JP2001326991

Patent Translate
Powered by EPO and Google
Notice
This translation is machine-generated. It cannot be guaranteed that it is intelligible, accurate,
complete, reliable or fit for specific purposes. Critical decisions, such as commercially relevant or
financial decisions, should not be based on machine-translation output.
DESCRIPTION JP2001326991
[0001]
BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an
audio processing apparatus for dividing an audio input signal into a plurality of frequency bands
for amplification by a plurality of amplifiers provided for each frequency band and outputting to
a plurality of speakers. .
[0002]
2. Description of the Related Art In a multi-channel speaker system in which an audio input
signal is divided into a plurality of frequency bands and dedicated speakers are driven by
dedicated amplifiers for each frequency band, a filter for dividing into a plurality of frequency
bands By using a digital filter instead of an analog filter, it is possible to realize a speaker
response with a linear phase and a flat frequency amplitude characteristic.
[0003]
FIG. 18 is a block diagram showing a multi-amplifier system speaker system disclosed in
Japanese Patent Laid-Open No. 3-143195 as a conventional example.
In FIG. 18, analog audio input signals are converted into respective digital signals by n A / D
converters 61 to 6 n, and then each digital signal converted by A / D converters 61 to 6 n is
divided into band division circuits. The frequency bands 41 to 4n are band-limited to the
09-05-2019
1
individual frequency bands, whereby the frequency bands are divided into n frequency bands.
The band division circuits 41 to 4n are all configured by FIR filters. Next, the output signals of
the band dividing circuits 41 to 4 n are respectively applied to the inverse filters 71 to 7 n to
cancel the amplitude and phase characteristics of the speaker units 31 to 3 n driven in each
band, and the amplitudes of the speaker units 31 to 3 n A frequency characteristic is added to
make the characteristic flat and make the phase characteristic straight. Similar to the band
dividing circuits 41 to 4 n, the inverse filters 71 to 7 n are constituted by FIR filters.
[0004]
Then, the output signals of the inverse filters 71 to 7 n are respectively applied to the delay
correction circuits 81 to 8 n and delayed so as to absorb the time difference generated in the
band division circuits 41 to 4 n and the inverse filters 71 to 7 n. Next, each output signal of the
delay correction circuits 81-8n is converted into an analog signal by the D / A converters 51-5n,
and then the analog signals converted by the D / A converters 51-5n are respectively amplified
by the amplifiers 21-2n. It is amplified and applied to the speaker units 31 to 3 n. In this method,
since the band dividing circuits 41 to 4n and the inverse filters 71 to 7n are formed by FIR
filters, the amplitude characteristics of the respective frequency bands can be made flat and the
phase characteristics can be made straight with high precision.
[0005]
However, in the above-mentioned conventional method, since the band dividing circuits 41 to 4n
and the inverse filters 71 to 7n are constituted by the FIR filter, many product-sum operations
and a delay memory are required. There is a problem of that. FIG. 3 shows the configuration in
the case where the FIR filter is realized by a digital signal processor (DSP). In FIG. 3, the input
signal S11 is delayed for each sample by the delay memories (z-1) 111 to 11k-1, each sample is
stored, and each of the input signal S11 and each of the memories stored in the delay memories
111 to 11k-1. The samples are respectively multiplied by the coefficient multipliers 120-12k-1
with the coefficients stored in the DSP. Then, the multiplication results are added by the adder 13
and the addition result S13 becomes the output of the FIR filter. The number of product-sum
operations is called the tap length of the FIR filter, which is k taps in FIG.
[0006]
09-05-2019
2
Here, since it is necessary to store a long signal in order to filter relatively low-pass components,
the filter coefficient length and the number of product-sum operations also increase. For this
reason, it is disadvantageous to the process which handles a wide frequency range like hearing.
Taking into consideration that the lower limit of the audio frequency band is 20 Hz, the tap
length k becomes 2400 (assuming that the sampling frequency fs is 48 kHz) when realizing an
FIR filter with a frequency resolution of 20 Hz. This process requires a DSP with an operation
speed of 105.2 MIPS, and it is difficult to realize this with a general-purpose one-chip DSP.
[0007]
As a conventional example for reducing the operation load in the low band, for example,
Japanese Patent Laid-Open No. 7-59186 proposes a method using a different sampling
frequency fs for each band, and the configuration is shown in FIG. However, in this method,
decimation processing (the first and third decimators in the figure) for reducing the sampling
frequency fs of the band-divided signal and interpolation processing for restoring the sampling
frequency fs of the signal whose characteristics are corrected Since (the first and third
interpolators in the figure) and filtering processing (the second and fourth LPFs in the figure) for
preventing aliasing are required, there is a problem that the efficiency is bad.
