Field of the Invention
The present invention relates to a tone control apparatus
provided with a digital filter.

Background of the Invention
There is a method of using a digital filter as a method
of adjusting the tone quality of a sound signal. Either of an FIR (Finite Impulse
Response) filter and an IIR (Infinite Impulse Response) filter is used as such a
digital filter which constitutes a tone control filter.

It is necessary to design a desired filter response using
a certain method when designing a tone control filter. For example, there are methods
of representing the desired filter response as a Fourier spectrum, and computing
a filter factor based on the Fourier spectrum. According to an example of such related
art methods, an impulse response is obtained by carrying out an inverse Fourier
transform of the Fourier spectrum, and the impulse response is then set as the filter
factor of the tone control filter. According to this method, the filter factor can
be easily calculated. As a filter provided with a filter factor which can be obtained
in this way, there is an FIR filter, for example (refer to, for example, patent
reference 1).

However, an FIR filter generally has a large amount of
arithmetic operation and a large delay time, as compared with an IIR filter. On
the other hand, an IIR filter can be formed as a filter with a relatively small
size. While a typical IIR filter has a low degree of flexibility, it cannot necessarily
have a desired filter characteristic. By the way, when designing an IIR filter based
on a Fourier spectrum, the Fourier spectrum is converted into a cepstrum and a filter
factor is determined from the cepstrum according to either of the following two
types of methods.

One of them is a method of further converting the cepstrum,
into which the Fourier spectrum has been converted, into a linear prediction coefficient
(referred to as "LPC" from here on), and using the coefficient as the filter factor
of the IIR filter (refer to, for example, patent reference 2). The other one is
a filter which uses the cepstrum into which the Fourier spectrum has been converted
as a coefficient, just as it is. This filter is known as a log magnitude approximation
(referred to as "LMA" from here on) filter (refer to, for example, "Log magnitude
approximation (LMA) filter" written by Satoshi Imai, Institute of Electronics and
Communication Engineers of Japan paper magazine, vol. J63-A, No. 12, pp. 886 to
893, 1980).

[Patent reference 1] JP,2-205106,A (Fig. 1)

[Patent reference 2] JP,7-36484,A (pp. 58 to 65 and Fig. 3)

The LMA filter described in the above-mentioned paper has
a high degree of flexibility in filter design, and a wide range of realizable filter
responses. Furthermore, the LMA filter has a relatively small amount of arithmetic
operation required for calculation of a filter factor thereof and filtering processing.

However, in a case of an IIR filter or an LMA filter having
a filter factor which is an LPC coefficient to which a cepstrum as mentioned above
is converted, a ripple which is contrary to a desired filter response may occur
in the response characteristic of the filter which is actually formed. Particularly,
when a steep characteristic change point, such as a sharp peak or dip, exists in
the desired filter response, such a ripple appears notably and becomes a factor
to reduce the accuracy of the tone control apparatus.

While setting of a long length of quefrency for the cepstrum
can suppress such a ripple, the tone control filter increases in its hardware scale.
On the other hand, when there is a limit on the hardware scale or the amount of
arithmetic operation of the filter, it is difficult to increase the length of quefrency
for the cepstrum.

The present invention is made in order to solve the above-mentioned
problems, and it is therefore an object of the present invention to provide a tone
control apparatus provided with a means for suppressing ripples with a fixed length
of quefrency for the cepstrum.

Disclosure of the Invention
In accordance with the present invention, there is provided
a tone control apparatus which constitutes a tone control filter having a desired
characteristic by inputting a filter factor into said tone control filter, and which
adjusts a sound signal by making the sound signal pass through said tone control
filter, said tone control apparatus including a smoothing means for smoothing a
Fourier spectrum of a desired filter response; a cepstrum calculating means for
calculating a cepstrum from the Fourier spectrum smoothed by said smoothing means;
and a filter factor calculating means for calculating said filter factor based on
the cepstrum calculated by said cepstrum calculating means.

As a result, the tone control apparatus can suppress occurrence
of ripples in the tone control filter and can provide high-precision tone control.

