JP4942755B2 - Audio signal processing apparatus and method - Google Patents

Description

  The present invention relates to an audio signal processing apparatus and method. In particular, the present invention relates to a technique for improving distortion caused by insufficient bits when quantizing a digital audio signal. More specifically, the present invention relates to a technique for generating data in which the number of quantization bits of a digital audio signal is expanded, and more particularly to suppression of occurrence of digital audio signal distortion.

In recent years, digitalization of audio signal processing in audio equipment (audio equipment), in-vehicle audio equipment (car audio equipment), and DVD recorders has become common. For example, 16-bit quantization for a CD or the like and a sampling frequency of 44.1 kHz (a digital audio tape standard is a sampling frequency of 48 kHz) are defined as standards.
On the other hand, a natural sound that is generated in an analog manner is generally used as the source of the audio signal, and the dynamic range of the input level is very wide. Therefore, waveform distortion occurs in an input signal that changes at a minute level near the minimum bit due to an insufficient number of quantization bits.

In the following Patent Document 1, in order to cope with such signal distortion due to the insufficient number of quantization bits, (a) a sequential sample value obtained by quantizing an audio signal in the form of an analog signal is a 1 / Nth power of 2. Obtaining code information of an N-bit digital audio signal converted into a digital signal with a resolution of (b), and (b) an additional code of (MN) bits in the least significant digit (LSB) of the N-bit code information It is disclosed that the number of bits is converted into M-bit code information having a relationship of M> N by making information continuous.

In the following Patent Document 2, in order to cope with such signal distortion due to the insufficient number of bits of quantization, (a) among the data changes for each sample data of the input digital data, Using at least one of the low-pass filters having different cut-off frequencies as a criterion for the change time length to obtain a signal output, and (b) the level change rate from the time interval of the LSB change It is described that after generating a data string lower than the LSB that discriminates and smoothes the waveform change in the discriminated period, data below the LSB is generated before and after the LSB change point to extend the bit.

Japanese Patent No. 3336823 Japanese Patent Publication No. 7-73186 Supervised by Takao Hinamoto, Mitsuji Muneyasu, Ryo Otaguchi, “Nonlinear Digital Signal Processing”, Asakura Shoten, March 20, 1999, p. 72-74, p. 106-108

  The above Non-Patent Document 1 will be mentioned later.

  For example, the method of the apparatus of Patent Document 1 performs a process of making additional code information continuous at the least significant digit. In this method, the additional code information is added only to the waveform portion where only the least significant digit changes. That is, if the original data is n bits and the bit width is M bits, the code information added for correction has only (M−n) bits. Therefore, the waveform distortion improvement effect is poor. That is, only the change between sampling points where only the least significant digit of the input signal changes is interpolated with a high quantization bit number.

  In the method according to Patent Document 2, the cutoff frequency is selected only by the time length of the LSB change. That is, the cut-off frequency changes every time there is an LSB change. In addition, it is difficult to accurately select the cut-off frequency only by the time length of the LSB change. For example, even with a sine wave having a constant frequency, the time length of the LSB change is not constant. For this reason, the process stitching does not change smoothly over a long time range, and the distortion is not sufficiently corrected.

  The present invention generates interpolated data for waveform distortion correction that is finer and has a higher degree of freedom than in the past while realizing that the reproducibility of an audio signal input waveform having a region where the signal amplitude changes sharply and greatly is not impaired, It is an object of the present invention to provide an audio signal processing device that approximates an audio signal input waveform.

The audio signal processing device of the present invention is
an original data bit extension unit for generating and outputting a sequence of n + α-bit audio signal data by extending each bit of the n-bit (n is an integer) audio signal data by α bits (α is an integer) bit;
An edge-preserving smoothing filter that smoothes a sequence of n + α-bit audio signal data generated by the bit extension ;
A frequency amplitude estimator for estimating the frequency and amplitude of the n-bit audio signal;
A filter coefficient generation unit that generates a low-pass filter coefficient based on the frequency and the amplitude,
The edge preserving smoothing filter unit performs smoothing using the filter coefficient generated by the filter coefficient generation unit,
The filter coefficient generation unit
A filter coefficient of a low-pass filter that removes harmonic components of the frequency of the estimated speech signal is generated, and the order of the low-pass filter decreases as the estimated amplitude increases. To do.

According to the present invention, the data width of information added for waveform distortion correction is configured to generate distortion correction data with a data width after bit number conversion (expansion). Therefore, it differs from the prior art of Patent Document 1 in that the data width to be corrected is wide. Further, an edge-preserving smoothing filter (for example, a one-dimensional mth-order ε filter) is applied as distortion correction data generation means. With these configurations, it is possible to generate interpolation data for distortion correction that is finer and has a higher degree of freedom than the conventional method, and to approximate the original sound waveform.
Further, by applying an edge preserving smoothing filter (for example, a one-dimensional m-order ε filter), the reproducibility of an audio signal having a region where the signal amplitude changes sharply and greatly is not impaired. This is because the threshold value for the filter operation can be set.

It is a block diagram which shows the structure of the audio | voice signal output device which concerns on Embodiment 1 of this invention. It is a figure which shows the function f (u) of Formula (1). (A), (b) is a figure for demonstrating operation | movement of an epsilon filter. (A), (b) is a figure for demonstrating operation | movement of an epsilon filter. 5 is a block diagram showing an audio signal output device according to another configuration example of Embodiment 1. FIG. 5 is a block diagram showing an audio signal output device according to still another configuration example of Embodiment 1. FIG. 3 is a block diagram illustrating a more specific configuration example of the first embodiment. FIG. (A) And (b) is a figure for demonstrating operation | movement of the input process part 2 used in Embodiment 1. FIG. (A) And (b) is a figure for demonstrating operation | movement of the original data bit expansion part 5 used in Embodiment 1. FIG. 4 is a diagram illustrating a configuration example of a data storage unit 7 used in Embodiment 1. FIG. 6 is a diagram illustrating an output of a data storage unit 7 used in the first embodiment. FIG. First difference calculation unit 8-1 to fourth difference calculation unit 8-4 used in the first embodiment, first ε determination unit 9-1 to fourth ε determination unit 9-4, and determination 3 is a table summarizing data generated by the weighted average unit 10. (A) And (b) demonstrates operation | movement of the 1st (epsilon) determination part 9-1-the 4th (epsilon) determination part 9-4 which are used in Embodiment 1, and operation | movement of the weighted average part 10 with determination. It is a figure for doing. It is a figure which shows the weighted average value EM corresponding to the audio | voice signal data DM. (A)-(c) is a figure for demonstrating operation | movement of the data addition part 11 used in Embodiment 1. FIG. (A)-(e) is a figure for demonstrating operation | movement when the signal which the signal amplitude is steep and is large (it is larger than a threshold value) is input into the audio | voice signal output device which concerns on Embodiment 1. FIG. . First difference calculation unit 8-1 to fourth difference calculation unit 8-4 used in the first embodiment, first ε determination unit 9-1 to fourth ε determination unit 9-4, and determination 4 is a table showing data generated by the weighted average unit 10. It is a block diagram which shows the audio | voice signal output device of Embodiment 1 of this invention. 4 is a flowchart illustrating processing steps of the audio signal output device according to the first embodiment. It is a block diagram which shows the audio | voice signal output device which concerns on Embodiment 1 of this invention. It is a block diagram which shows another structural example of the audio | voice signal output device of Embodiment 2 of this invention. It is a block diagram which shows the structure of the audio | voice signal processing apparatus which concerns on Embodiment 3 of this invention. (A) And (b) is a figure which shows an example of the n-bit audio | voice signal of Embodiment 3. FIG. (A) And (b) is a figure for demonstrating the operation | movement of an epsilon filter. (A) And (b) is a figure which shows the operation | movement of the original data bit expansion part 24. FIG. (A) And (b) is a figure which shows operation | movement of the edge preservation | save type | mold smoothing filter part 25. FIG. 3 is a block diagram illustrating an example of a frequency amplitude estimation unit 22. FIG. (A)-(d) is a figure which shows operation | movement of the frequency amplitude estimation part 22 of FIG. (A)-(d) is a figure which shows operation | movement of the frequency amplitude estimation part 22 of FIG. (A)-(c) is a figure for demonstrating operation | movement of the frequency amplitude estimation part 22 when the audio | voice signal on which the different frequency component was superimposed is input. (A)-(d) is a figure for demonstrating operation | movement of the frequency amplitude estimation part 22 when the audio | voice signal on which the different frequency component was superimposed is input. (A)-(d) is a figure which shows operation | movement of the example from which the frequency amplitude estimation part 22 differs. 6 is a diagram illustrating an example of a coefficient table stored in a filter coefficient generation unit 23. FIG. (A) And (b) is a figure for demonstrating the filter coefficient table which stores the some low-pass filter coefficient from which a cut-off frequency and an order differ. 10 is a flowchart illustrating processing steps of the audio signal processing device according to the third embodiment.

Explanation of symbols

  1 input terminal, 2 input processing unit, 3 extended data generation processing unit, 4 output processing unit, 5 original data bit expansion unit, 6 one-dimensional m-order ε filter unit, 7 data storage unit, 8-1 to 8-4 difference Calculation unit, 9-1 to 9-4 ε determination unit, 10 weighted average unit with determination, 11 data addition unit, 12 weighted average unit with coefficient programmable determination, 13 weighted average unit with variable degree determination, 15 output processing unit, 15a Programmable amplifier, 16 comparison unit, 17-bit extension and weighted average unit with determination, 18-bit extension data addition unit, 18a-bit extension unit, 18b data addition unit, one-dimensional m-order ε filter unit with 19-bit extension, 21 input terminal , 22 Frequency amplitude estimation unit, 23 Filter coefficient generation unit, 24 Original data bit extension unit, 25 Edge Presence smoothing filter section, 27 the inflection point detecting unit, 28 a frequency estimation unit, 29 an amplitude estimating unit, 31 a primary differential calculating section, 32 sign change point detecting unit.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing a configuration of an audio signal output apparatus according to Embodiment 1 of the present invention. The audio signal output device according to Embodiment 1 includes an input terminal 1, an input processing unit 2, an extended data generation processing unit 3, and an output processing unit 4. Among these, the input signal processing unit 2 and the extended data generation processing unit 3 constitute an audio signal processing device.

