JP4206409B2 - Audio processing apparatus, method thereof, program, and recording medium recording the program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To attain acceleration of a reaction speed by suppressing or intensifying an input signal in accordance with a change in the character of the relevant signal. <P>SOLUTION: The input signal is converted into a discrete frequency domain signal x(&omega;n) for each frame and divided into a plurality of groups so as to include one or more signals, an average root AX of total power sums of input signals within the frame is determined, an amplitude average AHm of signals within the group is determined for each group, and magnitudes of a first target amplitude TX and AX are compared. If AX is greater, a first standardized average value AHm&times;TX/AX is determined and the magnitude of the standardized average value is compared with that of amplitudes ¾x(&omega;n)¾ of signals in that group. If ¾x(&omega;n)¾ is greater, a first compression function is applied to that amplitude and the signal is output after the amplitude is suppressed to become closer to TX. If ¾x(&omega;n)¾ is not greater, the signal is output while keeping its amplitude as it is, and the entire output signal is converted into a time domain signal. <P>COPYRIGHT: (C)2007,JPO&amp;INPIT

Description

The present invention relates to an audio processing apparatus for compressing a dynamic range of a digitized input signal to a desired range, a method and a program thereof, and a recording medium on which the program is recorded.

In teleconferences and hearing aid systems, the audio signal to be played is used as an input signal so that the audio content to be played can be easily understood even if the playback volume (sound adjustment amount) for the speaker is kept constant. In addition, there is an audio processing method in which the dynamic range of the input signal is compressed, processed so as to fall within a certain volume range, and output by a speaker or the like.
For example, conventionally, in a technique called a limiter or a compressor, when an audio signal having a certain level or more is input based on a predefined nonlinear compression function, a signal whose size is forcibly suppressed is output. These techniques have the problem that the distortion in the audibility of the sound increases as the amount of suppression increases. As shown in “Non-patent Document 1,” it is expected that the above-described distortion problem can be reduced by dividing or converting a signal into a plurality of frequency bands and applying different compression functions for each frequency band. The The technique disclosed in “Non-patent Document 1” has a fixed compression function given in advance for each frequency band, but these compression functions include not the input signal itself converted into each frequency band but each of these discrete functions. The time average value of the amplitude of the frequency domain signal is given as an input. By calculating the ratio between the average time value of the compressed amplitude obtained from this and the time average value of the amplitude before applying the compression function for each frequency band, and multiplying these discrete frequency domain signals by these ratios, The compression process in each frequency band is completed. The signals compressed for each frequency band are subjected to frequency synthesis processing and become final output signals.
T.A. Schneider and R.M. Brennan, “A multichannel compression attribution for a digital healing aid”, 1997 IEEE International Conference on Acoustics, Speech, and Signal Processing 97 (IC). 1, pp. 411-414, April 1997

  Although the technique shown in “Non-patent Document 1” is expected to realize finer and higher quality processing as the frequency band is divided and converted more finely, the amplitude of each discrete frequency domain signal and the original input signal is expected. The compression function given by each frequency band is complex and the compression function that is given fixedly for each frequency band cannot flexibly cope with changes in the signal properties. When a signal with a certain property is input, it can be compressed with high quality. When a signal having a characteristic is input, there is a problem that the signal cannot always be compressed with an equivalent quality. Here, the difference in signal characteristics includes, for example, a difference in signals having a wide range of frequency components such as a highly periodic signal such as a vowel part of speech and a consonant part of one speech.

  In addition, since the time average value of the amplitude of the discrete frequency domain signal is applied as the input of the compression function of each frequency band, there is a problem in that the reaction rate is delayed according to the time constant required for the time average.

  A discrete-time input signal is converted into a discrete frequency domain signal for each frame, and the discrete frequency domain signal is divided into a plurality of groups so that at least one group includes a plurality of discrete frequency domain signals. The square root of the total power of the input signal is obtained, and for each of the divided groups, the average amplitude of the discrete frequency domain signals in the group is obtained, and the desired desired first target amplitude after compression processing and the power When the first determination is made to compare the value of the square root of the sum, and when the signal indicating that the square root of the power sum is larger is input by the first determination, the first target The amplitude average value for each group is normalized by the ratio of the amplitude and the square root of the power sum to obtain a first normalized average value. For each group, the first normalized average value and the group's A second determination comparing the magnitudes of the amplitudes of the discrete frequency domain signals is made. If the amplitude of the discrete frequency domain signals is greater than the second judgment, the output signal is compared to the amplitude. A first compression function that suppresses the first target amplitude so as to approach the first target amplitude is applied by the first compression function calculation unit, and the determination by the second determination unit must be greater in amplitude of the discrete frequency domain signal. For example, the discrete frequency domain signal is output as it is, and the entire discrete frequency domain signal output for each group is converted into a time domain signal.

  According to the above configuration, the response speed can be improved by converting the input signal into a discrete frequency domain signal, and the discrete frequency domain signal is divided into groups, and the characteristics of the signal for each group are divided. Stable or dynamic range compression is possible by suppressing or emphasizing with a compression function corresponding to.