Brief Description of the Figures

- Fig. 1 is a block diagram showing the structure of a tone control apparatus
in accordance with embodiment 1 of the present invention;
- Fig. 2 is a diagram for explaining a spectrum smoothing method which the tone
control apparatus in accordance with embodiment 1 of the present invention uses;
- Fig. 3 is a diagram for explaining an advantage provided by a smoothing method
of smoothing a Fourier spectrum which a tone control apparatus in accordance with
the present invention uses;
- Fig. 4 is a diagram for explaining a spectrum smoothing method which the tone
control apparatus in accordance with embodiment 2 of the present invention uses;
- Fig. 5 is a diagram for explaining a spectrum smoothing method which the tone
control apparatus in accordance with embodiment 3 of the present invention uses;
- Fig. 6 is a flowchart for explaining the smoothing method spectrum smoothing
method which the tone control apparatus in accordance with embodiment 3 of the present
invention uses;
- Fig. 7 is a block diagram showing the structure of a tone control apparatus
in accordance with embodiment 4 of the present invention;
- Fig. 8 is a block diagram showing the structure of a tone control apparatus
in accordance with embodiment 5 of the present invention; and
- Fig. 9 is the block diagram for explaining an analysis method of analyzing an
input sound signal to obtain a spectral envelope, which the tone control apparatus
in accordance with embodiment 5 of the present invention uses.

Preferred Embodiments of the Invention
Hereafter, in order to explain this invention in greater
detail, the preferred embodiments of the present invention will be described with
reference to the accompanying drawings.

Embodiment 1.
A tone control apparatus in accordance with embodiment
1 of the present invention will be explained hereafter with reference to Figs. 1
to 3. Fig. 1 is a block diagram showing the structure of the tone control apparatus
in accordance with embodiment 1, Fig. 2 is a diagram for explaining a smoothing
method of smoothing a Fourier spectrum, which this tone control apparatus uses,
and Fig. 3 is a diagram for explaining an advantage provided by the smoothing method
of smoothing a Fourier spectrum which the tone control apparatus uses.

This embodiment 1 is an example in which a desired filter
response to the tone control apparatus is predetermined, and a filter factor of
the tone control filter is calculated for the desired filter response. Assume that
an LMA filter is used as the tone control filter in this embodiment.

As shown in Fig. 1, the tone control apparatus in accordance
with embodiment 1 is provided with a desired filter response input unit 11 for inputting
the desired filter response into the tone control apparatus, a spectrum calculating
unit 12 for calculating a Fourier spectrum for the set filter response which is
inputted by the desired filter response input unit 11, a spectrum smoothing processing
unit 13 for performing smoothing processing on the Fourier spectrum calculated by
the spectrum calculating unit 12, a cepstrum calculating unit 14 for converting
the Fourier spectrum smoothed by spectrum smoothing processing unit 13 into a cepstrum,
and an LMA filter 15 to which the cepstrum calculated by the cepstrum calculating
unit 14 is set as the filter factor of the LMA filter. The tone control apparatus
also has an input terminal 16 for inputting a sound signal into the LMA filter 15,
and an output terminal 17 for outputting a sound signal from the LMA filter.

Next, the operation of the tone control apparatus will
be explained.

The desired filter response is inputted into the desired
filter response input unit 11. The filter response inputted into the desired filter
response input unit 11 can take the form of the impulse response or a transfer function
of a filter, or a gain function with frequency on the horizontal axis, for example.
The spectrum calculating unit 12 converts the filter response inputted into the
desired filter response input unit 11 into a Fourier spectrum given by the following
equation (1):
$$H\left({e}^{j2\mathrm{\&pgr;}k/N}\right)\left(k=-N/2,\cdots ,0,\cdots ,N/2\right)$$

The length N of the Fourier spectrum is equal to the length
of quefrency of a cepstrum which will be explained below.