[0008]
An object of the present invention is to provide an audio processing apparatus capable of
reducing the memory capacity and the number of operations of a circuit for processing a low
band. In particular, the present invention has an object to provide an audio processing apparatus
capable of realizing band division processing and speaker characteristic correction processing by
simple hardware.
[0009]
SUMMARY OF THE INVENTION In order to achieve the above object, the present invention is to
divide the low band into bands and correct the speaker characteristics with a warp filter that has
high resolution in the low band even with a small amount of calculation. . That is, according to
the present invention, one or more warp filters respectively dividing the low frequency side of
the audio input signal into one or more frequency bands and the high frequency side of the audio
input signal into one or more frequency bands 1 There is provided a speech processing
09-05-2019
3
apparatus including the above-described FIR filter.
[0010]
Further, according to the present invention, among audio signals divided into a plurality of
frequency bands, one or more warp filters for correcting the low band side characteristics for
each band according to the speaker characteristics, and a plurality of frequency bands There is
provided a voice processing apparatus comprising: one or more FIR filters for correcting, for each
band, the characteristics of the high frequency band of the divided audio signals in accordance
with the characteristics of the speaker.
[0011]
BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be
described below with reference to the drawings.
FIG. 1 is a block diagram showing a first embodiment of a speech processing apparatus
according to the present invention.
[0012]
In FIG. 1, analog audio input signals are converted into digital signals by A / D converter 206,
and then digital signals converted by A / D converter 206 are respectively band-splitting filters
2041 to 204k and 204k + 1 to 204n. Are band-limited to individual frequency bands and divided
into n frequency bands. Then, the output signals of the band division filters 2041 to 204k and
204k + 1 to 204n are respectively applied to the speaker characteristic correction filters 2071 to
207k and 207k + 1 to 207n, and are driven for each of the band speaker units 2031 to 203k,
Frequency characteristics are added for canceling the amplitude phase characteristics of 203k +
1 to 203n, making the amplitude characteristics of the speaker units 2031 to 203k and 203k + 1
to 203n flat and making the phase characteristics linear.
[0013]
The low band side band division filters 2041 to 204k and the speaker characteristic correction
09-05-2019
4
filters 2071 to 207k are both configured by the warp filter as shown in detail in FIG. 2, and the
high band side band division filters 204k + 1 to 204n and the speaker characteristic correction
The filters 207k + 1 to 207n are all configured by FIR filters as shown in detail in FIG. Note that
the band division filter 2041 on the lowest frequency side may be a low pass filter (LPF), and the
band division filter 204 n on the highest frequency side may be a high pass filter (HPF).
[0014]
The output signals of the speaker characteristic correction filters 2071 to 207k and 207k + 1 to
207n are respectively applied to the delay correction circuits 2081 to 208k and 208k + 1 to
208n, and the band division filters 2041 to 204k, 204k + 1 to 204n and the speakers The
characteristic correction filters 2071 to 207 k and 207 k + 1 to 207 n are delayed so as to
absorb the time difference generated. Next, the respective output signals of the delay correction
circuits 2081 to 208k and 208k + 1 to 208n are converted into analog signals by the D / A
converters 2051 to 205k and 205k + 1 to 205n, respectively, and then D / A converters 2051 to
205k, The analog signals converted by 205k + 1 to 205n are respectively amplified by the
amplifiers 2021 to 202k and 202k + 1 to 202n, and applied to the speaker units 2031 to 203k
and 203k + 1 to 203n.
[0015]
Here, in general, the FIR filter has uniform frequency resolution characteristics over all
frequencies, but when processing an audio signal according to the band as in the present
invention, the frequency resolution needs to be uniform over the entire band When band limiting
is performed or after band limiting, it is only necessary to process signals within each limited
band at an optimum frequency resolution from an auditory point of view for that band. That is,
the frequency resolution in the band is fine, and in the other bands (high band), the purpose can
be sufficiently achieved even if the frequency resolution is coarse, and the warp filter satisfies
this purpose in the low band. For the working principle of the warp filter, see "Comparison of
Loudspeaker Equalization Methods Based on DSP Techniques" by Matti Karjalainen et al., J. Audio
Eng. Soc., Vol. 47, No. 1/2, (1999 Jan. / Feb.) Pp. 14I am familiar with the paper entitled -31, but
here I will briefly explain its basic operation.
