatively long time of the input signal (the input signal stored in the input signal buffer unit 1). A long-time average level detection unit 2 that detects an average, a short-time average level detection unit 3 that detects a time average of an input signal (an input signal stored in the input signal buffer unit 1) in a relatively short time, and a long time The gain function Fi (i = 1, 2) is selected from a plurality of gain functions F1, F2,..., Fn prepared in advance according to the long-term average value of the input signal detected by the average level detector 2. ,..., N), and the gain function Fi selected by the gain function selection unit 4 and the short-time average value of the input signal detected by the short-time average level detection unit 3. Gain when amplifying An output value for calculating an output signal from the gain calculation unit 5 to be output and the gain calculated by the gain calculation unit 5 and the input signal stored in the input signal buffer unit 1, in other words, an output value for amplifying the input signal by the calculated gain And a calculation unit 6.
[0027]
The input signal buffer unit 1 stores an input signal (input data) input in a short time of about 16 ms (hereinafter referred to as “buffer time”), and the short-time average level detection unit 3 stores the input signal buffer. The arithmetic average value (short-time average value) of the input signal level for the buffer time stored in the unit 1 is calculated, and the long-time average level detection unit 2 inputs an input signal that is a fixed multiple (for example, 16 times) of the buffer time. The arithmetic average value (long-term average value) of the level is calculated. The long-time average level detection unit 2 includes a buffer having a buffer time (capacity) that is a fixed multiple (for example, 16 times) of the buffer time of the input signal buffer unit 1 in order to calculate a long-time average value. .
[0028]
The gain function selection unit 4 holds data of three types of gain functions F1, F2, and F3 as shown in FIG. The gain function F1 indicated by the solid line in FIG. 2 has the largest gain (slope of the solid line) among the three types, and the input level upper limit L1 is the smallest. Further, the gain function F3 indicated by a broken line in the figure has the smallest gain among the three types and the largest upper limit L3 of the input level. Further, in the gain function F2 indicated by a dashed line in the figure, the upper limit value L2 of the gain and the input level is set to an intermediate value among the three types. However, these gain functions F1 to F3 are examples, and the types and numbers thereof are not limited to those shown in FIG.
[0029]
  Next bookBasic exampleThe operation of will be described.
[0030]
The long-time average value of the input signal obtained by the long-time average level detection unit 2 is given to the gain function selection unit 4, and the gain function selection unit 4 selects an appropriate gain function Fi (i = 1, 1) according to the long-time average value. 2, 3) is selected. Here, taking a telephone call or other communication device as an example, the average value of the long-term average value is generally determined mainly by noise components (components other than voice) included in the input signal, and the short-time average value is mainly The level is considered to be determined by the signal component (sound component) included in the input signal. Therefore, if the long-time average value is a relatively large value, there are many noise components contained in the input signal, and conversely if the long-time average value is a relatively small value, the input signal contains few noise components. Since it can be determined, the gain function selection unit 4 selects the gain function F1 if the long-term average value is relatively small, selects the gain function F3 if it is relatively large, and gains if it is relatively intermediate. Select function F2. The gain calculation unit 5 obtains an output level by substituting the short-time average value calculated by the short-time average level detection unit 3 into the input level of the selected gain function Fi, and determines the output level as the input level (short-time average value). ) To calculate the gain.
[0031]
Then, the output value calculation unit 6 calculates an output value from the gain calculated by the gain calculation unit 5 and the input signal stored in the input signal buffer unit 1, in other words, the input signal with the gain calculated by the gain calculation unit 5. The output signal is obtained by amplifying the signal.
[0032]
Thus, as the noise component included in the input signal, for example, ambient noise of a communication device such as a telephone becomes relatively large, the gain decreases and the upper limit value Li (i = 1, 2, 3) of the input level increases. Since the gain function Fi is selected by the gain function selection unit 4, the output signal level is higher even when the level fluctuation of the input signal is larger than in the conventional example in which gain control is performed only with a single gain function. Gain control can be maintained. In addition, the gain function Fi is selected according to the long-term average value of the input signal determined by the level of the noise component, and the short-time average value determined by the signal component (speech component) other than the noise component and the selected gain function. In order to calculate the gain when amplifying the input signal, a decrease in the signal / noise ratio of the output signal is suppressed as compared with the case where the gain is determined by selecting the gain function Fi according to the level of the input signal itself. Can do.
[0033]
  (Basic example2)
  BookBasic exampleAs shown in FIG. 3, the short time peak detection unit 7 that detects the peak value of the input signal within a unit time (buffer time) when the short time average level detection unit 3 detects the short time average value, It is characterized in that it includes a gain correction unit 8 that corrects the gain calculated by the gain calculation unit 5 when the peak value detected by the short-time peak detection unit 7 exceeds a predetermined threshold value. About the configurationBasic exampleThe description is omitted because it is common to 1.
[0034]
  Thus,Basic example1, when the peak value of the input signal increases to some extent, an overflow may occur when the output value calculation unit 6 calculates the output value. If an overflow occurs in the output value calculation unit 6, so-called clipping noise is generated. May cause a deterioration in call quality.
[0035]
  Book against itBasic exampleThen, when the peak value detected by the short time peak detection unit 7 exceeds a predetermined threshold value, the gain correction unit 8 corrects the gain calculated by the gain calculation unit 5 to a relatively small value. Yes. As a result, even when an input signal having a high level exceeding the threshold value is input, it is possible to prevent clipping noise from occurring in the output signal or distortion of the output signal.
[0036]
  (Basic example3)
  BookBasic example4 includes an input signal discriminating unit 9 that discriminates whether or not the input signal is an audio signal, and changes the gain function selected by the gain function selecting unit 4 according to the discrimination result of the input signal discriminating unit 9 There is a feature in the point thatBasic exampleThe description is omitted because it is common to 2.
[0037]
The input signal discriminating unit 9 can be realized by using a conventionally known technique. For example, the short-time average value obtained by the short-time average level detection unit 3 and the long-time average obtained by the long-time average level detection unit 2 are used. By comparing the ratio with the value to a predetermined threshold value, it is determined whether or not the input signal is an audio signal. For example, H is obtained when the input signal is an audio signal, and L when the input signal is not an audio signal (non-audio signal). A binary signal (discrimination signal) is output.
[0038]
  The gain function selector 4Basic exampleAs described in 1, the gain function F1 is selected if the long-term average value is relatively small, the gain function F3 is selected if it is relatively large, and the gain function F2 is selected if it is relatively intermediate. However, even when the gain function F1 or F2 is selected, when the determination signal output from the input signal determination unit 9 is L, that is, when the input signal is a non-speech signal, the selected gain function F1 or F2 The gain function F3 has a smaller gain than that.