The calculated Fourier spectrum is smoothed by the spectrum
smoothing processing unit 13. The smoothing of the Fourier spectrum can be done
by filtering a series of sampled values of the Fourier spectrum using a median filter
with a low pass characteristic. The smoothing using this method is given by the
following equation (2):
$${S}^{\prime}\left(n\right)={\displaystyle \sum _{k=-N/2}^{N/2}S\left(k\right)W\left(n-k\right)}$$

In the above-mentioned equation (2), S(k) is the series
of sampled values of the Fourier spectrum [equation (1)] which has not been smoothed
yet, and W(n) is a coefficient sequence of the median filter which is used for the
smoothing. Furthermore, S'(n) is the series of sampled values of the Fourier spectrum
which has been smoothed. Fig. 2 shows this smoothing method, Fig. 2(a) shows yet-to-be-smoothed
log magnitude values, and Fig. 2(b) shows smoothed log magnitude values.

The reason why a ripple may occur in the filter response
of the tone control filter is that the tone control filter cannot follow a steep
change in the Fourier spectrum of the desired filter response due to discontinuation
of the cepstrum. The tone control apparatus can suppress the reverse effect by filtering
the series of sampled values of the spectrum using the low pass filter. The tone
control apparatus can perform this smoothing on either a power spectrum or a log
square magnitude spectrum alternatively, instead of the simple Fourier spectrum.

Fig. 3 (a) shows the log square magnitude spectrum of the
filter response of the tone control filter when the desired filter response has
a shelving characteristic in which the gain of the filter increases rapidly at a
certain frequency and the LMA filter is formed so as not to smooth the Fourier spectrum.
Fig. 3(b) shows the log square magnitude spectrum of the response of the tone control
filter at the time of smoothing the Fourier spectrum using the smoothing method
shown by the equation (2).

The smoothed Fourier spectrum is then converted into a
cepstrum according to the following equation (3) by the cepstrum calculating unit
14:
$$\begin{array}{c}c\left(m\right)=\frac{1}{N}{\displaystyle \sum _{k=-N/2}^{N/2}\mathrm{ln}{S}^{2}\left(k\right){e}^{j2mk\mathrm{\&pgr;}/N}}\end{array}$$
*m* = 0,1,···,*N*

In the above-mentioned equation (3), c (m) is the cepstrum
and m is the quefrency.

The cepstrum calculated by the cepstrum calculating unit
14 is used as the filter factor in the LMA filter 15. The transfer function of the
LMA filter is defined by the following equation (4):
$$F\left(z\right)=\mathrm{exp}\left(\frac{c\left(0\right)}{2}+2{\displaystyle \sum _{m-1}^{M}c\left(m\right){z}^{-m}}\right)$$

It is impossible to implement a filter circuit with the
transfer function of exponential function type as shown in the equation (3). Therefore,
the transfer function is developed into a form with which a filter circuit can be
implemented according to a predetermined method of approximating an exponential
function with a rational polynomial using modified Pade approximation. The predetermined
method is described by, for example, "Speech signal processing" written by Satoshi
Imai, Morikita Shuppan Co., Ltd., pp. 144 to 148, 1996.

Assume that the above-mentioned processing is carried out
before a sound signal is input into the input terminal 16. The above-mentioned cepstrum
can be calculated for each of two or more desired filter responses, and any one
of the cepstrums thus calculated can be selected at the time when the tone control
apparatus is used. The sound signal is inputted into the LMA filter 15 to which
the cepstrum calculated by the cepstrum calculating unit 14 is set as the filter
factor of the LMA filter via the input terminal 16, and the sound signal filtered
by the LMA filter 15 is outputted via the output terminal 17.

Therefore, the tone control apparatus in accordance with
this embodiment 1 can suppress ripples of the tone control filter and can make the
tone control filter be faithful to the desired filter response by performing smoothing
processing on a Fourier spectrum of the desired filter response, thereby providing
high-precision tone control.

Since the tone control apparatus in accordance with this
embodiment 1 uses a median filter with a low pass characteristic for the smoothing
processing, it can perform the smoothing processing with a small amount of arithmetic
operation.