  In the present embodiment, the input processing unit 2 is described as an A / D converter that converts analog audio signals into digital audio signal data. However, the input processing unit 2 includes an analog processing circuit, and the input processing unit 2 May be converted into digital audio signal data after analog processing such as equalizing, or digital data is received from the input terminal 1 using the input processing unit 2 as a digital interface, and the n-bit audio signal data DI is It may be output.

  An analog audio signal SA is input from the input terminal 1 to the input processing unit 2, and the input processing unit 2 samples the analog audio signal SA at a predetermined sampling period and converts it into a sequence of n-bit audio signal data DI. And output to the extended data generation processing unit 3. The extension data generation processing unit 3 includes an original data bit extension unit 5 and a one-dimensional mth-order ε filter unit 6, converts each data in a sequence of n-bit audio signal data DI into n + α bits, and outputs an output processing unit 4. Output to. The original data bit extension unit 5 outputs to the one-dimensional m-order ε filter unit 6 a sequence of n + α-bit audio signal data DS obtained by bit-extending (also referred to as “bit shift”) n-bit audio signal data DI by α bits. To do.

  The audio signal data DS (although the other audio signal data DI and the like are the same) is composed of a sequence of data sequentially representing data values (sample values) for each sampling point (sampling timing) sampled in time series. The coordinate value i on the time axis increases by one for each sampling period. Therefore, data output from a circuit that outputs a digital signal, such as the input processing unit 2 and the original data bit expansion unit 5, constitutes a data string, but in the following description, it is simply referred to as data. is there.

  Conversion from the audio signal SA to the audio signal data DI in the input processing unit 2, supply of the audio signal data DI from the input processing unit 2 to the original data bit expansion unit 5, bit extension in the original data bit expansion unit 5, The supply of the audio signal data DS from the data bit extension unit 5 to the data storage unit 7 (the operation of other parts of the extension data generation processing unit 3 described below is the same) is supplied from a control unit (not shown). (Same frequency as the sampling frequency, or an integral multiple of the sampling frequency or a fraction of an integer).

  The one-dimensional m-th order ε filter unit 6 receives a sequence of n + α-bit audio signal data output from the original data bit extension unit 5 and simply selects a target sampling point (hereinafter, “target sampling point” as “target point”). A portion where there is a steep and large change based on the audio signal data of the other sampling points before and after the point of interest (that is, sampling point data aligned in the time axis direction). Is used as an edge-preserving smoothing filter that performs smoothing by treating a small amplitude component as noise, and a one-dimensional m-th order ε filter unit 6 shown in FIG. 1 difference calculation unit 8-1 to m-th difference calculation unit 8-m, first ε determination unit 9-1 to m-th ε determination unit 9-m, weighted average unit with determination 10, and data addition unit 11 is provided. By performing smoothing using the data width increased by the original data bit extension unit 5, the distortion due to the insufficient number of quantization bits is eliminated, and the effective number of quantization bits is increased.

The sequence of the audio signal data DO with the effective number of quantization bits increased by the one-dimensional m-order ε filter unit 6 is output to the output processing unit 4, and the output processing unit 4 outputs the n + α-bit audio signal data DO. For example, the data of the column is converted from digital to analog, and subjected to analog gain control, equalizing, and the like.
In addition, the output processing unit 4 is assumed to perform other processes such as conversion from an electrical signal to an optical signal, serial-parallel and parallel-serial conversion of digital data.

  Here, the ε filter applied to the audio signal will be described. The ε filter is described in Non-Patent Document 1 described above, for example.

ここで、ｘ（ｉ）はサンプリングタイミングｉの入力データ値、ｙ（ｉ）は入力データｘ（ｉ）が入力されたときの出力データ値、ｍは低域通過フィルタの次数、ａ_ｋはフィルタ係数、ｋはサンプリングタイミングｉの前後のサンプリングタイミングの、サンプリングタイミングｉからの時間差（相対的な位置）をサンプリング周期数で表す値、εは閾値である。
式（１）は、サンプリング点ｉを注目点としたときのフィルタリング処理を表すものである。One-dimensional processing in the time axis direction by the ε filter is expressed by, for example, Expression (1).

Here, x (i) is the input data value at the sampling timing i, y (i) is the output data value when the input data x (i) is input, m is the order of the low-pass filter, and _ak is the filter A coefficient, k is a value representing the time difference (relative position) from the sampling timing i of the sampling timing before and after the sampling timing i, and ε is a threshold value.
Equation (1) represents the filtering process when the sampling point i is the point of interest.

  The above equation (1) is for the case where m is an even number, and the data of the point of interest, the sampling data of the point of interest before the (m / 2) sampling cycle, the point of interest {(m / 2) −1 } It is described as an expression for filtering using a total of m pieces of sampling data up to after the sampling period. If ε filter processing is performed using equal sampling data before and after the attention point, m is defined as an odd number, and the attention point (m−1) / Expression (1) may be expressed so as to calculate sampling data up to two sampling periods later. Of course, if ε filter processing is performed using only the sampling data after the attention point, m is not defined as an odd number or an even number, and sampling data from the attention point data up to m sampling cycles after the attention point is obtained. Ε filter processing may be used. That is, what range of sampling data should be used for the ε filter processing may be set optimally according to the applied signal.

FIG. 2 is a diagram illustrating f (u) in the equation (1). When the absolute value of u is less than or equal to the threshold ε, f (u) = u, and when the absolute value of u is greater than the threshold ε, f (u) = 0. As a result, the filter input, the level, and the filter output level become the same, and the signal change larger than the threshold value ε before and after the input remains as it is.
The following description is based on the formula (1). In the above example, x (ik) -x (i) is supplied as u.

Difference x (ik) − between the speech signal data x (i) at the point of interest i and the speech signal data x (i−k) at the point (i−k) separated from the point of interest i by the k sampling period. When x (i) is small (difference is equal to or smaller than the threshold ε), f (u) = f {x (ik) −x (i)} is almost linear and the sum of the coefficients a _k is 1. Assuming equation (1) is rewritten as equation (2), which is equal to the weighted average filter.

Also, equation (1) can be rewritten as equation (3).

  When x ′ (ik) newly defined when the nonlinear function of FIG. 2 is applied to f (u) of Expression (3), Expression (3) is rewritten as Expression (4). Can do. In Expression (4), x ′ (ik) becomes x (ik) when the difference x (ik) −x (i) is less than or equal to the threshold ε, and the difference x (ik) − When x (i) is larger than the threshold ε, x (i) is obtained.

  3 and 4 are diagrams illustrating an operation when a signal that changes in a stepwise manner is input to the ε filter. 3A and 3B show the case where the step-like change width is equal to or smaller than the threshold ε, and FIGS. 4A and 4B show the case where the step-like change width is larger than the threshold ε. Indicates. 3A and 4A show the waveform of the input signal x (i) of the ε filter, and FIGS. 3B and 4B show FIGS. 3A and 4B, respectively. The output signals y (a) and y (b) at the sampling points a and b of the ε filter when the signal a) is input are shown together with the input signal x (i).

In these figures, the horizontal axis is the time axis and sampling points (sampling timing)
i (i =... (a-3), (a-2), (a-1), a, b, (b + 1), (b + 2)...), and the vertical axis represents signals x (i) and y (i). Indicates the data value level. x (b) is an input signal of the sampling point b next to the sampling point a of x (a), and a case where a step-like change (rise) occurs between the sampling point a and the sampling point b is assumed. Yes. Further, x (b + 1), x (b + 2)... Are signals of sampling points i (b + 1), i (b + 2). , X (a-2), x (a-3)... Are sampling points (a-1), (a-2), (a) one sampling period, two sampling periods, three sampling periods before the sampling point a. a-3)...

In the following description, it is assumed that m = 6 (in the description with reference to FIGS. 3A, 3B, 4A, and 4B).
When an input signal waveform as shown in FIG. 3A is processed, according to the ε filter using the function f in FIG. 2, the levels of x (a) and x (b) of the sampling points a and b of the signal Is equal to or smaller than the threshold ε, the filtered signal y (a) at the sampling point a is a sampling value x (a−3), x (a−2), x (a−1) in the range Wa, x (a), x (b), x (b + 1) is a weighted average, and the range Wa includes data x (b), x (b + 1) at the level of the signal x (b). Data is also used for the weighted average. As a result, the signal y (a) after filtering has a larger value than the signal x (a) before filtering.

  Further, the filtered signal y (b) at the sampling point b is a sampling value x (a−2), x (a−1), x (a), x (b), x (b + 1) within the range Wb, x (b + 2) is a weighted average, and within the range Wb, data x (a-2), x (a-1), and x (a) at the level of the signal x (a) are included. Is also used for the weighted average. As a result, the signal y (b) after filtering has a smaller value than the signal x (b) before filtering.

  For example, when the data value level difference between the sampling point a and the sampling point b before bit extension is 1 LSB difference of 16-bit width data, each output level is a data value level difference that cannot be expressed by 16-bit width data processing. Become. However, since the audio signal output apparatus of this embodiment performs bit expansion processing before ε filtering, it performs filtering to a waveform in which the data value level difference of the input waveform is smoothed as in the output waveform of FIG. can do.