Example 1
FIG. 1 shows a first embodiment of the present invention. When the input signal x (t) is input to the frequency domain conversion unit 2, a fixed time (frame), for example, when the sampling frequency of the input signal x (t) is 16 kHz, the number of samples is short for every 256 or 512. N discrete frequency domain signals corresponding to N frequencies from ω1 to ωN, such as x (ω1),. . . , X (ωN). However, the input signal is a digital signal sampled at a constant period and each sample is converted to a digital value, and N is an integer. N discrete frequency domain signals x (ω1),. . . , X (ωN) are input to the power sum square root calculation unit 4 and the group band division unit 6.

  In the group band dividing unit 6, the discrete frequency domain signals are divided into M groups in order for the frequency numbers 1 to N. However, M is smaller than N and is an integer of 1 or more. As a method of dividing the group, for example, a method of dividing the frequency equally or a method of dividing the low frequency group relatively finely and the high frequency region relatively coarsely can be considered. Further, a group including only one discrete frequency domain signal may exist in one group, but it is assumed that at least one group including two or more discrete frequency domain signals exists. For example, equal division is performed every 1 kHz. Discrete frequency domain signals x (ω1),. . . , X (ωN) is input to the group amplitude average calculation unit 8m, the first compression function control unit 10m, and the first compression function application unit 12m for each divided group. However, m is an integer from 1 to M.

  On the other hand, in the power sum square root calculation unit 4, x (ω1),. . . , X (ωN), and the square root AX of the sum of powers is input to the first compression function controller 10m and the first compression function application unit 12m for each group. The power sum square root calculation unit 4 may input the input signal x (t) as shown by a broken line in FIG. 1 and calculate the square root of the sum of the powers of the target samples in the frame. For each group, the group amplitude average calculation unit 8m calculates the average amplitude AHm of the discrete frequency domain signals in the group. The amplitude average AHm is input to the corresponding compression function control unit 10m and compression function application unit 12m, respectively. An expected desired amplitude after compression processing (hereinafter referred to as a first target amplitude) TX is input to the input signal from the input unit 14, and the first target amplitude TX is the first target amplitude for each group. The data is input to the compression function control unit 10m and the first compression function application unit 12m. The value of the first target amplitude TX is determined by the dynamic range of the speaker or output amplifier of the system to which this apparatus is applied, and a value that does not cause signal distortion is selected.

FIG. 2 shows a detailed example of the first compression function control unit 10m and the first compression function application unit 12m, and a diagram of parts related thereto. The first compression function control unit 10m includes an amplitude absolute value conversion unit 101m, a second determination unit 102m, and a first normalized average value calculation unit 104m.
The first compression function application unit 12m includes a changeover switch 106m, a first compression function calculation unit 130m, a phase applying unit 126m, and a phase calculation unit 128m.
The input unit 14 includes a TX input unit 1400 and an α input unit 1402.
The square root AX of the power sum from the power sum square root calculation unit 4, the first target amplitude TX from the TX input unit 1400, and the amplitude average AHm of the discrete frequency domain signals in the group from the group amplitude average calculation unit 8m, respectively. The first normalized average value is input to the first normalized average value calculation unit 104m, and the first normalized average value calculation unit 104m calculates AHm · TX / AX to calculate the first normalized average value. Is input to the determination unit 102m. The amplitude average value AHm of each group is normalized by the ratio TX / AX of the first target amplitude TX and the power sum square root AX.

  On the other hand, the input discrete frequency domain signal x (ωn) of the group is input to the amplitude absolute value converting unit 101m and the phase calculating unit 128m. The amplitude | x (ωn) | of the discrete frequency domain signal x (ωn) is obtained by the amplitude absolute value converting unit 101m and input to the second determination unit 102m and the changeover switch 106m. The second determination unit 102m compares | x (ωn) | with the magnitude of the values of AHm · TX / AX. If the amplitude | x (ωn) | is larger, the changeover switch 106m is switched to the fixed contact 1062m side by the output of the second determination unit 102m, and the amplitude | x (ωn) | is not larger. , Switching to the fixed contact 1061m side. When switched to the fixed contact 1062m side, the first compression function calculation unit 130m applies the first compression function to the amplitude x (ωn) of the discrete frequency domain signal. The first compression function can be expressed by the following equation, for example.

α | x (ωn) | + (1-α) AHm · TX / AX
Here, α is input from the α input unit 1402 and is a real number ranging from 0 to 1 that determines the degree of suppression. The smaller the value, the greater the suppression. Α is preferably 0.2 to 0.5. Note that AHm · TX / AX calculated by the first normalized average value calculation unit 104m is input to the first compression function calculation unit 130m to be used for calculation of the compression function. The calculation result is input to the phase applying unit 126m. The amplitude | x (ωn) | is directly input to the phase applying unit 126m from the fixed contact 1061m.