Since the means for calculating the filter factor and the
tone control filter can be independently formed by calculating the filter factor
in advance before the sound signal is inputted to the tone control filter, the size
of the tone control apparatus can be reduced.

In addition, since the tone control apparatus uses the
LMA filter as the tone control filter, the tone control apparatus can be scaled
down in size and can have a high degree of flexibility in design.

Embodiment 2.
A tone control apparatus in accordance with embodiment
2 of the present invention will be explained hereafter with reference to Fig. 4.
Fig. 4 is a diagram for explaining a smoothing method of smoothing a spectrum which
the tone control apparatus in accordance with embodiment 2 uses.

Instead of the median filter with a low pass characteristic,
which is disposed as the spectrum smoothing method in accordance with above-mentioned
embodiment 1, the tone control apparatus in accordance with this embodiment 2 uses
a method of replacing each odd or even-numbered sampled value of spectrum data as
shown in Fig. 4 with an average of two sampled values adjacent to each odd or even-numbered
sampled value. The other components and these components' workings of the tone control
apparatus in accordance with embodiment 2 are the same as those explained in embodiment
1, and therefore the explanation of the other components and these components' workings
will be omitted hereafter.

The smoothing method which the tone control apparatus in
accordance with this embodiment 2 uses is represented by the following equation
(5):
$$\begin{array}{c}{S}^{\prime}\left(2k+1\right)=\frac{1}{2}\left({S}^{\prime}\left(2k\right)+S\left(2k+2\right)\right)\}\\ {S}^{\prime}\left(2k\right)=S\left(2k\right)\end{array}$$

In this equation (5), S(k) is an amplitude component of
the Fourier spectrum which has not been smoothed yet, and S' (k) is an amplitude
component of the Fourier spectrum which has been smoothed. As an alternative, S(k)
can be either a power spectrum or a log square magnitude spectrum. The smoothed
Fourier spectrum is inputted into a cepstrum calculating unit 14 of Fig. 1. Fig.
4(a) shows the log magnitude of the yet-to-be-smoothed Fourier spectrum, and Fig.
4(b) shows the logarithmic magnitude of the smoothed Fourier spectrum in which each
odd or even-numbered sampled value of the spectrum data is replaced by an average
of two sampled values adjacent to each odd or even-numbered sampled value.

Since the tone control apparatus in accordance with this
embodiment 2 can smooth the Fourier spectrum of the desired filter response while
maintaining the gain of the filter at a peak and a dip of the desired filter response,
the tone control apparatus can provide a filter response in which ripples are suppressed
without spoiling the peak and dip of the desired filter response, thereby enhancing
the effects of the tone control apparatus.

Embodiment 3.
A tone control apparatus in accordance with embodiment
3 of the present invention will be explained hereafter with reference to Figs. 5
and 6. Fig. 5 is a diagram for explaining a spectrum smoothing method which the
tone control apparatus in accordance with embodiment 3 of the present invention
uses, and Fig. 6 is a flowchart for explaining the spectrum smoothing method which
the tone control apparatus in accordance with embodiment 3 of the present invention
uses.

Although the tone control apparatus in accordance with
either of above-mentioned embodiments 1 and 2 performs the smoothing processing
on a Fourier spectrum of a desired filter response over an entire frequency range,
a ripple which occurs in the tone control filter substantially results from a steep
change in the filter characteristic of the tone control filter, such as a peak and
a dip which appears in the Fourier spectrum. Therefore, the tone control apparatus
in accordance with this embodiment 3 does not smooth the Fourier spectrum of the
desired filter response over the entire frequency range which is supported by the
tone control filter, but smoothes only a frequency range of the Fourier spectrum
which includes a steep change causing ripples. The other components and these components'
workings of the tone control apparatus in accordance with this embodiment are the
same as those explained in embodiment 1, and therefore the explanation of the other
components and these components' workings will be omitted hereafter.