Next, when an input signal waveform as shown in FIG. 4A is processed, the signal x (a) at the sampling point a and the signal at the sampling point (b) are obtained according to the ε filter using the function f in FIG. Since the difference in the level of x (b) is larger than the threshold value ε, f {(x (ik) −x (i)} = 0 in equation (1), and the output after filtering of the sampling points a and b y (a) and y (b) are the same as the input levels x (a) and x (b) as shown in FIG.
When obtaining the filtering output y (a) at the sampling point a, the data x (b) and x (b + 1) at the level of the signal x (b) at the sampling point b are all replaced with the data value x (a) at the sampling point a. When the weighted average is obtained and the filtering output y (b) at the sampling point b is obtained, the data x (a-2), x (a-1), x (a) at the level of the signal x (a) at the sampling point a is obtained. By replacing all the data values x (b) at the sampling point b and obtaining a weighted average.
As a result of such processing, a sharp change in signal level is not deteriorated.
That is, the ε filter can generate data obtained by smoothing a small amplitude data value change in combination with a bit expansion process while maintaining a large amplitude and steep change in the input signal.

Here, when the coefficient _ak of the equation (1) is controlled, when the signal waveform of FIG. 3A is input, the output levels y (a) and y (b) at the sampling points a and b in FIG. ) From the input levels x (a) and x (b), that is, changes {y (a) −x (a)} and {y (b) −x (b)}.
The larger the coefficient value (weighted value) for the signal at the sampling point that is different from the sampling point of interest, the greater the amount of change in the output level from the input level.
If the amount of change is large, the input data value level change is smoothed to a smoothed output waveform.
On the other hand, if the amount of change is small, the input signal has a steep change waveform even when the input data value level changes below the threshold ε.
When the amount of change is reduced, a steep waveform change remains as in the case where the cutoff frequency of the low-pass filter is increased.

Weighted average unit with determination in Embodiment 1 (FIG. 1) so that the characteristics (smoothing process characteristic (low-pass filter characteristic)) of the edge preserving type smoothing filter part (one-dimensional m-order ε filter) can be adjusted. In place of 10, a weighted average unit 12 with coefficient programmable determination may be used as shown in FIG. 5, and by doing so, an audio signal having a function capable of freely setting the coefficient (weighted value) _ak An output device (FIG. 5) can be configured.
Such a weighted average unit with coefficient programmable determination 12 is, for example, an expression (1) or an expression for the outputs EE (1) to EE (m) of the first to mth determination units 9-1 to 9-m. The value of the coefficient _ak used when the calculation according to (5) is performed can be changed according to the setting from the outside.

  The weighted average unit with coefficient programmable determination 12 shown in FIG. 5 is the same as the weighted average unit with determination 10 in FIG. 1 except that the coefficient is variable. The other components shown in FIG. 5 are the same as those shown in FIG.

  When the ε filter processing of Expression (1) is performed, when the signal waveforms of FIG. 3A and FIG. 4A are input, the output level y (a) in FIG. 3B and FIG. , Y (b) to determine the sampling data before and after the target point (from the data before the sampling period (m / 2) to the data after the {(m / 2) -1} sampling period with reference to the target point) The higher the order m, the smoother the data in a wider range with respect to the point of interest, and the higher frequency components are suppressed, that is, the cut-off frequency of the low-pass filter. For example, instead of the weighted average unit with determination 10 in the first embodiment (FIG. 1), the weighted average unit 13 with order variable determination shown in FIG. By one dimension It is possible to configure the audio signal output device having the following ε variable to smoothing processing characteristics the degree m of a filter function which can be changed (the low-pass filter characteristic).

For example, the weighted average unit 13 with variable order of determination uses only one selected from the outputs EE (1) to EE (m) of the first to mth determination units 9-1 to 9-m. Thus, the calculation according to the formula (1) or the formula (5) can be performed, and any one of the outputs EE (1) to EE (m) is selected and the coefficient a _k to be multiplied by these is selected. The value of can be changed according to the setting from the outside.

  6 is the same as the weighted average unit with determination variable 10 shown in FIG. 1 except that the order is variable. The other components shown in FIG. 6 are the same as those shown in FIG.

In this embodiment, edge preserving type smoothing filter processing is performed using the data width increased to n + α bits by bit expansion so that the small amplitude change portion changes smoothly. Along with this, small amplitude variation noise is also reduced by the filter processing. That is, when the bit extension from n bits to n + alpha bits, the threshold value ε of ε filter to the data after the bit extension and 2 ^alpha. (The minimum data value level change before the bit extension, becomes to the 2 ^alpha calculated as data value level change after alpha-bit extension.) In this way, epsilon filter holds the change of sudden loud input voice signal However, the effective number of quantization bits can be increased.

  Hereinafter, as an example, a case where α = 2, that is, a case where an n-bit audio signal is subjected to a quantization bit shortage improvement process with n + 2 bits will be described. Note that the standard is that digital audio signals such as CDs are encoded by 16-bit quantization, and there are cases where the number of quantization bits is increased to 24 bits by increasing 8 bits. However, hereinafter, in order to simplify the description, a case where 2 bits are increased will be described.

  FIG. 7 is a diagram illustrating the configuration of the audio signal processing device according to the first embodiment when m = 4. The description will be made assuming that the threshold ε is 2 to the power of α (square) = 4.

  In FIG. 1, DM (1) to DM (m) are used as codes of data input to the difference calculation units 8-1 to 8-m, but in FIG. 7, the difference calculation units 8-1 to 8-8 are used. DM (l2), DM (l1), DM (c), and DM (r1) are used as codes of data input to -4. Data represented by codes DM (l2), DM (l1), DM (c), DM (r1) is represented by codes DM (1), DM (2), DM (3), DM when m = 4. It is the same as the data represented by (4). The same applies to the outputs ED (l2) to (r1) of the difference calculation units 8-1 to 8-4 and the outputs EE (l2) to EE (r1) of the ε determination units 9-1 to 9-4.

  FIGS. 8A and 8B are diagrams for explaining the operation of the input processing unit 2. The horizontal axis indicates the time axis i. FIG. 8A shows the signal change of the analog audio input signal SA, and FIG. 8B shows the data value level change of the digital audio signal data DI.

  An analog audio input signal SA as shown in FIG. 8A is input from the input terminal 1 to the input processing unit 2. The input processing unit 2 converts the analog audio input signal SA into n-bit audio signal data DI as shown in FIG. 8B and outputs it to the extended data generation processing unit 3. In the extension data generation processing unit 3, the n-bit audio signal data DI is input to the original data bit extension unit 5.

  The audio input signal SA shown in FIG. 8 (a) has a gentle change in signal level, but the quantization resolution with respect to the change in signal level is low (the number of bits is small), so FIG. 8 (b). The audio signal data DI shown is converted into binary data values (Y and Y + 1).

FIGS. 9A and 9B are diagrams for explaining the operation of the original data bit extension unit 5.
The horizontal axis indicates the time axis i. FIG. 9A shows the change in the data value level of the n-bit audio signal data DI, and FIG. 9B shows the change in the data value level of the n + 2 bit audio signal data DS. The original data bit extension unit 5 bit-extends n bits of audio signal data DI as shown in FIG. 9A by 2 bits (this bit extension has a value on the lower side of the least significant bit of the data DI. This is done by adding two-digit bits of “00”), and outputs n + 2 bits of audio signal data DS as shown in FIG. 9B to the one-dimensional fourth-order ε filter unit 14.

  Since the n + 2 bit audio signal data DS as shown in FIG. 9B has been expanded by 2 bits, the data value level Y of the audio signal data DI in FIG. 9A is 4Y, and Y + 1 is 4 (Y + 1). ).

  As described above, most digital audio signals are standardized by 16 bits. The 16-bit audio signal data may be expanded by 8 bits. At this time, the data value level Y of the audio signal data DI is converted to 256Y, and Y + 1 is converted to 256 (Y + 1).

  The data storage unit 7 in FIG. 7 holds data in a sequence of n + α-bit audio signal data continuous in the time axis direction and outputs the data simultaneously, and is configured as shown in FIG. 10, for example. The illustrated data storage unit 7 includes four flip-flops FF_1 to FF_4 connected in cascade, and the input audio signal data DS is transferred by the four flip-flops FF_1 to FF_4 in synchronization with the sampling clock. . That is, the input audio signal data DS is input to the flip flop FF_1, the output of the flip flop FF_1 is input to the flip flop FF_2, the output of the flip flop FF_2 is input to the flip flop FF_3, and the output of the flip flop FF_3 is flip flop. FF_4 is supplied to the input and held.

As a result, the input audio signal data at four consecutive sampling points can be simultaneously held and output simultaneously by the four flip-flops FF_1 to FF_4.
That is, the data held in the flip-flop FF2 is output as data representing the data value level DM (c) of the attention point (timing) c, and the data held in the flip-flop FF_3 is one sampling of the attention point. The data that is output as data representing the data value level DM (l1) of the audio signal before the cycle and held in the flip-flop FF_4 is the data value level DM (l2) of the audio signal two samples before the target point. The data that is output as data that is held and held in the flip-flop FF_1 is output as data that represents the data value level DM (r1) of the audio signal after one sampling period of the target point.

  FIG. 11 shows the output of the data storage unit 7 when the sampling point c is the point of interest. The horizontal axis represents time i, and the vertical axis represents the audio signal data DS input to the data storage unit 7 and the data value level of the audio signal data DM output from the data storage unit. In the illustrated example, the data storage unit 7 outputs DM (l2) = 4Y, DM (l1) = 4Y, DM (c) = 4Y, DM (r1) = 4Y + 4, which are the first difference calculation unit. 8-1 to the fourth difference calculation unit 8-4. As described above, the data storage unit 7 stores the audio signal data DM (c) at the point of interest c and the audio signal data DM (l2) at the sampling points around (at the front and rear of) the point of interest that is continuous (before and after) the point of interest. ), DM (l1), and DM (r1) are output simultaneously.