[0039]
Thus, when the functions F1 and F2 having a large gain are selected when the input signal is a non-voice signal, the level of the non-voice signal becomes larger than necessary and the noise on the call increases. Thus, it is determined whether the input signal is an audio signal or a non-audio signal, and in the case of a non-audio signal, the gain function selection unit 4 changes the gain function Fi to a small gain. It is possible to prevent the gain from becoming unnecessarily large and to suppress a decrease in the signal / noise ratio of the input signal.
[0040]
  (Basic example4)
  BookBasic exampleAs shown in FIG. 5, the input signal discriminating unit 9 discriminates whether or not the input signal is an audio signal, and attenuates the output signal after the input signal discriminated by the input signal discriminating unit 9 as a non-audio signal is amplified. The output signal attenuating unit 10 is characterized in that it has a feature, other configurations areBasic exampleThe description is omitted because it is common to 2.
[0041]
  The input signal discrimination unit 9Basic example3, for example, a discrimination signal that is H when the input signal is an audio signal and L when the input signal is a non-audio signal is output. The output signal attenuating unit 10 attenuates the signal (output signal) output from the output value calculating unit 6 when the determination signal output from the input signal determining unit 9 is L, that is, when the input signal is a non-speech signal. It is something to be made. When the input signal is an audio signal, the output signal attenuation unit 10 passes the output signal as it is without being attenuated.
[0042]
  Thus,Basic exampleIf the functions F1 and F2 having large gains are selected when the input signal is a non-speech signal as described in FIG. 3, the level of the non-speech signal becomes larger than necessary and the noise on the call increases. As described above, it is determined whether the input signal is a voice signal or a non-voice signal. If the input signal is a non-voice signal, the output signal is attenuated by the output signal attenuating unit 10, so that the output signal other than the voice signal is attenuated. Therefore, noise can be reduced.
[0043]
By the way, before and after the input signal is switched from a voice signal to a non-voice signal or from a non-voice signal to a voice signal, the level of the output signal varies greatly depending on whether or not the output signal is attenuated by the output signal attenuation unit 10. There is a possibility that the voice quality may be deteriorated due to the voice being interrupted or suddenly increasing.
[0044]
Therefore, in the output signal attenuating unit 10, it is desirable to change the attenuation amount of the output signal stepwise when the discrimination result in the input signal discriminating unit 9 is switched. For example, an input signal as shown in FIG. 6A is input, and as shown in FIG. 6B, the input signal discrimination result by the input signal discrimination unit 9 is switched from a non-speech signal to a voice signal at time t1. Assume that the audio signal is switched to the non-audio signal at time t2. At this time, instead of switching the presence / absence of attenuation by the output signal attenuator 10 at times t1 and t2 as shown in FIG. 4C, a few ms from t1 and t2 as shown in FIG. For several tens of ms, output with an attenuation (indicated as “low attenuation” in FIG. 6) smaller than the attenuation when the input signal is a non-speech signal (indicated as “high attenuation” in FIG. 6). It attenuates the signal.
[0045]
Thus, if the attenuation amount of the output signal by the output signal attenuating unit 10 is changed stepwise when the discrimination result in the input signal discriminating unit 9 is switched, the input signal is switched from the non-speech signal to the audio signal. Therefore, it is possible to improve the call quality by preventing the occurrence of problems such as a decrease in the volume of the head portion of the audio signal.
[0046]
  (Basic example5)
  BookBasic exampleIs characterized by the configuration of the input signal discriminating unit 9, and for other configurations,Basic exampleSince it is common to 3 and 4, illustration and description are omitted.
[0047]
  BookBasic exampleAs shown in FIG. 7, the input signal determination unit 9 in FIG. 7 includes an average value ratio calculation unit 9 a for obtaining a ratio between the long-time average value and the short-time average value (= short-time average value / long-time average value) of the input signal, An average value difference calculating unit 9b that determines a difference between the long-time average value and the short-time average value and compares the difference value with a predetermined threshold value to determine whether the signal is an audio signal, and an average value ratio calculating unit A threshold value determination unit 9c that determines a threshold value in the average value difference calculation unit 9b according to the ratio of the average values obtained in 9a.
[0048]
  That is, with a conventionally known configuration, it is only necessary to determine whether or not the input signal is an audio signal by comparing the calculated value of the average value ratio calculating unit 9a or the average value difference calculating unit 9b with a predetermined threshold value. However, the threshold value determination unit 9c determines the threshold value in the average value difference calculation unit 9b according to the average value ratio obtained by the average value ratio calculation unit 9a.Basic exampleThere are features. For example, when the ratio calculated by the average value ratio calculation unit 9a is larger than 1, it is considered that the input signal is likely to be an audio signal. Conversely, when the ratio is smaller than 1, the input signal is audio. The possibility that it is not a signal is considered high.
[0049]
  BookBasic example, The threshold value determination unit 9c increases the threshold value when the ratio calculated by the average value ratio calculation unit 9a is larger than 1, and conversely the ratio calculated by the average value ratio calculation unit 9a is smaller than 1. In this case, by reducing the threshold value, the influence of the noise component (long-term average value) of the input signal on the determination of whether or not it is an audio signal is reduced.
[0050]
  BookBasic exampleThe threshold value determining unit 9c determines the threshold value in the average value difference calculating unit 9b according to the average value ratio obtained by the average value ratio calculating unit 9a. The threshold value determination unit 9c may determine the threshold value in the average value ratio calculation unit 9a according to the difference value obtained in 9b. Or you may make it determine each threshold value of the average value ratio calculation part 9a and the average value difference calculation part 9b adaptively according to a long-time average value.
[0051]
  (Basic example6)
  BookBasic example8 is characterized in that it includes an input signal discriminating unit 11 that discriminates whether or not the input signal is an audio signal from the frequency analysis result of the input signal, as shown in FIG.Basic exampleThe description is omitted because it is common to 4. That meansBasic exampleThe input signal determination unit 9 in FIG. 4 determines whether the input signal is an audio signal by comparing the ratio between the short time average value and the long time average value with a predetermined threshold, for example. The input signal is distinguished from theBasic exampleThe input signal discriminating unit 11 discriminates the input signal from the result of frequency analysis of the input signal.