As shown in Fig. 5, in order to implement the spectrum
smoothing method, the tone control apparatus in accordance with this embodiment
includes a characteristic change point detecting unit 22 for detecting a change
point where the filter characteristic of the tone control filter change steeply
from the Fourier spectrum inputted via a spectrum input terminal 21, and a spectrum
smoothing processing unit 23 for smoothing the Fourier spectrum inputted via the
spectrum input terminal 21 based on the change point detected by the characteristic
change point detecting unit 22. The tone control apparatus in accordance with this
embodiment outputs the smoothed Fourier spectrum via a spectrum output terminal
24. The smoothed Fourier spectrum is then inputted into a cepstrum calculating unit
14 shown in Fig. 1.

Next, a method of detecting a change point where the filter
characteristic of the tone control filter change steeply from the Fourier spectrum
will be explained with reference to the flowchart of Fig. 6.

First, k = 1 is set for the inputted Fourier spectrum S(n),
where n= 0, ..., N (in step ST101), and the difference D(k) between S(k), where
n=k, and S(k+1) adjacent to S(k) is calculated according to equation (6) (in step
ST102). As an alternative, S(n) can be either a power spectrum or a log square magnitude
spectrum.
$$D\left(k\right)=S\left(k+1\right)-S\left(k\right)$$

The characteristic change point detecting unit then determines
whether D(k) > &egr;_{1}, where &egr;_{1} is a threshold
provided beforehand (in step ST103). When, in step ST103, determining that D(k)
is larger than the threshold (i.e., if YES), the characteristic change point detecting
unit determines whether or not a condition defined by equation (7) is satisfied,
where E[S (n)] is an average value of summation of S(n), n= 0, ..., N (in step ST104).
$$\left|S\left(k\right)-E\left[S\left(n\right)\right]\right|>\left|S\left(k+1\right)-E\left[\mathit{S}\left(n\right)\right]\right|$$

When the equation (7) is satisfied (i.e., if YES), the
characteristic change point detecting unit outputs a current frequency point k as
a characteristic change point (in step ST105). On the other hand, when the equation
(7) is not satisfied (i.e., if NO), the characteristic change point detecting unit
detects a frequency point k+1 as a characteristic, change point (in step ST106),
increments k by 1 (in step ST107), and then determines whether or not k is equal
to N (in step ST108). When, in step ST108, determining that k = N (i.e., if YES),
the characteristic change point detecting unit ends the detection of a change point
where the filter characteristic of the tone control filter change steeply from the
Fourier spectrum. On the other hand, when, in step ST108, determining that k is
not equal to N (i.e., if NO), the characteristic change point detecting unit returns
to step ST102 in which it repeats the above-mentioned processing.

When, in step ST103, determining that D(k) does not exceed
the threshold (i.e., if NO), the characteristic change point detecting unit advances
to step ST108 in which it further determines whether or not k is equal to N. When,
in step ST108, determining that k = N (i.e., if YES), the characteristic change
point detecting unit ends the detection of a change point where the filter characteristic
of the tone control filter change steeply from the Fourier spectrum. On the other
hand, when determining that k is not equal to N (i.e., if NO), the characteristic
change point detecting unit returns to step ST102 in which it repeats the above-mentioned
processing.

The spectrum smoothing processing unit 23 then performs
the smoothing processing, which is explained in embodiment 1 or 2, only on a fixed
frequency range of the Fourier spectrum with a center frequency being equal to the
characteristic change point detected as explained above.

As previously mentioned, since the tone control apparatus
in accordance with this embodiment 3 detects a steep change point of the spectrum
which causes a ripple, and then performs the smoothing processing only on a certain
frequency range of the Fourier spectrum with a center frequency being equal to the
change point detected, the tone control apparatus can reduce the amount of change
in the Fourier spectrum caused by the smoothing to a minimum, and can make the tone
control filter be faithful to the desired filter response, thereby providing high-precision
tone control.

Embodiment 4.
A tone control apparatus in accordance with embodiment
4 of the present invention will be explained hereafter with reference to Fig. 7.
Fig. 7 is a block diagram showing the structure of the tone control apparatus in
accordance with embodiment 4.