  The difference calculation units 8-1 to 8-4 in FIG. 7 include the audio signal data DM (c) at the target sampling point and the sampling points before and after the target sampling point among the data output from the data storage unit 7. Difference data representing differences from each of the audio signal data DM (l2), DM (l1), and DM (r1).

  Specifically, the first difference calculation unit 8-1 generates difference data ED (l2) between the data value DM (l2) at the sampling point l2 and the data value DM (c) at the point of interest c (sampling point c). And output to the first ε determination unit 9-1 and the weighted average unit 10 with determination. Similarly, the second difference calculation unit 8-2 to the fourth difference calculation unit 8-4 also use the difference data ED (l1) to ED (based on the data values DM (l1) to DM (r1) before and after the attention point, respectively. r1) are generated and output to the second ε determination unit 9-2 to the fourth ε determination unit 9-4 and the weighted average unit with determination 10, respectively. As described above, the difference calculation units 8-1 to 8-4 each receive the audio signal data DM (c) at the point of interest and the audio signal data DM at the sampling points around the point of interest (before and after) of the point of interest. The difference between each of (l2), DM (l1), and DM (r1) is calculated, and difference data representing the difference is output.

  FIG. 12 shows the data DM (k) output from the data storage unit 7 when the sampling point c in FIG. 11 is the point of interest, and the first difference calculation unit 8-1 to the fourth difference based on the data DM (k). Data ED (k), EE (k), and EM generated by the calculation unit 8-4, the first ε determination unit 9-1 to the fourth ε determination unit 9-4, and the weighted average unit with determination 10 The value of f {ED (k)} obtained in the process of calculation in the weighted average unit with determination 10 is also shown.

The first difference calculation unit 8-1
ED (l2) = DM (l2) -DM (c) = 4Y-4Y = 0,
The second difference calculation unit 8-2
ED (l1) = DM (l1) -DM (c) = 4Y-4Y = 0,
The third difference calculation unit 8-3
ED (c) = DM (c) -DM (c) = 4Y-4Y = 0,
The fourth difference calculation unit 8-4
ED (r1) = DM (r1) -DM (c) = (4Y + 4) -4Y = 4,
Is output.

For each of the sampling points before and after the target sampling point, the ε determination units 9-1 to 9-4 use the difference data ED (l2) to ED (r1) output from the difference calculation units 8-1 to 8-4 as threshold values. Determination data indicating whether or not is larger than ε is generated.
Specifically, the first ε determination unit 9-1 to the fourth ε determination unit 9-4 receive the difference value ED (k) and determine whether ED (k) is greater than ε. The determination result is output to the weighted average unit 10 with determination as determination data EE (k) shown in FIG. In the illustrated example, “0” is output as EE (k) when ED (k) is greater than ε, and “1” is output as EE (k) when ED or less.

That is, the first ε determination unit 9-1 generates “0” when the difference data ED (l2) is larger than ε, and generates “1” when it is equal to or less than ε.
Similarly, the second ε determination unit 9-2 to the fourth ε determination unit 9-4 also generate determination data EE (l1) to EE (r1) based on the difference data ED (l1) to ED (r1). .
The generated determination data is supplied to the weighted average unit 10 with determination.
Since α = 2 here, the threshold ε = 4 is set.

  In the example shown in FIG. 12, the first, second, and third ε determination units 9-1, 9-2, and 9-3 include difference data ED (l2), ED (l1), and ED (c). “0” is input as “4”, and “4” is input as the difference data ED (r1) to the fourth ε determination unit 9-4, and these difference data ED (l2) to ED (r1) are all equal to or less than ε. Therefore, the determination data EE (l2) to EE (r1) are all “1”.

The weighted average unit with determination 10 includes determination data EE (l2) to EE (r1) output from the ε determination units 9-1 to 9-4 and the difference calculation unit 8-1 for each of the sampling points before and after the sampling point of interest. A weighted average value is generated based on the difference data ED (l2) to ED (r1) output by ˜8-4. Specifically, the weighted average unit 10 with determination performs weighted averaging of the difference data ED (l2) to ED (r1) based on the determination data EE (l2) to EE (r1), and the weight obtained as a result thereof Average data EM representing the average value is output. When the determination data is “1”, the coefficient _ak is multiplied by the difference data and added. When the determination data is “0”, no addition is performed. In other words, in accordance with the value of the determination data, an operation for obtaining a weighted average of the function f {ED (k)} having the relationship shown in FIG. 13B with respect to the difference value ED (k) is performed.

When the coefficients a _k (= a _l2 , a _l1 , a _c , a _r1 ) are all set to 0.25 (for the sake of simplicity, the sum is set to ¼ of the sum 1), the determination data EE (l2 ) To EE (r1) are all “1” in the above example, the output EM of the weighted average unit with determination 10 is given by the following equation (5).
EM
= A _l2 × ED (l2) + a _l1 × ED (l1) + a _c × ED (c) + a _r1 × ED (r1)
= 0.25 × 0 + 0.25 × 0 + 0.25 × 0 + 0.25 × 4
= 1
... (5)

The above processing is performed on the audio signal data DS supplied from the original data bit expansion unit 5 with each sampling point as an attention point in order, and the same calculation as in Expression (5) is performed.
FIG. 14 shows the weighted average value EM corresponding to the audio signal data DM of FIG. 11, that is, the weighted average value EM when each sampling point is a point of interest. In FIG. 14, the horizontal axis indicates the attention point, and the vertical axis indicates the data value of the weighted average value EM. The weighted average unit 10 with determination outputs the weighted average value EM to the data adding unit 11.

FIGS. 15A to 15C are diagrams for explaining the operation of the data adding unit 11. The horizontal axis is the time axis and indicates the sampling timing i.
FIG. 15A shows the data value level change of the audio signal data DS, FIG. 15B shows the data value level change of the weighted average value EM, and FIG. 15C shows the data value level of the audio signal data DO. Showing change.
The data adder 11 adds the audio signal data DS shown in FIG. 15A and the weighted average value EM shown in FIG. 15B, and outputs the audio signal data DO shown in FIG. To do. For example, at the sampling timing d in FIGS. 15A to 15C (when processing is performed with the sampling timing d as a point of interest), since DS (d) = 4Y and EM (d) = 1, DO (d) = DS (d) + EM (d) = 4Y + 1
It becomes.

As described above, the present embodiment can increase the effective number of quantization bits in a region where the data value level changes gradually. When an n-bit audio signal is converted into n + 2 bits, the analog input audio signal SA changes gently, and the data value level change of the audio signal data DS bit-extended corresponding to this changes is shown in FIG. As shown in FIG. 15C, when jumping from 4Y to 4Y + 4 (4 = 2 ² ), interpolation can be performed using 4Y + 1, 4Y + 2, 4Y + 3, and smoothing can be performed. That is, a signal change that cannot be expressed before bit expansion can be obtained, and the waveform of the original sound (analog input sound signal SA) can be made closer than before the filter processing.

Next, an operation when a signal whose signal amplitude changes suddenly and greatly (changes stepwise with a width larger than the threshold ε) is input to the ε filter will be described. FIGS. 16A to 16E are diagrams for explaining the operation in the case where a signal whose signal amplitude changes suddenly and greatly is input to the audio signal processing apparatus. The horizontal axis represents the sampling timing i on the time axis.
FIG. 16A shows the amplitude change waveform of the analog audio input signal SA, FIG. 16B shows the data value level change of the n-bit audio signal data DI, and FIG. 16C shows the n + α-bit audio. FIG. 16D shows a data value level change of the weighted average value EM, and FIG. 16E shows a data value level change of the n + α-bit audio signal data DO. Is shown.

  An analog audio signal SA as shown in FIG. 16A is input from the input terminal 1 to the input processing unit 2. The input processing unit 2 converts the analog audio signal SA into digital audio signal data DI (binary data value levels (Y and Y + 4 for n bits)) as shown in FIG. Output to the bit extension unit 5.

  In the original data bit expansion unit 5, the data value level Y of the audio signal data DI of FIG. 16B is converted to 4Y and Y + 4 is converted to 4 (Y + 4), and the audio signal data as shown in FIG. The DS is output to the one-dimensional fourth-order ε filter unit 6.

  FIG. 17 shows the data DM (k) output from the data storage unit 7 when the sampling point c in FIG. 16C is the point of interest, and the first difference calculation units 8-1 to 8-1 based on the data DM (k). Data ED (k), EE (k) generated by the difference calculation unit 8-4, the first ε determination unit 9-1 to the fourth ε determination unit 9-4, and the weighted average unit 10 with determination , EM, and the value of f {ED (k)} obtained in the process of calculation in the weighted average unit 10 with determination is also shown.

  The data storage unit 7 sets DM (l2) = 4Y, DM (l1) = 4Y, DM (c) = 4Y, DM (r1) = 4Y + 16 to the first difference calculation unit 8-1 to the fourth difference calculation unit 8. Output to -4.

The first difference calculation unit 8-1
ED (l2) = DM (l2) -DM (c) = 4Y-4Y = 0,
The second difference calculation unit 8-2
ED (l1) = DM (l1) -DM (c) = 4Y-4Y = 0,
The third difference calculation unit 8-3
ED (c) = DM (c) -DM (c) = 4Y-4Y = 0,
The fourth difference calculation unit 8-4
ED (r1) = DM (r1) -DM (c) = (4Y + 16) -4Y = 16,
Is output.

