close

Вход

Забыли?

вход по аккаунту

?

JP2010212887

код для вставкиСкачать
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 JP2010212887
An object of the present invention is to provide a signal correction device capable of preventing
an excessive reduction in volume while preventing clipping of a digital signal. A coefficient
calculating unit 102A obtains a coefficient p according to an input signal x [n], and according to
the coefficient p, a coefficient for controlling the level of a signal in a digital level correcting unit
101A, and a characteristic correcting unit By the balance setting unit 102B determining the
coefficient for changing the characteristic of the signal in 101B, the digital level correction unit
101A and the characteristic correction unit 101B can perform adaptive processing. [Selected
figure] Figure 1
Signal characteristic change device
[0001]
The present invention relates to a signal correction apparatus that changes the characteristics of
an input signal.
[0002]
Desired listening environment (for example, indoor transfer characteristic) by changing
characteristics (for example, volume, frequency characteristic, phase characteristic, tonality) of
an input signal (for example, audio signal, audio signal) using DSP (Digital Signal Processor) or
the like Development of signal characteristic change devices for providing a head-related transfer
function and a volume).
[0003]
08-05-2019
1
As an example, there is an equalizer device, in which the level (for example, power and signal
amplitude) can be adjusted for each frequency band.
When changing the level of the digital signal, raising the level may cause the signal to clip
beyond the maximum level of the digital signal system.
[0004]
For this reason, the equalizer device that raises the level of the digital signal performs level
control to lower the level of the input signal over the entire frequency band, and amplifies the
level by the amount reduced by processing the digital signal after performing analog conversion.
ing.
[0005]
Also, conventionally, in order to prevent the deterioration of the S / N ratio caused by the level
increase and decrease as described above, the level of the input digital signal is reduced so as not
to clip based on the maximum value of the level change amount. (See, for example, Patent
Document 1).
However, in the past, since the level was lowered by a fixed amount, an excessive reduction in
volume might occur depending on the input signal.
Japanese Patent Application Laid-Open No. 2002-345075
[0006]
In the conventional signal characteristic change apparatus, the level of the digital signal is
lowered to prevent the digital signal from being clipped, but the level is lowered by a fixed
amount, so an excessive volume reduction occurs. Was a problem. The present invention has
been made to solve the above-mentioned problems, and it is an object of the present invention to
provide a signal correction device capable of preventing an excessive decrease in volume while
08-05-2019
2
preventing clipping of digital signals.
[0007]
In order to achieve the above object, the present invention is a signal characteristic change
device for changing the characteristic of an input signal, comprising: level control means for
controlling the level of the input signal; and the input signal controlled in level by the level
control means. Coefficient control means for calculating the level control coefficient for adjusting
the level control of the level control means and the characteristic change coefficient for adjusting
the level control of the characteristic change means based on the characteristic change means for
changing the characteristics of And to be configured.
[0008]
As described above, in the present invention, when changing the characteristics of the input
signal, the level of the input signal is controlled based on the feature quantity of the input signal
in the previous stage.
Therefore, according to the present invention, since adaptive processing can be performed
according to the input signal, it is possible to provide a signal characteristic change device
capable of preventing clipping while suppressing excessive volume reduction of the digital signal.
[0009]
Hereinafter, an embodiment of the present invention will be described with reference to the
drawings. First Example FIG. 1 shows the configuration of a signal characteristic change
apparatus according to an embodiment of the present invention. The signal characteristic change
device includes a signal correction unit 101 and a characteristic change unit 102. This signal
characteristic change device is mounted on, for example, a voice communication device such as a
mobile phone, an electronic device such as a portable audio player, which converts a digital voice
signal into an analog signal and outputs it.
[0010]
08-05-2019
3
For example, in the case of a voice communication device, the communication device provided at
the front stage of this circuit establishes a communication link with the communication device of
the other party, and enables two-way communication with the other party. The received data
received by the wireless communication unit is input signal x [n] (n = 1, 2, 3,...) For each unit of
processing time (one frame = N samples) determined in advance by a decoder (not shown). , N)
are decoded into digital signals.
[0011]
The input signal x [n] may be an audio signal or an audio signal. In the following description, the
input signal x [n] is an audio signal. The range of N may be an integer of 1 or more, for example,
a frame size of N = 160 samples. In the case of an audio signal, N is mainly used to the power of
2 such as 1024, 512, 256, and so on.