In accordance with above-mentioned embodiment 1, the filter
factor is calculated before the sound signal is inputted into the tone control apparatus,
and this filter factor is fixedly used by the tone control filter. In contrast,
the tone control apparatus in accordance with this embodiment 4 accepts input of
a desired filter response and updates the filter factor of the tone control filter
sequentially according to the change of the desired filter response while a sound
signal is inputted thereinto. The other components and these components' workings
of the tone control apparatus in accordance with this embodiment are the same as
those explained in embodiment 1, and therefore the explanation of the other components
and these components' workings will be omitted hereafter.

As shown in Fig. 7, the tone control apparatus in accordance
with embodiment 4 is provided with a desired filter response input unit 31 for inputting
a desired filter response into the tone control apparatus, a spectrum calculating
unit 32 for calculating a Fourier spectrum of the desired filter response inputted
by the desired filter response input unit 31, a spectrum smoothing processing unit
33 for performing smoothing processing on the Fourier spectrum calculated by the
spectrum calculating unit 32, a cepstrum calculating unit 34 for converting the
Fourier spectrum smoothed by the spectrum smoothing processing unit 33 into a cepstrum,
a cepstrum smoothing processing unit 35 for asymptotically updating another cepstrum
which is to be provided to the tone control filter based on the cepstrum calculated
by the cepstrum calculating unit 34, and an LMA filter 36 to which the cepstrum
updated by the cepstrum smoothing processing unit 35 is set as the filter factor
of the LMA filter. The tone control apparatus in accordance with embodiment 4 also
has an input terminal 37 for inputting a sound signal into the LMA filter 36, and
an output terminal 38 for outputting the sound signal.

Next, the operation of the tone control apparatus in accordance
with this embodiment of the present invention will be explained.

A desired filter response of the tone control filter is
inputted by the desired filter response input unit 31. The spectrum calculating
unit 32 calculates a spectrum of the response inputted by the desired filter response
input unit 31. The spectrum calculated can be a Fourier spectrum, or can be alternatively
a power spectrum or a log square magnitude spectrum. The spectrum smoothing processing
unit 33 smoothes the Fourier spectrum calculated by the spectrum calculating unit
32 using the spectrum smoothing method explained in either of above-mentioned embodiments
1 to 3. The smoothed spectrum is then converted into a cepstrum by the cepstrum
calculating unit 34. The cepstrum calculated by the cepstrum calculating unit 34
is inputted into the cepstrum smoothing processing unit 35, and the cepstrum smoothing
processing unit 35 updates the filter factor of the LMA filter 36 asymptotically
according to the following equation (8):
$$C\left(m,t\right)=a\times C\left(m,t-1\right)+\left(1-a\right)\times c\left(m,t\right)$$

where, 0<*a*<1

In the equation (8), m is quefrency, t is a time, c(m,t)
is the cepstrum calculated by the cepstrum calculating unit 34, and C (m, t) is
the filter factor which is to be provided to the LMA filter 36.

Since the tone control apparatus in accordance with this
embodiment 4 allows input of a desired filter response thereinto while the sound
signal is inputted thereinto, users can arbitrarily change and adjust the tone quality
of the sound signal while listening to the sound.

Since the tone control apparatus in accordance with this
embodiment 4 is provided with the means for smoothing a spectrum of the desired
filter response, the tone control apparatus can always suppress ripples of the tone
control filter and can provide stable tone control even if the desired fil ter response
varies from moment to moment.

Since the tone control apparatus in accordance with this
embodiment 4 is further provided with the means for updating the filter factor asymptotically,
the tone control apparatus can change the tone quality of the sound signal smoothly
without causing the user who is listening the output sound to get a feeling that
something is wrong even if the desired filter response inputted to the tone control
apparatus is arbitrarily changed.