  Accordingly, “0” is input to the first ε determination unit 9-1 to the third ε determination unit 9-3, and “16” is input to the fourth ε determination unit 9-4. When the threshold value ε is set to “4”, the determination data EE (l2) to EE (c) are “1” because ED (l2) to ED (c) are equal to or less than the threshold value, and ED (r1) is equal to the threshold value ε. Since it is larger, EE (r1) becomes “0”.

The weighted average unit 10 with determination multiplies the coefficient _ak by the difference data when the determination data is “1”, and does not add it when it is “0”. When the coefficients a _k (= a _l2 , a _l1 , a _c ) are all 0.25, the output EM of the weighted average unit with determination 10 is given by the following equation (6).
EM = a _l2 × ED (l2) + a _l1 × ED (l1) + a _c × ED (c)
= 0.25 × 0 + 0.25 × 0 + 0.25 × 0
= 0
(6)

The above processing is performed on the audio signal data DS supplied from the original data bit expansion unit 5 with each sampling timing as an attention point in order, and the same calculation as in Expression (6) is performed.
FIG. 16D shows the data level change of the weighted average value EM corresponding to the audio signal data DM of FIG. 16C, that is, the weighted average value EM when each sampling timing i is the point of interest.
In FIG. 16D, the horizontal axis represents the sampling timing i of the target point, and the vertical axis represents the data value of the weighted average value EM.
The weighted average unit 10 with determination outputs the weighted average value EM to the data adding unit 11.

  The data adder 11 adds the audio signal data DS shown in FIG. 16C and the weighted average value EM shown in FIG. 16D, and outputs the audio signal data DO shown in FIG. To do.

  As shown in FIG. 16E, the audio signal data DO has a steep change similarly to the input SA or DI, and the ε filter sharpens a sharply changing region when the signal amplitude changes greatly. The degree can be saved.

As described above, when the quantization bit shortage improvement process for increasing the n-bit audio signal data to n + α bits is performed, the threshold ε is set to 2 α so that the minimum data value level change before bit expansion can be further finely interpolated. If the power is set, the signal waveform distortion due to the insufficient number of quantization bits due to the difference in bit resolution between the input audio signal and the output audio signal is reduced. That is, the effective number of quantization bits can be increased. Further, it is possible to increase the number of quantization bits in a region where the signal level changes gradually, while preventing the dynamic amplitude change characteristics of the audio signal from being deteriorated.
Incidentally, to the threshold value 2 ^alpha is the minimum data value level change 2 bits before expansion, but is taken into consideration that when converted becomes 2 ^alpha as a data value level change after alpha bit extension, actual use The threshold may be set to a value other than 2α for optimization according to the state.

FIG. 18 is a block diagram showing an audio signal output device obtained by modifying the audio signal output device shown in FIG.
The audio signal output device of FIG. 18 is generally the same as the audio signal output device shown in FIG. 1, but instead of the output processing unit 4 of FIG. 1, a digital or analog type amplitude control programmable amplifier 15a is incorporated. An output processing unit 15 is provided.
The programmable amplifier 15a can change the gain (amplification factor) in accordance with a gain control signal GC from the comparison unit 16 described later.
The comparison unit 16 receives the data DM (c) of the target point output from the data storage unit 7 and the output DO of the data addition unit 11 and outputs a gain control signal GC corresponding to the ratio between them. The gain control signal GC is supplied to the amplitude control programmable amplifier 15a of the output processing unit 15.

The operation of the audio signal output apparatus shown in FIG. 18 will be described. The input processing unit 2, the original data bit extension unit 5, and the one-dimensional mth-order ε filter unit 6 operate in the same manner as described in the example of FIG.
The comparison unit 16 receives the data DM (c) of the attention point output from the data storage unit 7 and the data addition unit output DO, and compares these data. For example, when the ratio of these data exceeds a certain level, the gain of the amplitude control programmable amplifier 15a, and hence the output amplitude of the output processing unit 15, is controlled by the output data of the comparison unit 16.
For example, the ratio (DO / DM (c)) of each input is obtained in the comparison unit 16, and this reciprocal is used for control. That is, the smaller the input ratio DO / DM (c), the larger the gain of the amplitude control programmable amplifier 15a.

  With this configuration, it is possible to correct the influence of the performance of the low-pass filter that accompanies the smoothing operation of the one-dimensional mth-order ε filter unit 6. For example, when the high-frequency amplitude of the signal data is decreased due to the performance of the low-pass filter, it is determined that the amplitude of the data addition unit output DO is smaller than the amplitude of the data DM (c) of the attention point input to the comparison unit 16. Therefore, the control signal GC determined based on the ratio (DO / DM (c)) of these data is output, and the output processing unit 15 is operated to increase the amplitude while leaving the smoothed waveform. In this way, frequency correction can be performed.

  In addition, instead of controlling the output processing unit 15 with the output of the comparison unit 16, another amplitude control unit disposed downstream from the output processing unit 15 is controlled (thereby correcting the output of the data adding unit 11). In this case, the same amplitude correction effect as described above can be obtained.

  FIG. 19 is a flowchart showing the processing steps of the audio signal processing device according to the present embodiment described above. In the following description, it is assumed that the order of the ε filter section is m, and the same data code as that used in FIG.

First, an analog audio input signal SA is input to the input terminal 1, and the input processing unit 2 receives the analog audio input signal SA and outputs n-bit audio signal data DI (ST1).
The audio signal data DI output from the input processing unit 2 is input to the original data bit extension unit 5 of the extension data generation processing unit 3, and the original data bit extension unit 5 bit-extends the audio signal data DI by α bits to n + α. Bit audio signal data DS is output (ST2).

The data storage unit 7 receives n + α-bit audio signal data DS, holds m (m ≧ 2 ^α ) continuous data on the time axis, and outputs them as DM (1) to DM (m). (ST3).
The first difference calculation unit 8-1 to the m-th difference calculation unit 8-m receive m pieces of continuous data DM (1) to DM (m), and the data of the point of interest c and its front and back And difference data ED (1) to ED (m) are output (ST4).

Difference data ED (1) to ED (m) are input to the first ε determination unit 9-1 to the m-th ε determination unit 9-m, and determination data EE (1) indicating whether to perform addition. To EE (m) are output (ST5).
Difference data ED (1) to ED (m) and determination data EE (1) to EE (m) are input to the weighted average unit 10 with determination, and the weighted average unit 10 with determination uses the difference based on the determination data. Data is weighted and a weighted average value EM is output (ST6).

  The data adding unit 11 receives the weighted average value EM and the original data DM (c) of the point of interest c, and the data adding unit 11 adds these and outputs the audio signal data DO (ST7). The audio signal data DO is input to the output processing unit 4, and the output processing unit 4 performs, for example, D / A conversion processing based on the audio signal data DO and outputs an audio signal (ST8).

Embodiment 2. FIG.
FIG. 20 is a diagram showing an audio signal output apparatus according to Embodiment 2 of the present invention. The audio signal output device of FIG. 20 is generally the same as the audio signal output device of FIG. 1, but the original data bit extension unit 5 of FIG. 1 is not provided, and the weighted average unit 10 with determination and data of FIG. Instead of the addition unit 11, a weighted average unit 17 with bit extension and determination and a data addition unit 18 with bit extension are provided.

  The difference calculation units 8-1 to 8-m and the ε determination units 9-1 to 9-m perform processing on n-bit data. The threshold value in the ε determination units 9-1 to 9-m is set to “1”. This is because there is no bit extension part before the ε filter part.

The weighted average unit 17 with bit function extension and determination performs a weighted average operation and bit extension, and outputs a weighted average value EM that is bit-extended by α bits. For example, this bit is generated by generating as the average data EM data representing the numerical values of the decimal places that appear when the weighted average with determination similar to that described in the example of FIG. 7 is performed. Can be extended.
The data adder with bit extension 18 has a bit adder 18a and a data adder 18b. The bit extender 18a bit-extends the input DM (c) to n + α bits. The data DMa (c) expanded in bit is output.
The data adding unit 18b adds the bit expanded data DMa (c) and the output EM of the weighted average unit 17 with bit extension and determination, and outputs data DO representing the addition result.

  As described above, in the audio signal output device of FIG. 20, the bit expansion is not performed before the ε filter unit, but is performed during the calculation of the ε filter. By configuring as described above, the same effect as the example shown in FIG. 1 can be obtained.

FIG. 21 is a block diagram showing an audio signal output device obtained by modifying the audio signal output device shown in FIG.
The audio signal output device of FIG. 21 is generally the same as the audio signal output device shown in FIG. 20, but instead of the output processing unit 4 of FIG. 20, a digital or analog amplitude control programmable amplifier 15a is incorporated. An output processing unit 15 is provided.
This programmable amplifier 15a can change the gain (amplification factor) in accordance with a gain control signal GC from the comparison unit 16, which will be described later, as shown in FIG.

Similarly to the data addition unit 18 with bit extension of FIG. 20, the data addition unit 18 with bit extension has a bit extension unit 18a and a data addition unit 18b. The bit extension unit 18a receives the input DM (c) as n + α. Bit extended data DMa (c) is output.
The data adder 18b adds the bit expanded data DMa (c) and the average data EM, and outputs data DO representing the addition result.
The output of the bit expansion unit 18 a is also output to the outside of the data addition unit 18 with bit expansion and is supplied to the comparison unit 16.

The comparison unit 16 receives the bit-expanded point-of-interest data DMa (c) and the addition result output DO from the data addition unit 18b, and outputs a gain control signal GC corresponding to the ratio.
The gain control signal GC is supplied to the amplitude control programmable amplifier 15a of the output processing unit 15.