That is, the first compression function application unit 12m performs the following control (Equation 1) for the amplitude | x (ωn) | of each discrete frequency domain signal of the group.
When (a) | x (ωn) |> AHm · TX / AX, Fm (| x (ωn) |) = α | x (ωn) | + (1-α) AHm · TX / AX
(B) Otherwise, Fm (| x (ωn) |) = | x (ωn) | (Formula 1)
That is, the first compression function control unit 10m determines whether or not the amplitude of each discrete frequency domain signal for each group is suppressed based on the average amplitude AHm of the group. 1 compression function application unit 12m controls whether the amplitude | x (ωn) | is suppressed or left as it is.

The discrete frequency domain signal x (ωn) of the group m is input to the phase calculation unit 128m. The phase calculation unit 128m calculates the phase ∠x (ωn) of x (ωn), and the calculation result ∠x (ωn) is input to the phase applying unit 126m. In the phase applying unit 126m, Y (ωn) = Fm (| x (ωn) |) · で x (ωn) is calculated, output from the first compression function applying unit 12m, and input to the time domain converting unit 16. The
In addition, in the calculation of (Expression 1), x (ωn) / | x (ωn) | is multiplied in any case in the calculation of (Equation 1) without providing the phase adding unit 126m and the phase calculating unit 128m. The data may be output to the unit 16. The output of the first compression function application unit 12m of each group is converted into a time domain signal by the time domain conversion unit 16 by, for example, a short time inverse Fourier transform, and is output.

  According to the first embodiment, the input signal x (t) is converted into a signal for each frequency component by a short-time Fourier transform or the like, and is grouped into a group of each discrete frequency domain signal. Gives the compression function. At this time, the first compression function has a corresponding discrete frequency only when the instantaneous amplitude | x (ωn) | of the discrete frequency domain signal x (ωn) of each group is larger than the normalized average value AHm · TX / AX. The amplitude | x (ωn) | of the region signal x (ωn) is suppressed. Thereby, when the input signal has an amplitude larger than the first target amplitude, the compression standard of each group is automatically and instantaneously according to the macro frequency characteristics of the input signal captured by the average amplitude of each group. To be determined.

As indicated by a broken line in FIG. 1, the first determination unit 136 compares and determines the magnitude of the power sum square root AX and the first target amplitude TX. When TX> AX is determined, the determination output is obtained. Then, the switch 112 is turned on, and the input signal in the short-time Fourier transform section is output as it is. Further, for example, when the first determination unit 136 determines that TX> AX is not satisfied, the operation of the first compression function control unit 10m of each group may be prohibited.
Example 2
Next, a second embodiment of the present invention will be described.

In the first embodiment, when the square root of the power sum, that is, the estimated average value AX of the amplitude of the input signal is larger than the first target amplitude TX, the amplitude of the input signal is suppressed. However, when TX> AX, the amplitude suppression process is not performed. In the second embodiment, the second target amplitude DX equal to or lower than TX is determined in advance, and when DX> AX, the input signal is emphasized to further compress the dynamic range.
As described above, the first determination unit 136 compares the square root AX of the power sum from the power sum square root calculation unit 4 with the magnitude of the desired amplitude TX from the input unit 14, and the frame satisfies TX <AX. Are suppressed by the first compression function described in the first embodiment. In the second embodiment, for a frame that does not satisfy TX <AX, a desired amplitude (hereinafter referred to as a second target amplitude) DX to be emphasized is determined, and the input signal is enhanced so as to approach the second target amplitude to further increase the dynamic range. Compress.

FIG. 3 shows a specific configuration example of the second embodiment. Compared to the first embodiment described with reference to FIG. 1, the specific configurations of the first compression function control unit 10 m and the first compression function application unit 12 m are partially changed, and the second compression function control unit 135 m The second compression function application unit 125m is provided. The same reference numerals are assigned to the same functional components.
When the determination of the first compression determination unit 134 is TX> AX, a determination result output indicating this is input to the second compression function control unit 135m.
FIG. 4 shows a detailed example of the second compression function control unit 135m and the second compression function application unit 125m, and a diagram of parts related thereto. The same reference numerals are assigned to the same functional components.

The first compression determination unit 134 includes a first determination unit 136 and a third determination unit 137, and the second compression function application unit 125m includes a changeover switch 108m, a changeover switch 107m, and a second compression function calculation unit 132m. The phase adding unit 126m and the phase calculating unit 128m. The input unit 14 includes a TX input unit 1400, a DX input unit 1404, an α input unit 1402, and a β input unit 1406. The second compression function control unit 135m It is comprised by the 4th determination part 138m.
First, the first determination unit 136 compares the square root AX of the power sum from the power sum square root calculation unit 4 with the desired amplitude TX from the TX input unit 1400 and the magnitude of the value, and the result is (1) AX ≦ DX ( 2) DX <AX ≦ TX (3) Input to the three-state determination circuit 144 that determines three states of TX <AX.