[0052]
As shown in FIG. 9, the input signal discrimination unit 11 temporarily analyzes the frequency spectrum of the input signal obtained by the frequency analysis unit 11a and the frequency analysis unit 11a that performs frequency analysis by fast Fourier transform (FFT) of the input signal. Obtained by the delay unit 11b to store, the non-speech component spectrum estimation unit 11c that estimates the frequency spectrum of the component of non-speech (for example, ambient noise collected by the microphone of the other communication device), and the frequency analysis unit 11a A spectrum subtracting unit 11d that subtracts the frequency spectrum of the non-speech component estimated by the non-speech component spectrum estimating unit 11c from the frequency spectrum of the input signal, and the power of the frequency spectrum of the subtraction component (difference) obtained by the spectrum subtracting unit 11d. The subtraction component power calculation unit 11e to be calculated and the power value obtained by the subtraction component power calculation unit 11e are A sound signal is judged to be a sound signal when a certain threshold value is exceeded, and a non-voice signal is judged to be a non-voice signal when the threshold value is not exceeded (for example, “H” for a sound signal and “L” for a non-voice signal) A discrimination signal output unit 11f that outputs a signal).
[0053]
In the non-speech signal spectrum estimation unit 11c, when the discrimination signal of L level is output from the discrimination signal output unit 11f, that is, when the input signal is determined to be a non-speech signal, the non-speech estimated up to that point The average of the frequency spectrum of the component and the frequency spectrum of the input signal stored in the delay unit 11b is obtained, and the average value is used as the estimated value of the frequency spectrum of the non-voice component thereafter. In addition, when obtaining the average of the frequency spectrum of the non-speech component and the frequency spectrum of the input signal, it can be obtained by simple addition and division. However, the weight spectrum is multiplied and then added and divided. May be.
[0054]
  Book as aboveBasic exampleThe input signal discriminating unit 11 in FIG. 3 compares the input signal with the input signal discriminating unit 9 that treats the input signal as a time-series signal and discriminates whether or not it is an audio signal. Thus, there is an advantage that the accuracy of determination as to whether or not the sound signal is improved.
[0055]
  (Embodiment1)
  In the present embodiment, as shown in FIG. 10, the gain fluctuation that is corrected by the gain correction unit 8 is detected, and the gain fluctuation is corrected so that the fluctuation is suppressed when the detected fluctuation exceeds a predetermined range. There is a feature in that the correction unit 12 is provided, and for other configurations, for example,Basic example6 is the same as that of FIG.
[0056]
Usually, when the input signal is an audio signal, the signal level increases. Therefore, in order to keep the output signal constant, the gain tends to decrease as the level of the audio signal increases. On the other hand, the gain is considered to vary greatly when the input signal is switched from a non-speech signal to a speech signal or from a speech signal to a non-speech signal. Therefore, the gain tends to increase greatly at the time when the input signal is switched from the audio signal to the non-audio signal, and noise may be amplified if the gain remains at a large value after switching to the non-audio signal. is there.
[0057]
  Therefore, in the present embodiment, the gain variation corrected by the gain correction unit 8 is detected by the gain variation correction unit 12, and when the detected variation exceeds a predetermined range, the gain is increased in a direction to suppress the variation. I am trying to correct it. For example, the gain variation correction unit 12 detects the time point when the input signal switches from the audio signal to the non-speech signal based on the gain variation, and corrects and decreases the gain after switching to the non-speech signal. It is possible to prevent the non-voice signal (noise) from being greatly amplified and to prevent the signal / noise ratio of the input signal from being lowered. In this embodiment, the gain corrected by the gain correction unit 8 is corrected by the gain variation correction unit 12, but the gain calculated by the gain calculation unit 5 is corrected by the gain variation correction unit 12, for example, ,Basic exampleEven when the gain variation correction unit 12 is added to the configuration of 1, the same effect can be obtained.
[0058]
  (Embodiment2)
  As shown in FIG. 11, the present embodiment is characterized by including a noise removing unit 13 that removes a noise component from an input signal, and using the output signal of the noise removing unit 13 as an input to the short-time average level detecting unit 3. Other configurations are embodiments.1Since it is common, explanation is omitted.
[0059]
As shown in FIG. 12, the noise removing unit 13 performs a Fourier transform on the input signal to obtain a frequency spectrum, and a delay unit 13b that temporarily stores the frequency spectrum of the input signal obtained by the Fourier transform unit 13a. A non-speech component spectrum estimation unit 13c that estimates a frequency spectrum of a component of non-speech (for example, ambient noise collected by a microphone of a partner communication device), and a frequency spectrum of an input signal obtained by the Fourier transform unit 13a A subtractor 13d for subtracting the frequency spectrum of the non-speech component estimated by the non-speech component spectrum estimator 13c, and an inverse Fourier transform unit 13e for performing an inverse Fourier transform on the subtracted component (difference) obtained by the spectrum subtractor 13d. The signal output from the inverse Fourier transform unit 13e is miscellaneous from the input signal. A signal component (component of the ambient noise) has been removed. Here, in the non-speech signal spectrum estimation unit 13c, when the L-level discrimination signal is output from the input signal discrimination unit 11, that is, when the input signal is discriminated as a non-speech signal, estimation is performed up to that point. The average of the frequency spectrum of the non-speech component and the frequency spectrum of the input signal stored in the delay unit 13b is obtained, and the average value is used as the estimated value of the frequency spectrum of the non-speech component thereafter.
[0060]
If the signal from which the noise component is removed from the input signal by the noise removing unit 13 as described above is used as the input to the short-time average level detecting unit 3, the short-time average is detected from the input signal after the noise component is removed. Thus, an accurate gain can be calculated without being affected by noise components.
[0061]
  (Embodiment3)
  The present embodiment is characterized in that the long-time average level detection unit 2 ′ detects the long-time average value from the estimated value of the noise component included in the input signal as shown in FIG. Embodiment2Since it is common, explanation is omitted.
[0062]
The long-time average level detection unit 2 ′ in the present embodiment returns the frequency spectrum of the non-speech component (ambient noise) estimated by the non-speech component spectrum estimation unit 13c of the noise removal unit 13 to the time axis signal by inverse Fourier transform. Then, a long-term average value is obtained for the signal.
[0063]
Thus, in the long-time average level detection unit 2 in the previous embodiment, since the simple arithmetic average of the input signal is set to the long-time average value, the rate of including not only noise components but also speech components is high. However, since the long-term average level detection unit 2 ′ in the present embodiment obtains the long-term average value from the estimated value of the noise component included in the input signal, the rate at which the voice component is included can be significantly reduced. Thus, the long-term average value of the input signal determined by the level of the noise component can be accurately detected, and the signal / noise ratio can be improved. Further, since the long-time average value can be accurately detected, an appropriate gain can be calculated so as to suppress amplification of noise components included in the input signal.
[0064]
  (Embodiment4)
  In the present embodiment, as shown in FIG. 14, the long time average level detection unit 2 ″ averages a plurality of short time average values obtained by the short time average level detection unit 3 to detect the long time average of the input signal. There is a feature in the point, and for other configurations, the embodiment2Since it is common, explanation is omitted.