The operation of the audio signal output device shown in FIG. 21 will be described. The input processing unit 2 and the one-dimensional m-order ε filter 19 with bit extension operate in the same manner as described in the example of FIG.
The comparison unit 16 receives the α-bit expanded point-of-interest data DMa (c) output from the bit expansion unit 18a and the addition result output DO of the data addition unit 18b, and compares these data. For example, when the ratio of these data exceeds a certain level, the gain of the amplitude control programmable amplifier 15a, and hence the output amplitude of the output processing unit 15, is controlled by the output data of the comparison unit 16.
For example, the ratio (DO / DM (c)) of each input is obtained in the comparison unit 16, and this reciprocal is used for control. That is, the smaller the input ratio DO / DM (c), the larger the gain of the amplitude control programmable amplifier 15a.

  With this configuration, it is possible to correct the influence of the low-pass filter performance that accompanies the smoothing operation of the one-dimensional m-order ε filter unit 19 with bit extension. For example, when the high-frequency amplitude of the signal data is reduced due to the performance of the low-pass filter, the amplitude of the addition result output DO of the data addition unit 18b is smaller than the amplitude of the data DMa (c) of the target point input to the comparison unit 16. Since the control signal GC determined based on the ratio of these data (DO / DM (c)) is output, the output processing unit 15 increases the amplitude while leaving the smoothed waveform. To work.

  As described in the example of FIG. 18, instead of controlling the output processing unit 15 with the output of the comparison unit 16, another amplitude control unit arranged downstream from the output processing unit 15 is controlled ( In this case, the same amplitude correction effect as described above can be obtained.

  20 and FIG. 21, the same modification as that described with reference to FIG. 5 and FIG. 6 (use of weighted average unit 12 with coefficient programmable determination, or variable order) Use of the weighted average unit 13 with determination) can be added.

Embodiment 3 FIG.
FIG. 22 is a diagram showing a configuration of an audio signal processing device according to Embodiment 3 of the present invention. The audio signal processing apparatus according to Embodiment 3 includes an input terminal 21, a frequency amplitude estimation unit 22, a filter coefficient generation unit 23, an original data bit extension unit 24, and an edge preserving smoothing filter unit 25.

An n-bit (n is a positive integer) audio signal X is supplied to the input terminal 21. The audio signal X corresponds to the audio signal data DI in FIG. FIG. 23 is a diagram illustrating an example of an n-bit audio signal according to the third embodiment. In FIG. 23A, the horizontal axis represents time i, and the vertical axis represents signal level. As shown in FIG. 23A, this audio signal is a sequence of digital signals for each successive sampling point, and the time (representing numerical value) i increases by one for each sampling point.
In FIG. 23B, the horizontal axis indicates frequency f, and the vertical axis indicates power and gain. FIG. 23 (a) shows a sine wave audio signal of frequency f1 quantized with n bits, the solid line in FIG. 23 (b) is the frequency spectrum of the audio signal in FIG. 23 (a), and the dotted line is (described later). The frequency characteristic of a low-pass filter is shown. As shown in the frequency spectrum of FIG. 23 (b), harmonics such as f2 and f3 exist due to quantization even though it is a sine wave of frequency f1.

  In this embodiment, harmonic components of an n-bit audio signal are removed, and waveform distortion of the audio signal due to quantization is corrected.

  Specifically, first, a frequency is estimated from the audio signal, and a filter coefficient of a low-pass filter having frequency characteristics for removing harmonics of the estimated frequency is generated. Further, not only the frequency but also the amplitude is estimated, and the order of the filter coefficient of the low-pass filter is determined. The setting of the order will be described together with the edge preserving smoothing filter.

  For example, when the frequency f1 is estimated as indicated by the solid line in FIG. 23B, a low-pass filter coefficient having a cutoff frequency characteristic as indicated by the dotted line in FIG. 23B is generated. Since f2 is obvious if the frequency f1 can be estimated, fc1 may be set between f1 and f2.

  Next, smoothing is performed on the audio signal that has been bit-extended using the generated low-pass filter coefficient, thereby removing harmonic components and correcting waveform distortion due to quantization. An edge preserving smoothing filter is used for smoothing.

  As described in connection with the first embodiment, the edge-preserving smoothing filter is a filter that smoothes only small changes while preserving the sharpness of a portion where there is a steep and large change, and an ε filter. , Trimmed average filter (DW-MTM filter), bilateral filter, and the like. The ε filter and the trimmed average value filter (DW-MTM filter) are described in Non-Patent Document 1 described above. Hereinafter, an edge preserving smoothing filter will be described as an ε filter as an example.

FIG. 24 is a diagram for explaining the operation of the ε filter. FIG. 24A shows an input signal x (i) of the ε filter, and FIG. 24B shows f {x (i1-k) −x (when the position i1 in FIG. i1)}
Is shown.

  When the signal having a gentle slope as shown in FIG. 24A is the input signal of the ε filter, if the order is too large, the difference between the attention point and the surrounding points exceeds the threshold value, as shown in FIG. Since the area is set to 0 and the weighted average is taken, the smoothing effect is weakened.

  Many edge-preserving smoothing filters such as the ε filter have the advantage of being able to preserve steep and large changes compared to low-pass filters. However, when the signal has a gentle slope, the smoothing performance is low. In the third embodiment, the inclination of the signal is taken into account by estimating the frequency and amplitude of the audio signal. When the frequency is the same and the amplitude is large, the slope becomes steep, so the order of the filter is reduced and the smoothing performance of the ε filter is improved.

  In the configuration of the third embodiment, an n-bit audio signal X is first input from the input terminal 21 to the frequency amplitude estimation unit 22 and the original data bit extension unit 24. The frequency amplitude estimation unit 22 estimates the frequency and amplitude of the input audio signal X and outputs the frequency F and the amplitude A to the filter coefficient generation unit 23. The filter coefficient generation unit 23 generates a low-pass filter coefficient C based on the frequency F and the amplitude A, and outputs the low-pass filter coefficient C to the edge preserving smoothing filter unit 25. The original data bit extension unit 24 is the same as the original data bit extension unit 5 shown in FIG. 1 and the like, and an edge-preserving smoothing is performed on an n + α bit audio signal X ′ obtained by extending the n bit audio signal X by α bits. To the filter 25. The edge-preserving smoothing filter unit 25 smoothes the n + α-bit speech signal X ′ using the low-pass filter coefficient C using the edge-preserving smoothing filter, and converts the n + α-bit speech signal Y into the n + α-bit speech signal Y. Output. As the edge preserving smoothing filter unit 25, the one-dimensional m-order ε filter unit 6 including the weighted average unit 12 with coefficient programmable determination shown in FIG. 5 can be used.

  The operation of the third embodiment when a sinusoidal audio signal X having a frequency f1 quantized with n bits as shown in FIG. 23A is input will be described. The audio signal X is input to the frequency amplitude estimation unit 22 and the original data bit extension unit 24.

  The frequency amplitude estimation unit 22 estimates the frequency F = f1 and the amplitude A = 2 from the audio signal X, and outputs the frequency F = f1 and the amplitude A = 2 to the filter coefficient generation unit 23.

  The filter coefficient generation unit 23 generates a filter coefficient having a cut-off frequency fc1 that removes the harmonic component of the frequency f1, and an order based on the frequency F = f1 and the amplitude A = 2, and performs edge-preserving smoothing Output to the filter unit 25.

  FIG. 25 is a diagram for explaining the operation of the original data bit extension unit 24. The horizontal axis indicates time i, and the vertical axis indicates the signal level. FIG. 25A shows an n-bit audio signal X, and FIG. 25B shows an n + α-bit audio signal X ′. The original data bit extension unit 24 bit-extends the n-bit audio signal as shown in FIG. 25A by α bits, and converts the n + α-bit audio signal as shown in FIG. To the filter 25.

  The edge preserving type smoothing filter unit 25 performs a smoothing process using an ε filter. The ε filter treats the number of bits increased by bit expansion as a small amplitude component. That is, when bit expansion is performed from n bits to n + α bits, the threshold ε of the ε filter is set to 2 to the power of α. Thereby, a step of 1 LSB or less in n bits is smoothed.

  FIG. 26 is a diagram for explaining the operation of the edge preserving smoothing filter unit 25. The horizontal axis indicates time i, and the vertical axis indicates the signal level. 26A shows an n + α-bit audio signal X ′ (i), and FIG. 26B shows an n + α-bit audio signal Y (i) subjected to the ε filter process. The edge preserving smoothing filter unit 25 corrects the waveform distortion by removing harmonics from the n + α-bit audio signal as shown in FIG. 26 (a), and the n + α-bit as shown in FIG. 26 (b). Output audio signals.

  As described above, the third embodiment can output an n + α-bit audio signal in which waveform distortion is corrected by removing harmonics from the n-bit audio signal.

  FIG. 27 is a diagram illustrating a detailed configuration of the frequency amplitude estimation unit 22. As shown in FIG. 27, the frequency amplitude estimation unit 22 includes an inflection point detection unit 27, a frequency estimation unit 28, and an amplitude estimation unit 29.

The inflection point detection unit 27 detects a point at which the n-bit audio signal starts or ends a series of monotone increases and a point at which a series of monotone decreases starts or ends as inflection points. Here, “a series of monotonous increases” means a continuous increase in which no decrease occurs in the middle (the same value may continue). Within this series of monotonically increasing intervals,
X (i) ≦ X (i + 1)
Are continuously satisfied (at all sampling points i).
Similarly, “a series of monotonous decreases” means a series of decreases in which no increase occurs in the middle (the same value may continue). Within this series of monotonically decreasing intervals,
X (i) ≧ X (i + 1)
Are continuously satisfied (at all sampling points i).