  The second target amplitude DX input from the DX input unit 1404 is input to the third determination unit 137 and the second compression function calculation unit 132m. The third determination unit 137 compares the values of the square root AX of the power sum from the power sum square root calculation unit 4 and the second target amplitude DX, and the result is input to the three-state determination circuit 144. However, DX ≦ TX, and DX is preferably about one-half of TX. When the output of the three-state determination circuit 144 is (1), the selector switches 108m of all groups are switched to the fixed contact 1082m side, and in the case of (3), the switches are switched to the fixed contact 1081m. In the case of (2), the input signal is output without being suppressed. When switched to the fixed contact 1081m side, the amplitude | x (ωn) | of each discrete frequency domain signal of the corresponding group m from the amplitude absolute value converting unit 101m is changed to the changeover switch 106m in the first compression function applying unit 12m. To the fixed contact 1061m. That is, in this case, the amplitude | x (ωn) | is not subjected to suppression compression or enhancement compression.

  In the fourth determination unit 138m, the magnitude of the amplitude average AHm from the group amplitude average calculation unit 8m and the amplitude | x (ωn) | from the amplitude absolute value conversion unit 101m are compared, and the amplitude | x (ωn) | If it is larger, the changeover switch 107m is switched to the fixed contact 1072m side, and if the amplitude | x (ωn) | is not larger, it is switched to the fixed contact 1071m side. When the changeover switch is switched to 1072 m, that is, when DX> AX and | x (ωn) |> AHm are satisfied, the amplitude | x (ωn) | of the discrete frequency domain signal is calculated by the second compression function calculation. In part 132m, the second compression function is applied. For example, the second compression function can be expressed by the following equation.

β | x (ωn) | + (1-β) AHm · DX / AX
Here, β is input from the β input unit 1406, β is a real number in the range of 0 to 1 that determines the degree of emphasis, and the smaller the value, the larger the input amplitude | x (ωn) | Greater emphasis. Note that β is preferably 0.2 to 0.5. The calculation result is input to the phase applying unit 126m. When the fixed contact 1071m is input to the phase applying unit 126m and the changeover switch 107m is connected to the fixed contact 1071m side, the amplitude | x (ωn) | is directly input to the phase applying unit 126m.

That is, the second compression function application unit 125m controls the amplitude | x (ωn) | of each discrete frequency domain signal of the group by the following (Equation 2).
When (a) | x (ωn) |> AHm, Fm (| x (ωn) |) = β | x (ωn) | + (1-β) AHm · DX / AX
(B) Otherwise, Fm (| x (ωn) |) = | x (ωn) | (Formula 2)
That is, the second compression function control unit 135m determines whether or not the amplitude of each discrete frequency domain signal for each group is enhanced based on the average amplitude AHm of the group. The compression function applying unit 125m 2 controls whether the amplitude | x (ωn) | is emphasized or left as it is.

Subsequent compression in the second compression function application unit 125m is performed by adding the phase ∠x (ωn) to the signal subjected to enhancement processing as it is in the same manner as in the first embodiment and is input to the time domain conversion unit 16. The
FIG. 5A shows the characteristics of the suppression process according to (Expression 1) for a frame satisfying AX> TX, and FIG. 5B shows the characteristics of the enhancement process according to (Expression 2) for a frame not satisfying AX> TX. In both FIGS. 5A and 5B, the vertical axis is Fm (| x (ωn) |) that is output, and the horizontal axis is | x (ωn) | that is input.

In FIG. 5A, according to the above (Equation 1), in the region of | x (ωn) |> AHm · TX / AX (shown as the suppression region in FIG. 5A),
Fm (| x (ωn) |) = α | x (ωn) | + (1-α) AHm · TX / AX is suppressed, and | x (ωn) | ≦ AHm · TX / AX,
Since Fm (| x (ωn) |) = | x (ωn) |, no suppression is performed.
In FIG. 5B, in the region of | x (ωn) |> AHm (shown as an emphasized region in FIG. 5B) according to the above (Equation 2),
Fm (| x (ωn) |) = β | x (ωn) | + (1-β) AHm · DX / AX
And | x (ωn) | ≦ | AHm,
Since Fm (| x (ωn) |) = | x (ωn) |, it is not emphasized at all.
In FIG. 5A, in the suppression processing according to (Equation 1), the normalized average value AHm · TX / AX normalized by TX / AX is set as the lower limit value of suppression for | x (ωn) | In the emphasis by (Expression 2), the lower limit value of emphasis is given by AHm. When suppressing a signal, it is necessary to suppress a signal smaller than AHm depending on the required degree of suppression. On the other hand, when emphasizing, it is included in a signal smaller than the average value AHm. The intention is to avoid unnecessary amplification such as possible noise components. In the characteristic of (Equation 2) shown in FIG.
In the range of | x (ωn) |> AHm · TX / AX, the effect of suppressing the reverse is provided. However, when the first target amplitude TX is selected to be close to the second target amplitude DX, the reverse is achieved. Therefore, continuity with the characteristic of (Equation 1) is maintained.