[0065]
  Basic Examples 1 to 6 and Embodiments 1 and 2As shown in FIG. 16A, when the input signal buffer unit 1 has a buffer length of, for example, 128 bits, the long-time average level detection unit 2 has x times the input signal buffer unit 1 (for example, , 16 times) buffer length (128 × 16 = 2048 bits).
[0066]
On the other hand, since the long-time average level detection unit 2 ″ in this embodiment calculates the long-time average value from the short-time average value detected by the short-time average level detection unit 3, as shown in FIG. It is only necessary that x (for example, 16) short-time average values can be stored in the buffer of the long-time average level detection unit 2 ″. Therefore, in the present embodiment, the buffer length of the long-time average level detection unit 2 ″ can be greatly reduced (in the above example, from 2048 bits to 16 bits). There is an advantage that the buffer capacity is reduced, and the arithmetic average obtained for each y of the total x short-time average values is stored in the buffer, and the average value of these y values is averaged for a long time. If the value is obtained, the buffer capacity of the long-time average level detection unit 2 ″ can be further reduced.
[0067]
  (Embodiment5)
  In this embodiment, as shown in FIG. 17, the long-time average level detection unit 2 ″ detects the long-time average of the input signal only when the input signal determination unit 11 determines that it is not a voice signal, that is, a non-voice signal. There is a feature in the point to do, for the configuration other than this embodiment4Since it is common, description is abbreviate | omitted.
[0068]
Thus, the long time average level detection unit 2 ″ does not calculate the long time average value when the determination signal output from the input signal determination unit 11 is at the H level (when the input signal is an audio signal). Since the long-term average value is calculated when the signal is at L level (when the input signal is a non-speech signal), the background noise component contained in the voice signal can be detected as the long-term average value, and the background noise component It is possible to select an appropriate gain function according to the level.
[0069]
  (Embodiment6)
  The present embodiment is characterized in that noise component removing means for removing noise components other than the audio signal from the input signal is provided in the previous stage of the input signal buffer unit 1. However, regarding the configuration and operation after the input signal buffer unit 1Basic exampleThe description is omitted because it is common to 1.
[0070]
As shown in FIG. 18, the noise component removing means in the present embodiment includes a non-speech section detector 14 that detects a non-speech section in which a speech signal is not included in the time-axis input signal, and a frequency spectrum of the input signal by frequency conversion And a non-speech spectrum estimation unit that estimates a frequency spectrum of a non-speech signal that is not a speech signal from a frequency spectrum obtained by the frequency transform unit 15 in a non-speech segment detected by the non-speech segment detection unit 14 16, a frequency spectrum subtraction unit 17 that subtracts the frequency spectrum of the non-speech signal estimated by the non-speech spectrum estimation unit 16 from the frequency spectrum obtained by the frequency conversion unit 15, and a frequency spectrum obtained by the frequency spectrum subtraction unit 17 And a frequency inverse transform unit 18 for obtaining a time-axis signal by frequency inverse transform.
[0071]
The non-speech section detection unit 14 can be realized by using a conventionally known technique. For example, the non-speech section detection unit 14 estimates the instantaneous power of a speech component included in the input signal and estimates the level of the non-speech component (background noise component). Based on the ratio between the instantaneous power estimation value and the non-speech component level, a section (non-speech section) that can be regarded as containing no speech component in the input signal is detected. The frequency converter 15 obtains the frequency spectrum of the input signal using Fourier transform, wavelet transform, or the like.
[0072]
The non-speech spectrum estimation unit 16 obtains an average value of frequency spectra corresponding to the input signal of the non-speech section detected by the non-speech section detection unit 14 among the frequency spectra of the input signal obtained by the frequency conversion unit 15; The average value is set as the frequency spectrum estimation value of the non-speech signal. Then, by subtracting the frequency spectrum estimation value of the non-speech signal estimated by the non-speech spectrum estimation unit 16 from the frequency spectrum of the input signal obtained by the frequency conversion unit 15 in the spectrum subtraction unit 17, noise components other than the speech component are subtracted. It can be removed from the input signal. Then, frequency inverse transformation is performed on the frequency spectrum output from the spectrum subtraction unit 17 in the frequency inverse transformation unit 18 to obtain an input signal from which noise components other than the audio signal are removed.
[0073]
Thus, in the present embodiment, noise components other than the voice component included in the input signal are removed by the noise component removing unit in the previous stage of amplification, thereby preventing the quality deterioration of the voice signal due to the amplification of the noise component. be able to.
[0074]
  (Embodiment7)
  This embodiment is an embodiment as shown in FIG.6Is characterized in that a noise detection unit including a non-speech section detection unit 14, a frequency conversion unit 15, and a non-speech spectrum estimation unit 16 is used instead of the long-time average level detection unit 2. However, for other configurations and operationsBasic example1 and embodiments6Since it is common, description is abbreviate | omitted.
[0075]
  The non-speech spectrum estimation unit 16 of the present embodiment obtains the power spectrum of the frequency spectrum in the non-speech section, and further calculates the average value of the power spectrum for each band. Here, since the level of the average value of the power spectrum calculated by the non-speech spectrum estimation unit 16 is determined by the noise component included in the input signal,Basic exampleThe gain function selection unit 4 can select an appropriate gain function Fi (i = 1, 2, 3) according to the average value of the power spectrum instead of the long-time average value in 1.
[0076]
  Thus, as the noise component included in the input signal, for example, ambient noise of a communication device such as a telephone becomes relatively large, the gain decreases and the upper limit value Li (i = 1, 2, 3) of the input level increases. By selecting the gain function Fi in the gain function selection unit 4 as compared with the conventional example in which the gain control is performed only with a single gain function,Basic exampleAs in the case of 1, the gain control can be performed to keep the output signal level constant even when the level fluctuation of the input signal is large. In addition, the gain function Fi is selected according to the power spectrum average value of the frequency spectrum in the non-speech section of the input signal determined by the level of the noise component, and the short-time average value determined by the signal component (speech component) other than the noise component In order to calculate the gain when amplifying the input signal from the selected gain function, the signal / of the output signal is compared with the case where the gain is determined by selecting the gain function Fi according to the level of the input signal itself. A decrease in noise ratio can be suppressed. In the present embodiment, there is an advantage that the configuration can be simplified because a part of the noise component removing means is also used as the noise detection unit.
[0077]
  (Embodiment8)
  FIG. 20 shows a block diagram of the present embodiment. This embodimentBasic example4, the detection result of the non-speech segment by the non-speech segment detector 14 is input to the output signal attenuator 10 instead of the input signal discrimination unit 9 in FIG. 4, and the output signal after the input signal in the non-speech segment is amplified is output signal attenuated It is characterized in that it is attenuated by the part 10. Other configurations and operations areBasic example4 and embodiments7Since it is common, description is abbreviate | omitted.