The illustrated inflection point detection unit 27 includes a first derivative calculation unit 31 and a sign change point detection unit 32.
The primary differential calculation unit 31 calculates and outputs primary differential data D from the input n-bit audio signal. For example, the first derivative calculation unit 31
The n-bit audio signal at each sampling point is represented by X (i),
When the n-bit audio signal at the next sampling point is represented by X (i + 1),
D (i) = X (i + 1) −X (i)
D (i) obtained in step 1 is output as first-order differential data.

The sign change point detection unit 32 detects a point where the sign of the primary differential data D has changed positively and a point where the sign has changed negatively as an inflection point. More specifically, the sign change point detection unit 32 detects that the primary differential data D has changed from a state where the sign is negative or a state where the sign is negative to a state where the sign is positive (hereinafter referred to as “change to positive”). Point)), and the point where the sign changes from a positive or zero state to a negative sign (hereinafter referred to as “change point to negative”) Detect as “point”. The output of the sign change point detection unit 32 is binary data. From the “change point to positive” to the “change point to negative”, the first value (for example, a high level value) is taken. From the “change point to negative” to the “change point to positive”, a second value (for example, a low level value) is taken.
Such an operation of the sign change point detection unit 32 can be said to convert the primary differential data D into binary data with only a sign (however, in the case of 0, the sign of the previous data is used).
The inflection point detection unit 27 configured by the first-order differential calculation unit 31 and the sign change point detection unit 32 has a point where an n-bit audio signal starts a series of monotone increases and a point where a series of monotone decreases starts. Is detected as an inflection point.

The frequency estimation unit 28 calculates the section length of the inflection point of the audio signal and the previous (immediately preceding) inflection point and outputs it as the frequency F.
The amplitude estimation unit 29 outputs the difference in level between the audio signal inflection point and the previous (immediate) inflection point audio signal as an amplitude A.

FIG. 28 is a diagram for explaining the operation of the frequency amplitude estimator 22. The horizontal axis indicates time i. 28A shows an n-bit audio signal X (i), FIG. 28B shows primary differential data D (i) of the audio signal, FIG. 28C shows binary data, 28 (d) indicates the inflection point position S of the audio signal. When the n-bit audio signal is input in FIG. 28A, the primary differential calculation unit 31 calculates primary differential data as shown in FIG. The sign change point detection unit 32 converts the primary differential data D into binary data having only a sign as shown in FIG. 28C, and a position S where the binary data as shown in FIG. 28D changes. i1, i2, and i3 (the positions on the time axis are indicated by dotted lines extending in the vertical direction) are detected. Therefore, the frequency estimation unit 28 and the amplitude estimation unit 29 obtain F = 1 / (i2−i1) as the frequency and A = | X (i2) −X (i1) |
In the interval from i2 to i3, F = 1 / (i3−i2) is obtained as the frequency and A = | X (i3) −X (i2) | is obtained as the amplitude.
The obtained frequency F and amplitude A are supplied to the filter coefficient generation unit 23.

FIG. 29 is a diagram for explaining the operation of the frequency amplitude estimation unit 22 when an audio signal different from that in FIG. 28 is input. The horizontal axis indicates time i. 29A shows an n-bit audio signal X (i), FIG. 29B shows primary differential data D (i) of the audio signal, FIG. 29C shows binary data, 29 (d) shows the inflection point position S of the audio signal. The primary differential calculation unit 31 calculates primary differential data as shown in FIG. 29B when an n-bit audio signal is input in FIG. The sign change point detector 32 converts the primary differential data into binary data having only a sign as shown in FIG. 29C, and a position S = i4 where the binary data as shown in FIG. 29D changes. , I5, i6 (the positions on the time axis are indicated by dotted lines extending in the vertical direction). Therefore, the frequency estimation unit 28 and the amplitude estimation unit 29 obtain F = 1 / (i5-i4) as the frequency and A = | X (i5) −X (i4) | In the interval i5 to i6, F = 1 / (i6-i5) is obtained as the frequency, and A = | X (i6) −X (i5) |
The obtained frequency F and amplitude A are supplied to the filter coefficient generation unit 23.

As described with reference to FIGS. 28 and 29,
The position where the sign of the primary differential data of the audio signal calculated by the frequency amplitude estimation unit 22 shown in FIG. 27 changes (“change point to positive” and “change point to negative”) is the inflection point of the audio signal. It can be estimated that the reciprocal of the time interval between the inflection points that are treated and the phase is the frequency of the audio signal. In addition, since the inflection point is the start point of a series of monotone increases or the start point of a series of monotone decreases, it can be estimated that the difference between the signal values of successive inflection points is the amplitude.

In the example of FIGS. 28 and 29, the audio signal is composed of a single frequency component.
Next, the case where the audio signal is composed of a plurality of frequency components will be described with reference to FIGS. 30 and 31. FIG.
30A shows the signal component of frequency f11 and amplitude a11, FIG. 30B shows the signal component of frequency f12 (> f11) and amplitude a12, and FIG. 30C shows FIG. 30A. The signal component of FIG. 30 and the signal component of FIG. The dotted line extending in the vertical direction in the figure indicates the temporal position of the inflection point. The time interval (section length) between the inflection points that follow each other is used as a half cycle of the signal component for frequency estimation.
At this time, in most regions as shown in FIG. 30 (c), the higher frequency f12 is estimated as the frequency, and a12 is estimated as the amplitude. Thus, in the section where f12 and a12 are estimated, A filter coefficient is selected based on the frequency f12 and the amplitude a12, and the waveform distortion is accurately corrected.

  In a section where the frequency is not estimated to be f12 (sampling result is a value larger than f12) like the sampling points i11 to i12, the frequency is estimated by approximating the overlapped signal with one sine wave. However, since the sections i11 to i12 are very short and a frequency higher than f12 is estimated, the problem caused by this is not significant.

FIG. 31 shows an example different from FIG.
FIG. 31A shows signal components of frequency f13 and amplitude a13, FIG. 31B shows signal components of frequency f14 (> f13) and amplitude a14, and FIG. 31C shows FIG. 31A. The signal component shown in FIG. 31 and the signal component shown in FIG. The dotted line extending in the vertical direction in the figure indicates the temporal position of the inflection point. The time interval (section length) between the inflection points that follow each other is used as a half cycle of the signal component for frequency estimation.
FIG. 31D shows an output of n + α bits obtained by processing the n-bit signal (FIG. 31C) in the third embodiment. As shown in FIG. 31 (c), f13, which is the higher frequency, is estimated in most regions, and the amplitude a13 is estimated.

However, in a section where f3 is not estimated as in the section of sampling points i13 to i14 (the estimation result is a value lower than f3), the frequency is estimated by approximating the overlapped signal with one sine wave. .
However, the correction amount in the interval from i13 to i14 is not more than 1/2 of the LSB of the signal before bit extension (n-bit signal), and the signal after bit extension (n + α-bit signal) is assumed to be n bits. When quantized, the correction amount becomes zero. That is, if an n + α-bit signal after bit expansion is quantized and returned to an n-bit signal, the input signal can be reproduced even in the period from i13 to i14.
Further, only the sections i13 to i14 have an influence, and are limited to only a part of the section as a whole.

  As described with reference to FIGS. 30 and 31, the positions (“change points to positive” and “negative” of the sign of the first derivative data of the audio signal calculated by the frequency amplitude estimation unit 22 shown in FIG. ")" Is treated as the inflection point of the highest frequency signal component of the audio signal, and the reciprocal of the time interval between the inflection points of the audio signal is the frequency of the highest frequency component of the audio signal. It can be estimated that. In addition, since the inflection point is the start point of a series of monotone increases or the start point of a series of monotone decreases, the difference between the signal values of successive inflection points is the highest frequency component contained in the audio signal. It can be estimated that it is an amplitude.

  In the frequency amplitude estimation unit 22 shown in FIG. 27, the frequency can be calculated by estimating the half cycle of the highest frequency component included in the audio signal, and at the same time, the amplitude is calculated from the difference in signal value at the inflection point. can do.

  The filter coefficients determined based on the frequency and the amplitude estimated in this way are used for filtering in the signal values of the respective sections (intersections between successive inflection points). For example, for the data from the sample point next to the inflection point i1 to the inflection point i2, using the filter coefficient determined by the frequency and amplitude obtained by the inflection points i1 and i2 in FIG. Filtering is performed.

The inflection point detection unit 27 detects a point at which an n-bit audio signal starts a series of monotone increases and a point at which a series of monotone decreases starts as an inflection point. The inflection point detection unit may be configured to detect, as an inflection point, a point at which a bit sound signal ends a series of monotone increases and a point at which a series of monotone decreases ends. In that case, for example, as shown in FIG. 32 (b), as a first-order differential calculation unit that receives the signal shown in FIG. 32 (a),
D (i) = X (i) -X (i-1)
As shown in FIGS. 32 (c) and 32 (d), the sign of the primary differential data D changes to negative as the sign change point detection unit using the output of D (i) obtained in step 1 as the primary differential data. What is necessary is just to use the point that is the closest positive point before the point and the point that was the closest negative point before the point where the sign of the primary differential data D changed to positive.
In FIG. 32, the temporal positions of the detected inflection points are indicated by dotted lines with symbols i21, i22, i23.

  FIG. 33 is a diagram for explaining the operation of the filter coefficient generation unit 23. The filter coefficient generation unit 23 includes a filter coefficient table for storing a plurality of low-pass filter coefficients having different orders from the cutoff frequency as shown in FIG. 33, and the low-pass filter is generated based on the frequency F and the amplitude A from the filter coefficient table. The filter coefficient C is selected and output to the edge preserving smoothing filter unit 25.