  Moreover, the noise level estimation part 142m may be provided in the 2nd compression function control part 135m in FIG. In this case, for example, as indicated by a broken line in the second compression function control unit 135m in FIG. 4, a noise level estimation unit 142m is provided, and further, a fifth determination unit 140m is also provided. The frequency domain signal x (ωn) is input to the noise level estimation unit 142m for each group, and the noise level NLm obtained by multiplying the maximum value or the average value of the noise components that do not require enhancement by a constant larger than 1 is estimated. The noise level NLm is input to the fifth determination unit 140m. The fifth determination unit 140m compares the magnitudes of the amplitude average AHm and the noise level NLm from the group amplitude average calculation unit 8m for each corresponding group. Based on the comparison result from the fifth determination unit 140m, the larger value of the amplitude average AHm and the noise level NLm is input to the fourth determination unit 138m, and the input larger value is input to the fourth determination unit 138m. And the amplitude | x (ωn) |. In this way, unwanted enhancement of noise components can be more reliably suppressed.

Example 3
Next, a third embodiment of the present invention will be described. In the first and second embodiments, the amplitude value of the input signal is calculated by the square root of the sum of power, and thus a signal with low energy cannot be effectively suppressed even if the instantaneous amplitude is large, such as an impulsive signal. In the third embodiment, attention is paid to the property that the pulse signal converted into the frequency domain has substantially the same amplitude at each frequency, and is configured as shown in FIG. 6, for example. The overall block configuration is substantially the same as in the first embodiment, but the second compression determination unit 147 is provided, and the processing contents in the third compression function control unit 145m and the third compression function application unit 153m are different. . For each frame, it is determined whether or not the input signal is an impulsive signal, and the following processing is performed in the third embodiment for a frame that is an impulsive signal.

FIG. 7 shows details of the third compression function control unit 145m, the third compression function application unit 153m, and other related parts. The third compression function control unit 145m includes an amplitude absolute value conversion unit 101m, a second normalized average value calculation unit 148m, and a sixth determination unit 152m. The third compression function application unit 153m includes a changeover switch 110m, A third compression function calculation unit 154m, a phase applying unit 126m, and a phase calculation unit 128m are included. 6 and 7, the same reference numerals are assigned to the same functional components as those in the first and second embodiments.
The discrete frequency domain signal x (ωn) is input to the second compression determination unit 147 (impulsive signal determination unit), the phase calculation unit 128m, and the amplitude absolute value conversion unit 101m. The second compression determination unit 147 determines whether or not this frame is an impulsive signal, and the determination result is input to the sixth determination unit 152m. Further, the average amplitude AHm from the group amplitude average calculation unit 8m is input to the second normalized average value calculation unit 148m, and the square root AX of the power sum from the power sum square root calculation unit 4 and the impulsiveness from the PX input unit 1408 An expected amplitude (hereinafter, third target amplitude) PX after compression processing on the signal is input to the second normalized average value calculation unit 148m and the second compression determination unit 147, respectively. The third target amplitude PX is desirably about 1/10 of the first target amplitude TX.

The second normalized average value calculation unit 148m calculates the second normalized average value AHm · PX / AX and inputs it to the sixth determination unit 152m. In the sixth determination unit 152m, when the determination result from the second compression determination unit (impulsive signal determination unit) 147 is an impulse signal, the amplitude | x (ωn) | and the second normalized average value AHm · PX / AX values are compared in magnitude.
The amplitude | x (ωn) | is obtained by the amplitude absolute value converting unit 101m and input to the changeover switch 110m and the sixth determining unit 152m.
When the determination result of the sixth determination unit 152m is | x (ωn) |> AHm · PX / AX, the changeover switch 110m is switched to the fixed contact 1102m, and | x (ωn) |> AHm · PX / AX is not satisfied Switches the changeover switch 110m to the fixed contact 1101m.

When the changeover switch 110m is switched to the fixed contact 1102m, γ | x (ωn) | + (1-γ) AHm · PX / AX is the third compression function calculation unit with respect to the amplitude | x (ωn) | It is calculated at 154m. Note that γ is input by the γ input unit 1410, and preferably γ = 0.2 to 0.5. On the other hand, when the changeover switch 110m is switched to the fixed contact 1101m, the amplitude | x (ωn) | is left as it is.
That is, the third compression function application unit 153m controls the amplitude | x (ωn) | of each discrete frequency domain signal of the group by the following (Equation 3).
When (a) | x (ωn) |> AHm · PX / AX, Fm (| x (ωn) |) = γ | x (ωn) | + (1-γ) AHm · PX / AX
(B) Otherwise, Fm (| x (ωn) |) = | x (ωn) | (Formula 3)
That is, the third compression function controller determines whether or not the amplitude of each discrete frequency domain signal for each group is suppressed based on the second normalized average value AHm · PX / AX of the group. Based on the determination result, the third compression function applying unit 153m controls whether the amplitude | x (ωn) | is suppressed or left as it is.

A phase ∠x (ωn) is given to the processed amplitude Fm (| x (ωn) |) in the same manner as in the first and second embodiments, and is input to the time domain conversion unit 16. The phase ∠x (ωn) may be given by multiplying x (ωn) / | x (ωn) | by Fm (| x (ωn) |). The same applies to the second embodiment.
FIG. 8 shows a specific configuration example of the second compression determination unit 147. The second compression determination unit 147 includes, for example, an overall amplitude average calculation unit 1440, an overall maximum amplitude detection unit 1442, a seventh determination unit 1445, an eighth determination unit 1444, and an AND circuit 1446.