[0078]
  Thus, in this embodiment,Basic example4, the output signal after amplifying the input signal in the non-speech section is attenuated by the output signal attenuating unit 10, so that the output signal other than the audio signal can be attenuated to reduce noise. There is an advantage that the configuration can be simplified by using a part of the component removing means instead of the input signal determining unit 9.
[0079]
  (Embodiment9)
  FIG. 21 shows a block diagram of the present embodiment. This embodimentBasic example3, the detection result of the non-speech section by the non-speech section detection unit 14 is input to the gain function selection unit 4 instead of the input signal determination unit 9 in FIG. 3, and the gain function selection unit 4 according to the detection result of the non-speech section detection unit 14 Is different in the gain function to be selected.Basic example3 and embodiments8Since it is common, explanation is omitted.
[0080]
  Thus, in this embodiment,Basic example3, the gain function selection unit 4 changes the gain function Fi with a small gain to the input signal in the non-speech section, thereby preventing the gain for the non-speech signal from being increased more than necessary. Further, there is an advantage that the configuration can be simplified by using a part of the noise component removing means instead of the input signal determining unit 9.
[0081]
  (Embodiment10)
  FIG. 22 shows a block diagram of the present embodiment. The present embodiment is characterized in that the non-speech section detection unit 14 ′ constituting the noise component removing unit detects a non-speech section in which the speech signal is not included in the input signal from the frequency spectrum of the input signal obtained from the frequency conversion unit 15. There are features, and other configurations and operations are the embodiments6Since it is common, explanation is omitted.
[0082]
The non-speech section detector 14 ′ in this embodiment compares the frequency spectrum of the input signal obtained by the frequency converter 15 with a frequency spectrum model of non-speech (such as ambient noise) prepared in advance, and the frequency spectrum model In accordance with the degree of coincidence, it is determined whether it is voice or non-voice for each band, and when a certain number of bands or more are determined as non-voice, the section is determined as a non-voice section, and An average of the frequency spectrum and the frequency spectrum model is taken to update the non-voice frequency spectrum model. However, the non-speech section detection unit 14 'determines that the band is non-speech when the band determined to be speech by the above-described method deviates from the desired band. Hereinafter, the frequency spectrum estimation value of the non-speech signal is obtained by the non-speech spectrum estimation unit 16 from the frequency spectrum corresponding to the input signal of the non-speech segment detected by the non-speech segment detection unit 14 ′, and the frequency subtraction unit 17 By subtracting the frequency spectrum estimation value of the non-speech signal estimated by the non-speech spectrum estimation unit 16 from the frequency spectrum of the input signal obtained by the conversion unit 15, noise components other than the speech component are removed from the input signal, and further the frequency By performing frequency inverse transform on the frequency spectrum output from the spectrum subtracting unit 17 in the inverse transforming unit 18, an input signal from which noise components other than the audio signal are removed is obtained.
[0083]
  Thus, this embodiment is also an embodiment.6Similarly, since noise components other than the voice component included in the input signal are removed by the noise component removing means in the previous stage of amplification, qualitative deterioration of the voice signal due to amplification of the noise component can be prevented. In addition, the non-speech section detector 14 ′ in this embodiment detects a non-speech section by determining whether the frequency spectrum of the input signal is speech or non-speech for each frequency band.6There is an advantage that the non-speech section detection accuracy is higher than that of the non-speech section detector 14 in FIG.
[0084]
  (Embodiment11)
  FIG. 23 shows a block diagram of this embodiment. The present embodiment is characterized in that it includes a spectrum normalization unit 19 that obtains a normalized spectrum obtained by power normalizing the frequency spectrum obtained from the frequency conversion unit 15 and supplies the normalized spectrum to the non-speech section detection unit 14 ′. Configuration and operation of the embodiment10Since it is common, explanation is omitted.
[0085]
The spectrum normalization unit 19 calculates the power spectrum of the frequency spectrum of the input signal obtained by the frequency conversion unit 15 to obtain an average value thereof, and further divides (normalizes) the power spectrum value of each frequency band by the average value. To perform the process (power normalization process). Hereinafter, the non-speech section detection unit 14 ′ detects the non-speech section of the input signal, and the frequency spectrum corresponding to the input signal in the non-speech section detected by the non-speech section detection unit 14 ′ is transferred to the non-speech spectrum estimation unit 16. Thus, the frequency spectrum estimation value of the non-speech signal is obtained, and the spectrum subtraction unit 17 subtracts the frequency spectrum estimate value of the non-speech signal estimated by the non-speech spectrum estimation unit 16 from the frequency spectrum of the input signal obtained by the frequency conversion unit 15. Thus, noise components other than the speech component are removed from the input signal, and the frequency inverse transform unit 18 performs frequency inverse transform on the frequency spectrum output from the spectrum subtracting unit 17, thereby removing the noise component other than the speech signal. An input signal from which is removed is obtained.
[0086]
  Thus, this embodiment is also an embodiment.6 or 7Similarly, since noise components other than the voice component included in the input signal are removed by the noise component removing means in the previous stage of amplification, qualitative deterioration of the voice signal due to amplification of the noise component can be prevented. In addition, since the spectrum normalization unit 19 is provided in the present embodiment, characteristics (peak portions and valley portions) in the shape of the frequency spectrum are clarified by the power normalization process, and the non-speech interval detection unit 14 ′ There is an advantage that the detection system of non-speech section by can be improved.
[0087]
  (Embodiment12)
  FIG. 24 shows a block diagram of this embodiment. This embodiment is an embodiment11Is characterized in that a noise detection unit including a spectrum normalization unit 19, a non-speech interval detection unit 14 ′, a frequency conversion unit 15, and a non-speech spectrum estimation unit 16 is used instead of the long-time average level detection unit 2, For other configurations and operations, the embodiment11And embodiments7Since it is common, description is abbreviate | omitted.
[0088]
Thus, in the present embodiment, the detection accuracy of the non-voice section of the input signal is improved, so that the gain function selection unit 4 can be prevented from making a mistake in selecting the gain function, and an appropriate gain function corresponding to the noise component included in the input signal can be prevented. Can be selected.
[0089]
  (Embodiment13)
  FIG. 25 shows a block diagram of the present embodiment. This embodiment is an embodiment8And embodiment11And the detection result of the non-speech segment by the non-speech segment detector 14 'is input to the output signal attenuator 10, and the output signal after amplifying the input signal in the non-speech segment is output to the output signal attenuator 10 is characterized by attenuation. Other configurations and operations are the embodiment.8And embodiment11Since it is common, description is abbreviate | omitted.