[0016]
09-05-2019
5
The configuration of the warp filter will be described with reference to FIG. The input signal S15
is delayed by delay elements (D1) 151 to 151-1 formed by an all-pass filter, and the signals
delayed by the input signal S15 and delay elements 151 to 151-1 are coefficient multipliers 160
to 161-1, respectively. Each coefficient is multiplied by. Then, the multiplication results are
added by the adder 17, and the addition result S17 becomes the output of the warp filter.
[0017]
Here, while the delay element is configured to simply delay one sample in the FIR filter shown in
FIG. 3, it is characterized in that it is configured by an all pass filter in the warp filter. The all-pass
filter is a filter capable of changing the phase characteristic while maintaining the frequency
amplitude characteristic, and the configuration of a first-order all-pass filter is shown in FIG. 4 as
an example. Further, this transfer function D1 (z) can be expressed by the following equation (1).
[0018]
In the equation (1), λ is called a warping parameter, which is a parameter for determining the
resolution in the frequency band. When the frequency characteristic of the all-pass filter is
calculated from Equation (1), the amplitude characteristic is always 1, while the group delay
characteristic depends on the frequency as shown in FIG. Further, the group delay characteristic
can be changed by setting the warping coefficient λ appropriately, and by setting λ> 0, it is
possible to configure a delay element having a larger delay time as the frequency band becomes
lower. If this all-pass filter is replaced with the 1-sample delay memories 111 to 11k-1 in the FIR
filter shown in FIG. 3, a filter having fine frequency resolution in the low band and coarse
frequency resolution in the high band can be configured.
[0019]
An example of band division filters 2041 to 204k based on a warp filter will be described. FIG. 6
shows the frequency amplitude response of band-splitting filters 2041 to 204k using a 256-tap
warp filter (λ = 0.9), and FIG. 7 shows band-splitting filters 204k + 1 to 204n using a 4096-tap
FIR filter. Shows the frequency amplitude response of The cutoff frequency fkL on the low
frequency side is 100 Hz, the cutoff frequency fkH on the high frequency side is 200 Hz, and the
sampling frequency fs is 44.1 kHz. It can be seen from FIGS. 6 and 7 that almost equal frequency
amplitude responses can be obtained in the pass band.
09-05-2019
6
[0020]
Further, FIG. 8 shows the time response of the band division filters 2041 to 204k using the same
256-tap warp filter, and FIG. 9 shows the time response of the band division filters 204k + 1 to
204n using the same 4096-tap FIR filter. Show. As shown in FIG. 8 and FIG. 9, since these two
time responses are almost identical, it is understood that not only the frequency amplitude
response but also the phase characteristic can be controlled in the warp filter.
[0021]
Next, an example of the speaker characteristic correction filters 2071 to 207k using the warp
filter will be described. FIG. 10 shows the frequency amplitude response characteristics of the
speaker units 2031 to 203 n. In order to correct this characteristic, the speaker characteristic
correction filters 2071 to 207 k using the same 256 tap warp filter and the same 4096 tap FIR
filter are used The frequency amplitude response characteristics of the speaker characteristic
correction filters 207k + 1 to 207n are shown in FIGS. 11 and 12, respectively. The frequency
amplitude response characteristics corrected by the respective filters are shown in FIGS. 13 and
14, respectively. From the figure, it is understood that although the correction in the high band is
better in the FIR filter, in the correction in the low band, the correction accuracy is almost the
same as that of the FIR filter even if the warp filter is used. Therefore, by using the warp filter,
the band division filters 2041 to 204k and the speaker characteristic correction filters 2071 to
207k in the low band can be realized with the number of taps smaller than the FIR filter.
[0022]
Next, the comparison result of the calculation cost of the filter and the used memory amount in
this embodiment and the conventional example is shown in FIG. Although the number of
operation instructions per tap is four times that of the FIR filter, the warp filter can reduce the
tap length more than the increase in the number of instructions in the low band, so the total
number of instructions can be reduced. Can. In addition, since the tap length can be reduced,
memory can also be saved.
[0023]
09-05-2019
7
Here, to which band the low band to the high band should be configured by the warp filter will
be described by taking an example of band division with an octave band width. When fs = 48 kHz
and up to 24 kHz are divided into 10 bands and the lowest band (band = 1) processing is realized
by an 8192-tap FIR filter, the frequency resolution in this band is about 5.9 Hz. The ratio of the
upper limit frequency f1H in band 1 to the frequency resolution f1R in this band is f1H / f1R =
about 46.9 / about 5.9 = 8. The number of taps of the FIR filter required in the processing of
each band is shown in FIG. 16 under the condition that the ratio is kept at 8 in consideration of
the nature that the auditory sensitivity is logarithmic. As shown in FIG. 16, the required number
of taps of the FIR filter can be shortened as the frequency becomes higher.