Embodiment 5.
A tone control apparatus in accordance with embodiment
5 of the present invention will be explained hereafter with reference to Figs. 8
and 9. Fig. 8 is a block diagram showing the structure of a tone control apparatus
in accordance with embodiment 5, and Fig. 9 is a block diagram for explaining an
analysis method of analyzing an input sound signal to obtain a spectral envelope,
which the tone control apparatus in accordance with this embodiment uses.

The tone control apparatus in accordance with above-mentioned
embodiment 4 accepts a desired filter response from outside the tone control apparatus.
In contrast, the tone control apparatus in accordance with this embodiment 5 analyzes
a sound signal inputted thereinto so as to determine a desired filter response based
on the analysis result depending on a given purpose. For example, the tone control
apparatus can be so constructed as to calculate a spectrum of the sound signal inputted
thereinto, analyze the spectrum of the sound signal so as to obtain a spectral envelope,
and determine a filter response so as to emphasize this spectral envelope.

The tone control apparatus in accordance with embodiment
5 is an example of emphasizing the spectral envelope of the inputted sound signal,
and, as shown in Fig. 8, is provided with a desired filter response input unit 41
for inputting a desired filter response into the tone control apparatus, a spectral
envelope analyzing unit 42 for analyzing the inputted sound signal so as to obtain
a spectral envelope, a spectrum calculating unit 43 for calculating a Fourier spectrum
of a filter response of a tone control filter based on the desired filter response
inputted by the desired filter response input unit 41, and the analysis result obtained
by the spectral envelope analyzing unit 42, a spectrum smoothing processing unit
44 for performing smoothing processing on the Fourier spectrum calculated by the
spectrum calculating unit 43, a cepstrum calculating unit 45 for converting the
Fourier spectrum smoothed by the spectrum smoothing processing unit 44 into a cepstrum,
a cepstrum smoothing processing unit 46 for updating a cepstrum which is to be provided
to the tone control filter asymptotically based on the cepstrum calculated by the
cepstrum calculating unit 45, and an LMA filter 47 to which the cepstrum updated
by the cepstrum smoothing processing unit 46 is set as the filter factor of the
LMA filter. The tone control apparatus in accordance with embodiment 5 also has
an input terminal 48 for inputting a sound signal into the LMA filter 47, and an
output terminal 49 for outputting the sound signal.

Next, the operation of the tone control apparatus in accordance
with this embodiment of the present invention will be explained.

A desired filter response of the tone control filter is
inputted into the desired filter response input unit 41. The spectral envelope analyzing
unit 42 analyzes a sound signal inputted into the input terminal 48 so as to obtain
a spectral envelope. The spectrum calculating unit 43 calculates a filter response
of the tone control filter based on the desired filter response inputted by the
desired filter response input unit 41 and the spectral envelope obtained by the
spectral envelope analyzing unit 42.

The filter response of the tone control filter is calculated
according to equation (9). Equation (10) in the equation (9) shows the Fourier spectrum
of the desired filter response inputted into the desired filter response input unit
41, equation (11) shows the spectral envelope obtained by the spectral envelope
analyzing unit 42, and equation (12) shows the Fourier spectrum outputted from the
spectrum calculating unit 43. &agr; is a constant and is a parameter for controlling
the degree of enhancement of the spectral envelope. The spectrum calculated can
be alternatively a power spectrum or a log square magnitude spectrum.
$${H}_{\mathit{filter}}\left({e}^{j2\mathrm{\&pgr;}k/N}\right)=\mathrm{\&agr;}\times {H}_{\mathit{in}}\left({e}^{j2\mathrm{\&pgr;}k/N}\right)\times {H}_{\mathit{env}}\left({e}^{j2\mathrm{\&pgr;}k/N}\right)$$
$${H}_{\mathit{in}}\left({e}^{j2\mathrm{\&pgr;}k/N}\right)$$
$${H}_{\mathit{env}}\left({e}^{j2\mathrm{\&pgr;}k/N}\right)$$
$${H}_{\mathit{filter}}\left({e}^{j2\mathrm{\&pgr;}k/N}\right)$$

The spectrum smoothing processing unit 44 smoothes the
Fourier spectrum calculated by the spectrum calculating unit 43 using the spectrum
smoothing method explained in either of above-mentioned embodiments 1 to 3. The
smoothed spectrum is then converted into a cepstrum by the cepstrum calculating
unit 45. The cepstrum calculated by the cepstrum calculating unit 45 is inputted
into the cepstrum smoothing processing unit 46, and the cepstrum smoothing processing
unit 46 updates the filter factor of the LMA filter 47 asymptotically according
to the method explained in above-mentioned embodiment 4.