The square root AX of the power sum from the power sum square root calculation unit 4 and the third target amplitude PX from the PX input unit 1408 are input to the seventh determination unit 1445, and the square root AX of the power sum and the third target amplitude PX The magnitude of the value is determined. The determination result is input to the AND circuit 1446.
Also, the discrete frequency domain signal x (ωn) of the entire band is input to the overall amplitude average calculating unit 1440 and the overall maximum amplitude detecting unit 1442, and the overall amplitude average calculating unit 1440 performs the discrete frequency domain signal x (ωn) of the entire band. The average amplitude x (ωn) _A is calculated, and the maximum maximum amplitude detection unit 1442 calculates the maximum amplitude x (ωn) _M of the discrete frequency domain signal x (ωn) of the entire band. The amplitude average x (ωn) _A and the maximum amplitude x (ωn) _M are respectively input to the eighth determination unit 1444, and the eighth determination unit 1444 uses the amplitude average x (ωn) _A and the maximum amplitude x (ωn) _M. Is within a predetermined range, for example, when the amplitude average x (ωn) _A is larger than the constant amplitude ε times the maximum amplitude x (ωn) _M , and the seventh determination unit 1445 determines that PX <AX If determined, the second compression determination unit 147 determines that the signal is an impulsive signal. The constant ε is a real number from 0 to 1, and is preferably about 0.9, for example. The amplitudes of all samples in the frame may be averaged to obtain an amplitude average x (ωn) _A.

Although the amplitude of the time waveform is excessive in time as in the case of the impulsive signal according to the third embodiment, it is dispersed into each frequency after being converted into the frequency domain, so only the implementation of the first embodiment described above. Then, even when sufficient suppression is difficult, an effect of enabling suppression can be obtained. By applying the third embodiment first, it is possible to perform the suppression in the case where it is difficult to sufficiently suppress the first embodiment alone.
For easy understanding of Example 3, FIGS. 6 and 7 are described independently. However, in practice, Example 3 is preferably used in combination with Example 1 and Example 2. The processing flow of the third embodiment in this case will be described with reference to FIG. First, the second compression determination unit 147 determines whether or not the signal is an impulsive signal (S20). If the signal is an impulsive signal, the third compression function applying unit 153m as described above for the frame, Processing is performed using (Expression 3) (S22), and output to the time domain conversion unit 16. On the other hand, for the frame determined not to be an impulsive signal in step S20, the first determination unit 136 determines the magnitude of the square root AX of the power sum and the first target amplitude TX (S24), and if AX> TX. The first compression function application unit 12m performs processing according to (Expression 1) (S26) and outputs the result to the time domain conversion unit 16. On the other hand, for frames in which AX> TX does not hold in step S24, the third determination unit 137 determines the magnitude of the second target amplitude DX and the square root AX of the power sum (S28), and if DX> AX. The frame is processed by (Expression 2) by the second compression function application unit 125m (S30) and output to the time domain conversion unit 16. For frames that are not DX> AX in S28, the discrete frequency domain signals of all groups are output as they are to the time domain transform unit 16 (S32).

  Moreover, the following order is also considered about the frame determined not to be an impulsive signal in step S20. That is, as shown in the broken line block B in FIG. 9, the first compression determination unit 134 determines the magnitude of the second target amplitude DX and the square root AX of the power sum (S34), and satisfies DX> AX. The frame is processed by the second compression function application unit 125m (S36) and output to the time domain conversion unit 16. For frames that do not satisfy DX> AX in step S34, the first determination unit 136 determines the magnitude of the first target amplitude TX and the square root AX of the power sum (S38), and for frames that satisfy AX> TX. Processing is performed by the first compression function application unit 12m (S40), and the result is output to the time domain conversion unit 16. Frames that do not satisfy AX> TX in step S38 are output to the time domain conversion unit 16 as they are (S42).

Moreover, when Example 3 is implemented, the process sequence shown in FIG. 10 can also be considered. The second target amplitude DX and the third target amplitude PX are compared in advance, and when PX ≧ DX is satisfied, the first compression determination unit 134 determines whether the second target amplitude DX and the square root AX of the power sum are Each frame is determined (S46), and a frame satisfying DX> AX is processed by the second compression function application unit 125m (S58) and output to the time domain conversion unit 16.
On the other hand, for a frame that does not satisfy DX> AX in step S46, the second compression determination unit 147 determines whether or not it is an impulsive signal (S48). If it is an impulsive signal, the frame is processed by the third compression function application unit 153m (S54) and input to the time domain conversion unit 16. If it is determined in step S48 that the signal is not an impulsive signal, the first determination unit 136 determines the magnitude of the first target amplitude TX and the square root AX of the power sum, and if it is determined that AX> TX, On the other hand, the first compression function application unit 12m performs processing (S56) and outputs the result to the time domain conversion unit 16. Frames determined not to have AX> TX in step S50 are output to the time domain conversion unit 16 as they are (S52).