[0024]
On the other hand, in the case of the warp filter, since the warping coefficient λ is selected
according to each band, the number of taps required in any band is substantially constant. As
described above, it is possible to replace the 4096-tap FIR filter with a 256-tap warp filter, and
the implementation requires 1024 instructions per fs. Therefore, whether to use the FIR filter or
the warp filter (FIG. 1 The critical frequency for determining k, k + 1) shown in is 375 Hz.
Therefore, the vicinity of this frequency may be used as a criterion for determining whether to
use an FIR filter or a warp filter.
[0025]
The correction of speaker characteristics by using a warp filter is described in detail in
Comparison of Loudspeaker Equalization Methods Based on DSP Techniques by Matti
Karjalainen et al. The compensation scheme is described. On the other hand, the present
invention is largely different in that the object using the warp filter is limited to the low band,
focusing on the property that the warp filter has high resolution even with a small amount of
operation in the low band.
[0026]
Next, a second embodiment will be described with reference to FIG. FIG. 17 is basically the same
as the first embodiment (FIG. 1), but among the plurality of speaker units 2031 to 203 n, the
09-05-2019
8
speaker characteristic correction filter 2071 to 207 m is only for the band 1 to m where the
characteristic correction is necessary. Is provided. That is, after processing up to the band
division in the same manner as in the first embodiment, the characteristic correction is
performed only on the band = 1 to m to which the speaker units 2031 to 203m requiring the
characteristic correction belong, and the other band = m + 1 to n Are passed as they are and
applied to the delay correction circuits 2081 to 208 n of the next stage. According to the second
embodiment, since the speaker characteristic correction filters 2071 to 207 m are provided only
in the band = 1 to m to which the speaker units 2031 to 203 m requiring the characteristic
correction belong, calculation load can be further reduced.
[0027]
As described above, according to the present invention, the low frequency band is divided and
the speaker characteristic is corrected by the warp filter with high resolution even with a small
amount of operation, so the memory of the circuit which processes the low frequency band The
capacity and the number of operations can be reduced.
[0028]
Brief description of the drawings
[0029]
1 is a block diagram showing a first embodiment of the speech processing apparatus according
to the present invention.
[0030]
2 is a block diagram showing the configuration of the warp filter of FIG.
[0031]
3 is a block diagram showing the configuration of the FIR filter of FIG.
[0032]
4 is a block diagram showing a configuration of an all pass filter as a delay element of FIG.
[0033]
5 is a graph showing the group delay characteristics of the all-pass filter of FIG.
09-05-2019
9
[0034]
6 is a graph showing the frequency amplitude response characteristics of the warp filter that
constitutes the lower band division filter of FIG.
[0035]
7 is a graph showing the frequency amplitude response characteristics when the low-pass band
division filter of FIG. 1 is configured by an FIR filter.
[0036]
8 is a graph showing the time response of the warp filter constituting the low band separation
filter of FIG.
[0037]
9 is a graph showing a time response when the low band separation filter of FIG. 1 is configured
by an FIR filter.
[0038]
10 is a graph showing the frequency amplitude response characteristics of the speaker unit.
[0039]
11 is a graph showing the frequency amplitude response characteristics when the speaker
characteristic correction filter is configured by a warp filter.
[0040]
12 is a graph showing the frequency amplitude response characteristics when the speaker
characteristic correction filter is configured by an FIR filter.
[0041]
13 is a graph showing a corrected frequency amplitude response characteristic when the speaker
characteristic correction filter is configured by a warp filter.
09-05-2019
10
[0042]
14 is a graph showing the corrected frequency amplitude response characteristics when the
speaker characteristic correction filter is configured by an FIR filter.
[0043]
FIG. 16 is an explanatory view showing comparison results of calculation cost of a filter and
amount of used memory in the embodiment in FIG. 15 and the conventional example.
[0044]
16 is an explanatory view showing the number of taps of the FIR filter required in each band.
[0045]
<Figure 17> It is the block diagram which shows the speech processing unit of 2nd execution
form.
[0046]
FIG. 18 is a block diagram showing a conventional speech processing apparatus.
[0047]
FIG. 19 is a block diagram showing another conventional speech processing apparatus.
[0048]
Explanation of sign
[0049]
2041 to 204k Band Division Filter (Warp Filter) 204k + 1 to 204n Band Division Filter (FIR
Filter) 2071 to 207k Speaker Characteristic Correction Filter (Warp Filter) 207k + 1 to 207n
Speaker Characteristic Correction Filter (FIR Filter)
09-05-2019
11