[0012]
Also, although the input signal x [n] is a signal represented by 16 bits, it is not limited to this, and
even if it is a signal represented by b bits (b = 1, 2, 3. It may be a floating point signal.
Hereinafter, the same conditions apply to signals to be described later (for example, d [n] and y
[n] in FIG. 1).
[0013]
Subsequently, the signal correction unit 101 and the characteristic change unit 102 will be
described. The signal correction unit 101 receives an input signal x [n] and outputs an output
signal y [n] (n = 1, 2, 3,..., N) in which the characteristics of the signal are changed.
[0014]
The signal correction unit 101 includes a digital level correction unit 101A and a characteristic
correction unit 101B. The digital level correction unit 101A performs dynamic range control
(DRC) of the input signal x [n] using a level control coefficient TH input from the characteristic
08-05-2019
4
change unit 102 described later to calculate the signal level (eg, amplitude, power). It controls
and outputs the signal d [n]. Specifically, when the level of the input signal x [n] is larger than the
level control coefficient TH, the level of the input signal x [n] is attenuated and output according
to the level control coefficient TH.
[0015]
On the other hand, when the level of the input signal x [n] is smaller than the level control
coefficient TH, the level of the input signal x [n] is output without being changed. By the control
as described above, it is possible to suppress in advance only a signal that is likely to be clipped.
The digital level correction unit 101A may be any device that controls the signal level, and may
be an automatic volume adjustment (AGC).
[0016]
Characteristic correction unit 101B changes the frequency characteristic of signal d [n] using
characteristic change coefficient G (ω) input from characteristic change unit 102 described later.
As a result, for example, it is possible to compensate for a degraded frequency band such as a
speaker for a mobile phone handset or a speaker for a mobile phone hands-free, or to provide a
special effect such as an indoor transfer characteristic or head transfer function. A listening
environment can be realized.
[0017]
Specifically, signal d [n] is converted to a frequency domain signal and then multiplied by
characteristic change coefficient G (ω), and the result is converted to a time domain signal and
output as output signal y [n]. Do. The output signal y [n] is converted into an analog signal by a
digital analog converter (D / A converter) in a subsequent stage not shown, and is amplified and
output from the speaker through the amplifier. The characteristic correction unit 101B may
change the frequency characteristic without converting to the frequency domain.
[0018]
In the present embodiment, the signal correction unit 101 is described as controlling the signal
level, but the signal correction unit 101 is not limited to the characteristics of a signal such as an
08-05-2019
5
audio signal or an audio signal (for example, volume, frequency characteristics, phase
characteristics, tonality). The configuration is not limited to the above as long as it changes the
[0019]
Characteristic changing unit 102 receives input signal x [n] and has level control coefficient TH
that prevents clipping while suppressing excessive volume reduction of the signal for each frame,
and characteristic change coefficient G (ω) (ω = 1 , 2, 3 ...).
Here, the level control coefficient TH is a parameter used by the digital level correction unit
101A for amplifying or attenuating the level (for example, amplitude, power) of the signal to
obtain a desired volume setting. Here, when the input signal x [n] is a 16-bit audio signal, the
level control coefficient TH is in the range of 0 to 2 <16-1>.
[0020]
On the other hand, the characteristic change coefficient G (ω) is a parameter for changing the
frequency characteristic, which is used by the characteristic correction unit 101B. Although ω
indicates a frequency bin number, it may be a grouped frequency band unit. The level control
coefficient TH and the characteristic change coefficient G (ω) are output for each frame.
Therefore, the signal correction unit 101 can perform adaptive processing in accordance with the
input signal, and can prevent clipping while suppressing excessive volume reduction of the
signal.
[0021]
The characteristic change unit 102 includes a coefficient calculation unit 102A, a balance setting
unit 102B, a digital level adjustment unit 102C, and a characteristic adjustment unit 102D.
Hereinafter, each configuration will be described.