1 is a block diagram showing a configuration example of Embodiment 1 of the present invention. FIG. 3 is a detailed example of a first compression function control unit 10m and a first compression function application unit 12m according to the first embodiment of the present invention, and a block diagram of parts related thereto. The block diagram which shows the structural example of Example 2 of this invention. The detailed example of the 2nd compression function application part 125m and the 2nd compression function control part 135m of Example 2 of this invention, and the block diagram of the part relevant to this. (A) is the characteristic by the compression function of (Formula 1) by the 1st compression function application part 12m, (B) is the characteristic by the compression function of (Formula 2) by the 2nd compression function application part 125m. The block diagram which shows the structural example of Example 3 of this invention. The detailed example of the 3rd compression function application part 153m of Example 3 of this invention and the 3rd compression function control part 145m, and the block diagram of the part relevant to this. 9 shows a specific configuration example of a second compression determination unit 147 according to the third embodiment. The flowchart figure at the time of using combining Examples 1-3. The other flowchart figure at the time of using combining Examples 1-3.

Claims (12)

A frequency domain transform unit for transforming a discrete time input signal into a discrete frequency domain signal for each frame;
A group divider that divides the discrete frequency domain signal into a plurality of groups such that at least one group includes a plurality of discrete frequency domain signals;
A power sum square root calculation unit for obtaining the square root of the power sum of the input signals in the frame;
For each of the divided groups, a group amplitude average calculation unit that calculates an amplitude average of discrete frequency domain signals in the group;
A first determination unit that compares magnitudes of values of an expected desired first target amplitude after compression processing and the square root of the power sum;
A signal indicating that the square root of the power sum is larger than that of the first determination unit, and an amplitude average value for each group is calculated by a ratio between the first target amplitude and the square root of the power sum. Normalize and obtain the first normalized average value by the first normalized average value calculation unit,
For each group, a first compression function control unit that compares the first normalized average value and the amplitude of each discrete frequency domain signal of the group by a second determination unit;
If the second determination unit determines that the amplitude of the discrete frequency domain signal is larger, the first compression function calculation unit applies a predetermined first compression function to the amplitude. If the second determination unit determines that the amplitude of the discrete frequency domain signal is not larger, the discrete frequency domain signal is suppressed so as to approach the first target amplitude. A first compression function application unit for each group,
A time domain transforming unit that transforms the entire discrete frequency domain signal output from the first compression function applying unit for each group into a time domain signal;
A speech processing apparatus comprising: The speech processing apparatus according to claim 1, wherein
A signal indicating that the square root of the total power from the first determination unit is not larger is input,
A first compression determination unit that compares the input desired amplitude to be emphasized (hereinafter referred to as a second target amplitude) and the power sum square root with a third determination unit;
A second compression function control unit that compares the amplitude average value of each group and the amplitude of each frequency signal of the corresponding group by a fourth determination unit;
When the determination by the first determination unit is greater in the first target amplitude, the determination by the third determination unit is greater in the second target amplitude, and the fourth determination unit If the amplitude of the frequency signal is greater than
Applying a predetermined second compression function to the amplitude in the second compression function calculation unit, emphasizing it so as to approach the second target amplitude, and outputting it to the time domain conversion unit,
If the determination of the fourth determination unit determines that the amplitude of each frequency signal is not greater, the second compression function application unit outputs the amplitude as it is to the time domain conversion unit. Voice processing device. The speech processing apparatus according to claim 2, wherein
The second compression function control unit is configured to estimate a noise component estimation unit that estimates a value obtained by multiplying a maximum value or an average value of noise levels by a constant larger than 1 for each group;
A fifth determination unit that compares the noise level and the group amplitude average value;
If the fifth determination unit determines that the noise level is higher, the speech processing apparatus uses the noise level instead of the group amplitude average value. The speech processing apparatus according to any one of claims 1 to 3,
A second compression determination unit that determines for each frame whether or not the input signal is an impulsive signal;
When the second compression determination unit determines that the discrete frequency domain signal is an impulsive signal, the operation of the first compression function control unit is stopped, and for each group of the frame, The ratio of the desired amplitude after compression processing (referred to as the third target amplitude) and the square root of the power sum is normalized, and the second normalized average value calculation unit calculates the second amplitude average value. Find the normalized average value,
For each group, a third compression function control unit that compares the second normalized average value and the amplitude of each discrete frequency domain signal of the group with a sixth determination unit;
If the determination by the sixth determination unit is greater in the amplitude, a third compression function is applied to the discrete frequency domain signal by a third function calculation unit, and the third desired value is obtained. To the time domain conversion unit, and output to the time domain conversion unit. If the determination by the sixth determination unit is not larger than the amplitude, the discrete frequency domain signal is output to the time domain conversion unit as it is. A third compression function application unit that
A speech processing apparatus comprising: The speech processing apparatus according to claim 4, wherein
The second compression determination unit
A seventh determination unit that compares the third target amplitude and the square root of the power sum;
An overall amplitude average calculator for calculating an average amplitude of the input signal of the frame;