Next, the analysis processing performed by the spectral
envelope analyzing unit 42 will be explained. The sound signal inputted, via the
input terminal 48, into the spectral envelope analyzing unit 42 of Fig. 8 in units
of frames is converted into an FFT cepstrum by an FFT cepstrum calculating means
52. The FFT cepstrum is calculated according to the following equation (13). The
input terminal 48 corresponds to an input terminal 51 of Fig. 9.
$${c}_{\mathit{fft}}\left(n\right)=\frac{1}{N}{\displaystyle \sum _{k=-N/2}^{N/2}\mathrm{ln}\left|X\left(k\right)\right|{e}^{j2\mathrm{\&pgr;}k/N}}$$

In this equation (13), C_{fft}(n) is the FFT cepstrum
of the incoming sound signal, and X(k) is the Fourier spectrum of the incoming sound
signal.

A liftering means 53 performs liftering on the FFT cepstrum
so that high-quefrency components of the FFT cepstrum are cancelled and low-quefrency
components of the FFT cepstrum are left. A spectral envelope calculating means 54
acquires a spectral envelope by performing FFT on the FFT cepstrum, on which the
liftering has been performed, again. The spectral envelope acquired as mentioned
above is outputted from an output terminal 55, and is then inputted into the spectrum
calculating unit 43 of Fig. 8.

Since the tone control apparatus in accordance with this
embodiment 5 has the means for analyzing the spectral envelope of the input sound
signal in units of frames so as to emphasize the spectral envelope, the tone control
apparatus can make the sound signal clear.

In addition, since the tone control apparatus is provided
with the means for smoothing the spectrum of the filter response, it can always
suppress ripples of the tone control filter even when the target filter response
varies from moment to moment dependently upon changes in the sound signal with time,
thereby providing stable tone control.

Embodiment 6.
A tone control apparatus in accordance with embodiment
6 of the present invention will be explained hereafter. The tone control apparatus
in accordance with either of above-mentioned embodiments 1 and 4 uses an LMA filter
as the tone control filter. The tone control apparatus in accordance with this embodiment
6 uses, as the tone control filter, an IIR filter having a filter factor which is
an LPC coefficient calculated from a cepstrum. The other components and these components'
workings of the tone control apparatus in accordance with this embodiment are the
same as those explained in embodiment 1, and therefore the explanation of the other
components and these components' workings will be omitted hereafter.

In this case, the filter factor is acquired according to
the following equation (14):
$$\begin{array}{c}a\left(1\right)=-c\left(1\right)\\ a\left(n\right)=-c\left(n\right)-{\displaystyle \sum _{m=1}^{n-1}\left(1-\frac{m}{n}\right)a\left(m\right)c\left(n-m\right)}\end{array}\mathrm{\hspace{1em}}(\mathrm{l}<\mathit{n}\le \mathit{p})\}$$

In the above-mentioned equation (14), a(n) is the LPC coefficient
and p is the order of the IIR filter.

A transfer function F(z) of the IIR filter which is the
tone control filter is given by the following equation (15):
$$F\left(z\right)=\frac{1}{1+{\displaystyle \sum _{k=1}^{p}a\left(k\right){z}^{-k}}}$$

Since the tone control apparatus in accordance with this
embodiment 6 uses a common IIR filter as the tone control filter, the tone control
apparatus can be scaled down in size and can have a small amount of arithmetic operation.

Industrial Applicability
As mentioned above, the tone control apparatus in accordance
with the present invention is suitable for tone control using a digital filter.