[0090]
Thus, in the present embodiment, by improving the detection accuracy of the non-voice section of the input signal, the noise reduction effect due to the attenuation of the output signal other than the voice signal by the output signal attenuator 10 can be enhanced and the configuration is simplified. There is an advantage that can be achieved.
[0091]
  (Embodiment14)
  FIG. 26 shows a block diagram of this embodiment. This embodiment is an embodiment9And embodiment11And the detection result of the non-speech segment by the non-speech segment detector 14 ′ is input to the gain function selector 4 and the gain function selector 4 according to the detection result of the non-speech segment detector 14 ′. The feature is that the gain function selected by is changed. For other configurations, the embodiment9And embodiment11Since it is common, explanation is omitted.
[0092]
Thus, in this embodiment, by improving the detection accuracy of the non-voice section of the input signal, the gain function selection system in the gain function selection unit 4 is improved, and the signal / noise ratio of the input signal is more reliably reduced. There is an advantage that the configuration can be simplified and the configuration can be simplified.
[0093]
【The invention's effect】
  The invention according to claim 1 is an automatic gain control device for controlling a gain when amplifying the input signal in accordance with the level of the input signal, a noise detection unit for detecting a noise component included in the input signal, One of a short-time average level detection unit for detecting a time average in a short time of about 16 milliseconds and a plurality of gain functions prepared in advance according to the magnitude of the noise component detected by the noise detection unit. Gain function selection unit for selecting the gain function, short-time average value of the input signal detected by the short-time average level detection unit, and gain for amplifying the input signal from the gain function selected by the gain function selection unit A gain calculatorA gain fluctuation correction unit that detects a fluctuation amount of the gain calculated by the gain calculation unit and corrects the gain in a direction to suppress the fluctuation when the detected fluctuation amount exceeds a predetermined range;The noise detection unit includes a long-time average level detection unit that detects a time average in a time obtained by multiplying the short time of the input signal by 16 times, and the gain function selection unit is detected by the long-time average level detection unit. The long-time average level detection unit estimates a noise component included in the input signal by selecting any one of a plurality of gain functions prepared in advance according to the long-time average value of the input signal. It is characterized by detecting the long-term average value from the value, and the level of the short-time average value is determined mainly by the signal components other than the noise component contained in the input signal. In response, the gain is selected with only a single gain function by selecting one of a plurality of gain functions and amplifying the input signal with the gain calculated from the selected gain function and the short-time average value. Compared to the conventional example in which the control, it is possible to gain control to maintain the output signal level even when the level variation is large in the input signal constant. In addition, the gain function is selected according to the degree of noise contained in the input signal, and the gain for amplifying the input signal is calculated from the short-time average value determined by the signal component other than the noise component and the selected gain function. Therefore, a decrease in the signal / noise ratio of the output signal can be suppressed as compared with the case where the gain is determined by selecting the gain function according to the level of the input signal itself. In addition, the long-term average value is determined mainly by the noise component contained in the input signal, so that it can be used as a value representing the degree of noise contained in the input signal, and the degree of noise contained in the input signal is relatively low. It can be easily detected. Further, the long-term average value of the input signal determined by the level of the noise component can be accurately detected, and the signal / noise ratio can be improved.Moreover, since the fluctuation of the gain generally increases when the state of the input signal changes, the noise component is amplified by correcting the gain in a direction to suppress the fluctuation when the fluctuation of the gain exceeds a predetermined range. The signal / noise ratio can be prevented from decreasing.
[0096]
  Claim2The invention of claim1In the invention, the long-time average level detection unit detects a long-time average of the input signal by averaging a plurality of short-time average values obtained by the short-time average level detection unit, and directly from the input signal Compared with the case of detecting the average for a long time, the labor and configuration required for detection can be simplified.
[0097]
  Claim3The invention of claim1 or 2In the invention, an input signal discriminating unit for discriminating whether or not the input signal is an audio signal is provided, and the long-time average level detecting unit determines the length of the input signal only when the input signal discriminating unit discriminates that the input signal is not an audio signal. It is characterized by detecting the time average, the background noise component included in the audio signal can be detected as the long-term average value, and an appropriate gain function can be selected according to the level of the background noise component.
[0098]
  Claim4In the invention of claim 1, in the invention of claim 1, the noise detection unit includes a non-speech section detection unit that detects a non-speech section in which an input signal does not include a speech signal, a frequency conversion unit that obtains a frequency spectrum of the input signal, A non-speech spectrum estimation unit that estimates a frequency spectrum of a non-speech signal that is not a speech signal from a frequency spectrum obtained by a frequency conversion unit in a non-speech segment detected by a non-speech segment detection unit; The frequency spectrum of the non-speech signal is characterized by selecting any one gain function from a plurality of gain functions prepared in advance according to the frequency spectrum of the non-speech signal estimated by the non-speech spectrum estimation unit. Since its level is determined mainly by the noise component contained in the input signal, it can be used as a value that represents the degree of noise contained in the input signal. The degree of noise included in the relatively easy to detect.
[0099]
  Claim5The invention of claim 1 to claim 14In any of the inventions, the short-time peak detection unit that detects the peak value of the input signal within a unit time when the short-time average level detection unit detects the short-time average value, and the short-time peak detection unit detects And a gain correction unit that corrects the gain calculated by the gain calculation unit when the peak value to be exceeded exceeds a predetermined threshold value, and an input signal having a high level exceeding the threshold value is provided. It is possible to prevent the output signal from being distorted when input.
[0100]
  Claim6The invention of claim 1 to claim 15In any of the inventions, an input signal discriminating unit that discriminates whether or not the input signal is an audio signal, and the gain function selecting unit changes the gain function to be selected according to the discrimination result of the input signal discriminating unit. As a feature, it is possible to prevent a gain for a non-voice signal that is not a voice signal from becoming unnecessarily large, and to suppress a decrease in the signal / noise ratio of the output signal.
[0101]
  Claim7The invention of claim 1 to claim 16In any of the inventions, an input signal discriminating unit for discriminating whether or not the input signal is an audio signal, and an output signal for amplifying the output signal after the input signal discriminated to be an audio signal by the input signal discriminating unit An attenuation unit is provided, and noise can be reduced by attenuating output signals other than audio signals.
[0102]
  Claim8The invention of claim7In the invention, the output signal attenuating unit changes the attenuation amount of the output signal stepwise when the discrimination result in the input signal discriminating unit is switched, and the input signal is switched from the non-speech signal to the audio signal. Therefore, it is possible to improve the call quality by preventing the occurrence of problems such as a decrease in the volume of the head portion of the audio signal.