[0022]
08-05-2019
6
The coefficient calculation unit 102A receives the input signal x [n] and outputs, as a coefficient
p, a threshold of the minimum necessary level control coefficient for preventing clipping. The
coefficient p is a feature of the input signal x [n] for each frame (for example, spectrum power
absolute value, spectrum power average value, spectrum power dispersion value, spectrum
power standard deviation value, amplitude average, amplitude maximum value, zero crossing
number, Amplitude variance, amplitude standard deviation, amplitude difference dispersion
between samples, amplitude difference standard deviation between samples, etc.) and minimum
threshold of level control coefficients to prevent clipping Calculated based on the equation of
[0023]
As a result, since the coefficient p can be obtained by adaptive processing in accordance with the
input signal for each frame, it is possible to prevent a clip while suppressing an excessive volume
reduction of the signal. A specific configuration example of the coefficient calculation unit 102A
will be described in detail later.
[0024]
The balance setting unit 102B receives the coefficient p as an input and makes a condition
judgment so that sound quality deterioration (for example, distortion due to level control,
unnaturalness in volume due to automatic volume control, sense of noise, etc.) does not occur.
The temporary level control coefficient THp and the temporary characteristic change coefficient
Gp (ω) are output.
[0025]
Here, the condition determination is to compare the magnitude relation between a value
(minimum level control coefficient α) preset as the minimum value of the level control
coefficient TH and the coefficient p.
The minimum level control coefficient α is set so that the sound quality deterioration due to the
control of the signal level does not occur. By calculating the temporary level control coefficient
THp and the temporary characteristic change coefficient Gp (ω) based on the comparison result
of the minimum level control coefficient α and the coefficient p, the sound quality deterioration
due to the level control is prevented while the sound volume is excessively reduced. You can
prevent the clip while suppressing the
08-05-2019
7
[0026]
The specific operation of the balance setting unit 102B will be described. The balance setting unit
102B compares the coefficient p with the minimum level control coefficient α, and if the
coefficient p is larger than the minimum level control coefficient α (α <p), it means that no
speech degradation due to level control occurs. Therefore, the coefficient p is output as the
temporary level correction coefficient THp as it is, and the temporary characteristic change
coefficient G1 (ω) one frame before is output as the temporary characteristic change coefficient
Gp (ω).
[0027]
Here, the temporary characteristic change coefficient G1 (ω) one frame before is the temporary
characteristic change coefficient Gp (ω) calculated one frame before the present. The temporary
characteristic change coefficient G1 (ω) one frame before may be information in which the
characteristic one frame before the present is changed, and for example, the characteristic
change coefficient G (ω) calculated one frame before the current ) May be.
[0028]
On the other hand, if the coefficient p is smaller than the minimum level control coefficient α
(α> p), this means that voice deterioration occurs due to level control, so the minimum level
control coefficient α is used as the provisional level correction coefficient THp. Output. Further,
the amplification level and the attenuation level of the temporary characteristic change
coefficient G1 (ω) one frame before are suppressed as shown in equation (1), and are output as
the temporary characteristic change coefficient Gp (ω).
[0029]
[0030]
08-05-2019
8
In Equation (1), β is a parameter for weighting the suppression of the amplification level and the
attenuation level of the temporary characteristic change coefficient Gp (ω), and β ≦ 1.
If the minimum level control coefficient α is much larger than the balance control coefficient p
(α >> p), the amplification level and the attenuation level of the temporary characteristic change
coefficient Gp (ω) are greatly suppressed, and the processing does not substantially change the
characteristics It becomes.
[0031]
If the minimum level control coefficient α and the coefficient p have the same value, the
correction coefficient THp and the correction coefficient Gp (ω) may be determined under any of
the above conditions.
[0032]
FIG. 2 is a diagram illustrating an example of the provisional level control coefficient THp set by
the balance setting unit 102B.
If the coefficient p is larger than the minimum level control coefficient α (1) (α <p), the
temporary level control coefficient THp is determined based on the coefficient p. Here, the
provisional level control coefficient THp may be determined by the coefficient p, and may not be
output as the coefficient p as it is.
[0033]
On the other hand, when the coefficient p is smaller than the minimum level control coefficient
α (2) (α> p), the temporary level control coefficient THp is determined based on the minimum
level control coefficient α. Here, the provisional level control coefficient THp may be determined
by the minimum level control coefficient α, and may not be output as the minimum level control
coefficient α as it is.
[0034]
08-05-2019
9
FIG. 3 is a diagram illustrating an example of the temporary characteristic change coefficient Gp
(ω) set by the balance setting unit 102B. If the coefficient p is larger than the minimum level
control coefficient α (1) (α <p), the temporary characteristic change coefficient Gp (ω) is
determined based on the temporary characteristic change coefficient G1 (ω) one frame before .