  FIG. 34 is a diagram for explaining a filter coefficient table that stores a plurality of low-pass filter coefficients having orders different from the cutoff frequency. The low-pass filter coefficient held in this table is determined based on the cut-off frequency characteristic of the low-pass filter based on the frequency of the audio signal, and the order of the low-pass filter based on the frequency and amplitude. For example, in the case of an audio signal having a frequency characteristic such as a solid line in FIG. 34A, a low-pass filter coefficient having a cutoff frequency characteristic such as a dotted line in FIG. (If the frequency f1 can be estimated, f2 is self-explanatory, so fc1 is set between f1 and f2.) In the case of an audio signal having a frequency characteristic as shown by the solid line in FIG. 34B, a low-pass filter coefficient having a cutoff frequency characteristic as shown by the dotted line in FIG. 34B is generated. (If frequency f4 can be estimated, f5 is self-explanatory, so fc4 is set between f4 and f5). In this way, the low-pass filter has a cut-off frequency characteristic corresponding to the frequency.

  Also, the order is determined by obtaining the slope from the frequency and amplitude. When the frequency is the same and the amplitude is large, the gradient becomes large, so the order of the filter needs to be small.

  As described above, the filter coefficient generation unit 23 can generate a filter coefficient with a lower order of the low-pass filter as the estimated amplitude A is larger by cutting the estimated harmonic component of the frequency F.

By estimating the frequency and amplitude of the third the speech signals embodiment, the frequency and for smoothing and generates a low-pass filter coefficients with a cutoff frequency characteristics and the order based on the amplitude quantization of the audio signal The waveform distortion due to can be accurately corrected. On the other hand, since the edge-preserving smoothing filter is applied at the time of smoothing, the reproducibility of the audio signal input waveform having a region where the signal amplitude changes sharply and greatly is not impaired.

  So far, the edge preserving smoothing filter unit 25 has been described as being an ε filter, but the third embodiment is not limited to this configuration. Similar effects can be expected with other edge-preserving filters such as a bilateral filter and a trimmed average value filter (DW-MTM filter).

  FIG. 35 is a flowchart showing processing steps of the audio signal processing device according to Embodiment 3 described above.

  First, an n-bit audio signal is input to the frequency amplitude estimation unit 22 from the input terminal 21. The frequency / amplitude estimation unit 22 estimates the frequency and amplitude from the n-bit audio signal and outputs them to the filter coefficient generation unit 23 (ST11). The filter coefficient generation unit 23 outputs the low-pass filter coefficient to the edge preserving smoothing filter unit 25 based on the frequency and amplitude (ST12). The original data bit extension unit 24 outputs an n + α bit audio signal obtained by extending the n bit audio signal by α bits to the edge preserving smoothing filter unit 25 (ST13). The edge-preserving smoothing filter unit 25 performs edge preserving smoothing filter processing on the n + α-bit audio signal using the generated low-pass filter coefficient, and outputs an n + α-bit audio signal (ST14).

  The audio signal processing apparatus shown in FIG. 22 can also be realized by software, that is, by a programmed computer. In this case, the processing procedure by the computer is the same as that described above with reference to FIG.

  According to the third embodiment, the frequency and amplitude are estimated from an n-bit audio signal in a relatively long time range, and smoothed using the low-pass filter calculated based on the frequency and amplitude. Harmonic components generated by quantization can be removed, waveform distortion of the audio signal due to quantization can be corrected, and audio signal processing that approximates the audio signal waveform before quantization than before can be realized. Further, by applying an edge preserving smoothing filter, it is possible to prevent the reproducibility of an audio signal having a region where the signal amplitude changes sharply and greatly.

  The edge preserving smoothing using the filter coefficient generated based on the estimation result of the frequency and amplitude of the audio signal described in the third embodiment can be combined with either the first embodiment or the second embodiment. . For example, the coefficients generated by the filter coefficient generation unit 23 in FIG. 22 may be supplied to the weighted average unit 12 with coefficient programmable determination shown in FIG. Also, in the speech processing apparatus of FIG. 18, a coefficient programmable unit may be used as the weighted average unit 10 with determination, and the coefficient generated by the filter coefficient generation unit 23 of FIG. 22 may be used as the coefficient. Similarly, in the speech processing apparatus of FIG. 20, as the weighted average unit 17 with bit extension and determination, a coefficient programmable unit is used, and the coefficient generated by the filter coefficient generation unit 23 of FIG. 22 is used as the coefficient. Also good.

  As an application example of the present invention, the present invention can be applied to an audio signal processing apparatus such as a car audio and a PA (portable audio).

Claims (15)

an original data bit extension unit for generating and outputting a sequence of n + α-bit audio signal data by extending each bit of the n-bit (n is an integer) audio signal data by α bits (α is an integer) bit;
An edge-preserving smoothing filter that smoothes a sequence of n + α-bit audio signal data generated by the bit extension ;
A frequency amplitude estimator for estimating the frequency and amplitude of the n-bit audio signal;
A filter coefficient generation unit that generates a low-pass filter coefficient based on the frequency and the amplitude,
The edge preserving smoothing filter unit performs smoothing using the filter coefficient generated by the filter coefficient generation unit,
The filter coefficient generation unit
Generating a filter coefficient of a low-pass filter that removes harmonic components of the frequency of the estimated sound signal, and the order of the low-pass filter is reduced as the estimated amplitude is increased. apparatus. The filter coefficient generation unit
A filter coefficient table that stores a plurality of low-pass filter coefficients having different orders from the cutoff frequency,
The audio signal processing device according to claim 1 , wherein the low-pass filter coefficient is selected based on the estimated frequency and amplitude from the filter coefficient table. The audio signal processing apparatus according to claim 1 , wherein the frequency amplitude estimation unit detects an inflection point of the audio signal and estimates a frequency and an amplitude from the detected inflection point. The frequency amplitude estimator is
An inflection point detecting unit that detects a point at which the n-bit audio signal starts or ends a series of monotone increases and a point at which a series of monotonic decreases starts or ends as an inflection point;
A frequency estimation unit that estimates a frequency from the section length of the position of the detected inflection point;
The audio signal processing apparatus according to claim 3 , further comprising: an amplitude estimation unit that estimates a level difference of the detected inflection point as an amplitude. The inflection point detector is
A first derivative calculating unit for calculating first derivative data of the n-bit audio signal;
The audio signal processing apparatus according to claim 4 , further comprising: a sign change point detection unit that detects, as the inflection point, a point at which a sign of the primary differential data has changed positively and a point at which the sign has changed negatively. . The first derivative calculation unit
The n-bit audio signal at each sampling point is represented by X (i),
When the n-bit audio signal at the next sampling point is represented by X (i + 1),
D (i) = X (i + 1) −X (i)
The audio signal processing apparatus according to claim 5 , wherein D (i) obtained in step (2) is output as the first-order differential data. The sign change point detection unit is configured such that the first derivative data is changed from a state where the sign is negative or zero to a state where the sign is positive, and a state where the sign is positive or zero. The audio signal processing apparatus according to claim 6 , wherein a point at which the sign has changed to a negative state is detected as a point at which the sign has changed.   The audio signal processing apparatus according to claim 1, wherein the edge preserving smoothing filter unit includes an ε filter unit.   The threshold ε of the ε filter unit is 2 ^α Is
  The audio signal processing apparatus according to claim 8. The audio signal processing apparatus according to claim 8 , wherein the ε filter unit is a one-dimensional m-th order ε filter unit that performs processing on a sequence of input signal data. The audio signal processing apparatus according to claim 10 , wherein a weight value of an average calculation performed in the one-dimensional m-th order ε filter unit can be changed. The audio signal processing apparatus according to claim 10 , wherein the filter order m of the one-dimensional m-order ε filter unit is changeable. The one-dimensional m-order ε filter unit is
A data storage unit that holds data in a row of n + α-bit audio signal data continuous in the time direction and simultaneously outputs the data;
Among the data stored in the data storage unit, a difference calculation unit that generates difference data between the audio signal data of the target sampling point and each of the audio signal data of the sampling points before and after the target sampling point;
For each of the sampling points before and after the target sampling point, an ε determination unit that generates determination data indicating whether or not the difference data output by the difference calculation unit is greater than a threshold;
For each of the sampling points before and after the sampling point of interest, a weighted average unit with determination that generates a weighted average value based on the determination data output by the ε determination unit and the difference data output by the difference calculation unit,
The audio signal processing apparatus according to claim 10 , further comprising: an addition unit that generates an addition value of the audio signal data of the sampling point of interest and the weighted average value data output from the weighted average unit with determination. A comparison unit for obtaining a ratio between the audio signal data of the target sampling point input to the data storage unit or the audio signal data of the target sampling point output from the data storage unit and the data output from the data addition unit; ,
The audio signal processing apparatus according to claim 13 , wherein correction is performed on the addition value output from the data addition unit in accordance with the ratio obtained by the comparison unit.   an original data bit extending step for generating and outputting a sequence of n + α-bit audio signal data by extending each bit of n-bit (n is an integer) audio signal data by α bits (α is an integer);
  An edge-preserving smoothing filter step for smoothing a sequence of n + α-bit audio signal data generated by the bit extension;
  A frequency amplitude estimating step for estimating the frequency and amplitude of the n-bit audio signal;
  A filter coefficient generation step for generating a low-pass filter coefficient based on the frequency and the amplitude,
  The edge preserving type smoothing filter step performs smoothing using the filter coefficient generated in the filter coefficient generation step,
  The filter coefficient generation step includes:
  A filter coefficient of a low-pass filter that removes harmonic components of the frequency of the estimated audio signal is generated, and the order of the low-pass filter decreases as the estimated amplitude increases.
  An audio signal processing method.

Images