[0103]
  Claim9The invention of claim6 or 7 or 8In the invention, the input signal determination unit includes an average value ratio calculation unit that calculates a ratio between the long-time average value and the short-time average value of the input signal, and an average value difference calculation that calculates a difference between the long-time average value and the short-time average value. Of the audio signal depending on whether one of the ratio calculated by the average value ratio calculation unit and the difference calculated by the average value difference calculation unit exceeds a predetermined threshold corresponding to each of the ratio. It is characterized in that the threshold value in the other is changed based on one value of the ratio calculated by the average value ratio calculation unit and the difference calculated by the average value difference calculation unit, and By changing the threshold for determining the input signal when the time average value is relatively larger than the short-time average value, the influence of the noise component of the input signal on the determination of whether or not it is an audio signal is reduced. can do.
[0104]
  Claim10The invention of claim6 or 7 or 8In the invention, the input signal determining unit determines whether or not the input signal is an audio signal from the frequency analysis result of the input signal, and the accuracy of determining whether or not the input signal is an audio signal is improved.
[0105]
  The invention of claim 11 is claimed in claim5In the invention ofThe gain fluctuation correction unitThe gain variation corrected by the acquisition correction unit is detected, and when the detected variation exceeds a predetermined range, the gain is corrected so as to suppress the variation.RukoIn general, the gain fluctuation increases when the state of the input signal changes, so that the noise component is reduced by correcting the gain in a direction to suppress the fluctuation when the gain fluctuation exceeds a predetermined range. The signal / noise ratio can be prevented from being lowered by suppressing amplification.
[0106]
  Claim12The invention of claim 1 to claim 111In any of the inventions, a noise removal unit that removes a noise component from the input signal is provided, and the output signal of the noise removal unit is used as an input to the short-time average level detection unit, and after the noise component is removed By detecting the average for a short time from the input signal, an accurate gain can be calculated without being affected by the noise component.
[0107]
  Claim13The invention of claim 1 to claim 112The present invention further comprises a noise component removing unit that removes noise components other than the audio signal from the previous input signal input to the noise detecting unit, and the noise component removing unit includes the audio signal in the time axis input signal. A non-speech section detector for detecting non-speech sections, a frequency converter for obtaining a frequency spectrum of an input signal by frequency conversion, and a frequency obtained by the frequency converter for a non-speech section detected by the non-speech section detector A non-speech spectrum estimator that estimates the frequency spectrum of a non-speech signal that is not a speech signal from the spectrum, and a frequency that subtracts the frequency spectrum of the non-speech signal estimated by the non-speech spectrum estimator from the frequency spectrum obtained by the frequency converter Spectral subtraction unit and frequency spectrum subtraction unit to convert time domain signal from frequency spectrum Characterized by comprising the Mel frequency inverse transform unit, because removed from the input signal a noise component other than the voice signals, it is possible to prevent the quality degradation of the audio signal by the amplification of the noise component.
[0108]
  Claim14The invention of claim 1 to claim 112The noise component removing means removes a noise component other than the voice signal from the previous input signal inputted to the noise detecting section, and the noise component removing means has a frequency for obtaining a frequency spectrum of the input signal by frequency conversion. A conversion unit, a non-speech segment detection unit for detecting a non-speech segment in which the speech signal is not included in the input signal from the frequency spectrum of the input signal obtained from the frequency conversion unit, and a non-speech segment detected by the non-speech segment detection unit A non-speech spectrum estimation unit that estimates a frequency spectrum of a non-speech signal that is not a speech signal from the frequency spectrum obtained by the frequency conversion unit, and a non-speech spectrum estimation unit that estimates the frequency spectrum obtained by the frequency conversion unit. Frequency spectrum subtraction unit that subtracts the frequency spectrum of the audio signal and the frequency obtained by the frequency spectrum subtraction unit It consists of a frequency inverse transform unit that obtains a time-axis signal from a spectrum by frequency inverse transform, and removes noise components other than the speech signal from the input signal. Can be prevented. Further, the non-speech section detection unit can improve the detection accuracy by detecting the non-speech section from the frequency spectrum of the input signal.
[0109]
  Claim15The invention of claim14In the invention of claim 1, characterized in that it comprises a spectrum normalization unit that obtains a normalized spectrum obtained by power normalizing the frequency spectrum obtained from the frequency conversion unit and gives the normalized spectrum to the non-speech section detection unit. The characteristic of the spectrum becomes clear and the detection accuracy of the non-speech segment by the non-speech segment detector is improved.
[0110]
  Claim16The invention of claim13 or 14In the present invention, the non-speech spectrum estimation unit of the noise detection unit is also used as the non-speech spectrum estimation unit of the noise component removal means, and the configuration can be simplified.
[0111]
  Claim17The invention of claim13 or 14In the present invention, the non-speech section detecting unit of the noise detecting unit is also used as the non-speech section detecting unit of the noise component removing means, and the configuration can be simplified.
[Brief description of the drawings]
[Figure 1]Basic example of the present invention1 is a block diagram showing 1. FIG.
FIG. 2 is an explanatory diagram of a gain function in the same as above.
[Fig. 3]Basic example2 is a block diagram showing 2. FIG.
[Fig. 4]Basic example3 is a block diagram showing 3. FIG.
[Figure 5]Basic example4 is a block diagram showing 4. FIG.
FIG. 6 is an explanatory diagram of the above.
[Fig. 7]Basic example5 is a block diagram of an input signal determination unit in FIG.
[Fig. 8]Basic example6 is a block diagram showing 6. FIG.
FIG. 9 is a block diagram of an input signal determination unit in the same as above.
FIG. 10 shows an embodiment.1FIG.
FIG. 11 shows an embodiment.2FIG.
FIG. 12 is a block diagram of a noise removing unit in the same as above.
FIG. 13 is an embodiment.3FIG.
FIG. 14 shows an embodiment.4FIG.
FIG. 15 is an operation explanatory diagram of the long-term average level detection unit in the above.
FIG. 16 is an operation explanatory diagram of the long-term average level detection unit in the above.
FIG. 17 is an embodiment.5FIG.
FIG. 18 shows an embodiment.6FIG.
FIG. 19 shows an embodiment.7FIG.
FIG. 20 shows an embodiment.8FIG.
FIG. 21 Embodiment9FIG.
FIG. 22 shows an embodiment.10FIG.
FIG. 23 Embodiment11FIG.
FIG. 24 Embodiment12FIG.
FIG. 25 is an embodiment.13FIG.
FIG. 26 Embodiment14FIG.
FIG. 27 is an explanatory diagram of a gain function in a conventional automatic gain control device.