An overall maximum amplitude detector for detecting a maximum amplitude for all frequencies of the discrete frequency domain signal;
An eighth determination unit that determines whether the difference between the overall average amplitude value and the maximum amplitude is within a predetermined range;
If the square root of the power sum is larger in the seventh determination unit and falls within the range in the eighth determination unit, it is determined that the input signal of the frame is an impulsive signal. A voice processing device. A frequency conversion procedure for converting a discrete time input signal into a discrete frequency domain signal for each frame;
A group division procedure for dividing the discrete frequency domain signal into a plurality of groups so that at least one group includes a plurality of discrete frequency domain signals;
A power sum square root calculation procedure for obtaining the square root of the power sum of the input signals in the frame;
For each of the divided groups, a group amplitude average calculation procedure for calculating the average amplitude of the discrete frequency domain signals in the group;
A first determination procedure for comparing the magnitude of a value of an expected desired first target amplitude after the compression processing and the average of the power sum;
In the determination of the first determination procedure, the square root of the power sum is larger, and the average amplitude value for each group is normalized by the ratio of the desired first target amplitude and the square root of the power sum. Find the first normalized average value,
For each group, a second determination procedure that compares the first normalized average value with the amplitude of each discrete frequency domain signal of the group;
If the determination result of the second determination procedure indicates that the amplitude of the discrete frequency domain signal is larger, a predetermined first compression function is applied to the amplitude to determine the first target. A first compression function application procedure that suppresses the amplitude so as to approach the amplitude and leaves the discrete frequency domain signal as it is if the determination of the second determination procedure is not greater than the amplitude of the discrete frequency domain signal. When,
A time domain conversion procedure for converting the entire discrete frequency domain signal processed by the first compression function application procedure for each group into a time domain signal;
A voice processing method characterized by comprising: The voice processing method according to claim 6.
A third determination procedure for comparing the magnitude of the input desired amplitude to be emphasized (hereinafter referred to as a second target amplitude) and the total square root of the power;
A fourth determination procedure for comparing the amplitude average value of each group with the amplitude of each frequency signal of the corresponding group;
When the determination result of the first determination procedure is larger than the first target amplitude, the determination result of the third determination procedure is that the second target amplitude is large, and the fourth determination is performed. If the result of the procedure is greater than the amplitude of the frequency signal,
Applying a predetermined second compression function to the amplitude to emphasize the amplitude so as to approach the second target amplitude,
If the determination result of the fourth determination procedure determines that the amplitude of each frequency signal in the corresponding group is not greater, a second compression function application procedure that leaves the amplitude as it is;
And a procedure of converting the entire discrete frequency domain signal processed by the second compression function application procedure for each group into a time domain signal. The voice processing method according to claim 7,
A noise component estimation procedure for estimating a value obtained by multiplying a maximum value or an average value of noise levels by a constant larger than 1 for each group;
A fifth determination procedure for comparing the noise level with the group amplitude average value,
If the determination result of the fifth determination procedure determines that the noise level is larger, the fourth determination procedure uses the noise level instead of the group amplitude average value. . The voice processing method according to any one of claims 6 to 8,
A second compression determination procedure for determining whether or not the input signal is an impulsive signal for each frame;
When the discrete frequency domain signal is determined to be an impulsive signal by the second compression determination procedure, the compression processing for the impulsive signal is performed for each group of the frame without performing the first compression function application procedure. The ratio of the later desired amplitude (referred to as the third target amplitude) and the square root of the power sum is normalized to obtain the second normalized average value.
For each group, a sixth determination procedure for comparing the magnitude of the second normalized average value and the amplitude of each discrete frequency domain signal of the group;
If the determination result of the sixth determination procedure is larger than the amplitude, a third compression function is applied to the discrete frequency domain signal by the third function calculation unit, and the third desired vibration is applied. A third compression function application procedure that suppresses the value so as to approach the value and leaves the discrete frequency domain signal as it is if the determination result of the sixth determination procedure is not greater than the amplitude;
A procedure for converting the entire discrete frequency domain signal processed by the third compression function application procedure for each group into a time domain signal;
A voice processing method characterized by comprising: The voice processing method according to claim 9.
The second compression determination procedure is as follows.
A seventh determination procedure for comparing the magnitudes of the third target amplitude and the square root of the power sum;
An overall amplitude average calculation procedure for calculating an average amplitude value of the input signal of the frame;
An overall maximum amplitude detection procedure for detecting a maximum amplitude for all frequencies of the discrete frequency domain signal;
An eighth determination procedure for determining whether or not the difference between the overall amplitude average value and the maximum amplitude is within a predetermined range;
The input signal of the frame in which the determination result of the seventh determination procedure is larger in the square root of the total power and the determination result of the eighth determination procedure is within a predetermined range is an impulse signal A speech processing method characterized by determining that   A voice processing program for causing a computer to execute each procedure of the voice processing method according to claim 6. The computer-readable recording medium which recorded the audio | voice processing program of Claim 11.

Images