[Explanation of symbols]
1 Input signal buffer
2 Long-time average level detector
3 Short-time average level detector
4 Gain function selector
5 Gain calculator
6 Output value calculator

Claims (17)

In an automatic gain control device that controls a gain when amplifying an input signal according to the level of the input signal, a noise detection unit that detects a noise component included in the input signal, and a short time of about 16 milliseconds of the input signal A short-time average level detection unit that detects a time average, and a gain function that selects any one gain function from a plurality of gain functions prepared in advance according to the magnitude of the noise component detected by the noise detection unit A gain calculating unit that calculates a gain when the input signal is amplified from a selection unit, a short-time average value of the input signal detected by the short-time average level detection unit, and a gain function selected by the gain function selection unit ; variation detected detects the variation of the calculated gain calculating section and a gain variation correction unit for correcting the gain to suppress direction change when it exceeds a predetermined range, the noise detector, A long-time average level detection unit that detects a time average of a force signal obtained by multiplying the short time by 16 times, and the gain function selection unit sets the long-time average value of the input signal detected by the long-time average level detection unit In response, one of the gain functions prepared in advance is selected, and the long-term average level detection unit detects the long-term average value from the estimated noise component contained in the input signal. An automatic gain control device.   2. The automatic gain control according to claim 1, wherein the long time average level detection unit detects a long time average of the input signal by averaging a plurality of short time average values obtained by the short time average level detection unit. apparatus.   An input signal discriminating unit that discriminates whether or not the input signal is an audio signal, and the long-term average level detection unit detects the long-term average of the input signal only when the input signal discriminating unit discriminates that it is not an audio signal The automatic gain control device according to claim 1 or 2, wherein   The noise detector includes a non-speech section detector that detects a non-speech section in which an input signal does not include a speech signal, a frequency converter that obtains a frequency spectrum of the input signal, and a non-speech detected by the non-speech section detector A non-speech spectrum estimation unit that estimates a frequency spectrum of a non-speech signal that is not a speech signal from a frequency spectrum obtained by a frequency conversion unit in a section, and the gain function selection unit is a non-speech that is estimated by the non-speech spectrum estimation unit 2. The automatic gain control apparatus according to claim 1, wherein any one of a plurality of gain functions prepared in advance according to the frequency spectrum of the signal is selected.   The short-time peak detection unit for detecting the peak value of the input signal within a unit time when the short-time average level detection unit detects the short-time average value, and the peak value detected by the short-time peak detection unit are predetermined. 5. The automatic gain control device according to claim 1, further comprising a gain correction unit that corrects the gain calculated by the gain calculation unit when the threshold value is exceeded.   The input signal discriminating unit for discriminating whether or not the input signal is an audio signal, and the gain function selecting unit changes a gain function to be selected according to the discrimination result of the input signal discriminating unit. 6. The automatic gain control device according to any one of 5 above.   An input signal discriminating unit that discriminates whether or not the input signal is an audio signal, and an output signal attenuating unit that attenuates the output signal after amplifying the input signal that is discriminated to be an audio signal by the input signal discriminating unit An automatic gain control device according to any one of claims 1 to 6.   8. The automatic gain control device according to claim 7, wherein the output signal attenuating unit changes the attenuation amount of the output signal stepwise when the discrimination result in the input signal discriminating unit is switched.   The input signal determination unit includes an average value ratio calculation unit that calculates a ratio between the long-time average value and the short-time average value of the input signal, and an average value difference calculation unit that calculates a difference between the long-time average value and the short-time average value. And determining the audio signal based on whether one of the ratio calculated by the average value ratio calculation unit and the difference calculated by the average value difference calculation unit exceeds a predetermined threshold corresponding to each of the ratios. The threshold value in the other is changed based on one of the ratio calculated by the average value ratio calculation unit and the difference calculated by the average value difference calculation unit. Or the automatic gain control apparatus of 8.   9. The automatic gain control apparatus according to claim 6, 7 or 8, wherein the input signal determination unit determines whether or not the input signal is a voice signal from a frequency analysis result of the input signal. Said gain variation correction portion includes a characterized and Turkey to correct the gain to suppress direction change when the variation was detected and detects the variation of the corrected gain gain correction unit exceeds a predetermined range The automatic gain control device according to claim 5 .   The automatic gain control according to any one of claims 1 to 11, further comprising a noise removing unit that removes a noise component from the input signal, wherein the output signal of the noise removing unit is used as an input to the short-time average level detecting unit. apparatus.   A noise component removing unit that removes a noise component other than a voice signal from a previous input signal input to the noise detection unit, and the noise component removing unit includes a non-voice section in which the voice signal is not included in the time axis input signal; A non-speech section detecting unit for detecting a speech signal, a frequency converting unit for obtaining a frequency spectrum of an input signal by frequency conversion, and a speech signal from the frequency spectrum obtained by the frequency converting unit for a non-speech section detected by the non-speech section detecting unit A non-speech spectrum estimation unit that estimates a frequency spectrum of a non-speech signal that is not, a frequency spectrum subtraction unit that subtracts the frequency spectrum of the non-speech signal estimated by the non-speech spectrum estimation unit from the frequency spectrum obtained by the frequency conversion unit, , Frequency inverse transformation to obtain a time-axis signal by frequency inverse transformation from the frequency spectrum obtained by the frequency spectrum subtraction unit Automatic gain control apparatus according to any one of claims 1 to 12, characterized in that it consists of a part.   A noise component removing unit that removes a noise component other than a voice signal from a previous input signal input to the noise detecting unit, the noise component removing unit; a frequency converting unit that obtains a frequency spectrum of the input signal by frequency conversion; A non-speech section detecting unit for detecting a non-speech section in which an input signal does not include a speech signal from a frequency spectrum of an input signal obtained from a frequency transform unit, and a frequency converting unit for a non-speech section detected by the non-speech section detecting unit A non-speech spectrum estimator that estimates the frequency spectrum of a non-speech signal that is not a speech signal from the frequency spectrum obtained in step 1, and the frequency of the non-speech signal estimated by the non-speech spectrum estimator from the frequency spectrum obtained by the frequency converter The frequency spectrum subtraction unit that subtracts the spectrum and the frequency spectrum obtained by the frequency spectrum subtraction unit Automatic gain control apparatus according to any one of claims 1 to 12, characterized in that it consists of a frequency inversion section for obtaining the signal in the time axis by the number inverse transformation.   15. The automatic gain control apparatus according to claim 14, further comprising: a spectrum normalization unit that obtains a normalized spectrum obtained by power normalizing the frequency spectrum obtained from the frequency conversion unit and supplies the normalized spectrum to the non-speech section detection unit.   15. The automatic gain control apparatus according to claim 13, wherein the non-speech spectrum estimation unit of the noise detection unit is also used as the non-speech spectrum estimation unit of the noise component removing unit.   15. The automatic gain control device according to claim 13, wherein the non-speech section detection unit of the noise detection unit is also used as the non-speech section detection unit of the noise component removing unit.

Images