[0035]
Here, the temporary characteristic change coefficient Gp (ω) may be determined by the
temporary characteristic change coefficient G1 (ω) one frame before, and is output as the
temporary characteristic change coefficient G1 (ω) one frame before. It does not have to be.
[0036]
On the other hand, if the coefficient p is smaller than the minimum level control coefficient α (2)
(α> p), the temporary characteristic change coefficient Gp (ω) is the amplification level of the
temporary characteristic change coefficient G1 (ω) one frame earlier and Suppress and output
the attenuation level.
Here, the temporary characteristic change coefficient Gp (ω) is not limited to the equation (1) as
long as the amplification level and the attenuation level of the temporary characteristic change
coefficient G1 (ω) one frame before are suppressed.
[0037]
The digital level adjustment unit 102C receives the temporary level control coefficient THp as an
input to suppress a transient influence between adjacent frames. Specifically, the digital level
adjustment unit 102C performs smoothing processing (smoothing processing) between the
frames adjacent to the temporary level control coefficient THp according to Equation (2), and
uses the value obtained by the smoothing processing as the level control coefficient TH. It is
output to the correction unit 101A. Specifically, for R frame and subsequent ones, level control
coefficient TH obtained by smoothing temporary level control coefficient THp according to
equation (2) to suppress transient influence between adjacent frames. Ask and output this.
[0038]
08-05-2019
10
[0039]
The digital level correction unit 101A outputs the level control coefficient TH0 from the start of
the call to the R-th frame (R ≧ 1).
TH0 is a level control coefficient TH set in advance in the digital level correction unit 101A
before the start of a call. Further, in the equation (2), THR indicates the level control coefficient
TH calculated before the R frame. However, THR should just be information which controls the
signal level before R frame, for example, suppose that it may be temporary level control
coefficient THp before R frame.
[0040]
Also, φRj and φp are smoothing coefficients, and the weighting for the smoothing process can
be changed. The digital level adjustment unit 102C only needs to suppress the transient
influence between adjacent frames, and is not limited to the above example.
[0041]
Characteristic adjustment unit 102D receives the temporary characteristic change coefficient Gp
(ω), performs smoothing processing (smoothing processing) between adjacent frames, and
outputs characteristic change coefficient G (ω) set in characteristic correction unit 101B. .
Specifically, the characteristic change obtained by smoothing the temporary characteristic
change coefficient Gp (ω) in accordance with the equation (3) in order to suppress a transient
influence between adjacent frames for the R frame and subsequent ones. The coefficient G (ω) is
obtained and output.
[0042]
[0043]
08-05-2019
11
Note that from the call start to the R-th frame (Rω1), the characteristic correction unit 101B
outputs the characteristic change coefficient G0 (ω).
G0 (ω) is a characteristic change coefficient G (ω) set in advance in the characteristic correction
unit 101B before the start of a call.
[0044]
In Equation (3), the characteristic change coefficient GR (ω) indicates the characteristic change
coefficient G (ω) calculated before R frames. However, the characteristic change coefficient GR
(ω) may be information in which the characteristics of the signal before R frame are changed,
and may be, for example, a temporary characteristic change coefficient Gp (ω) before R frame.
[0045]
Also, φRj and φp are smoothing coefficients, and the weighting for the smoothing process can
be changed. This does not have to be the same as the smoothing coefficient of the digital level
adjustment unit 102C. The characteristic adjustment unit 102D may suppress transient influence
between adjacent frames, and is not limited to this.
[0046]
Next, FIG. 4 shows a configuration example of the coefficient calculation unit 102A. The
coefficient calculation unit 102A includes a feature amount extraction unit 102A1 and a
regression operation unit 102A2. Furthermore, the feature quantity extraction unit 102A1
includes the extraction unit 102A11 to the extraction unit 102A14. Hereinafter, each
configuration will be described.
[0047]
The feature amount extraction unit 102A1 receives the input signal x [n] as the feature amount
08-05-2019
12
(for example, spectrum power absolute value, spectrum power average value, spectrum power
dispersion value, spectrum power standard deviation value, amplitude average, maximum
amplitude value, A plurality of zero cross numbers, amplitude variance, amplitude standard
deviation, amplitude difference variance between samples, amplitude difference standard
deviation between samples, etc. are extracted for each frame, and the feature quantities A (m) (m
= 1, 2, 3,. S) Output (S ≧ 1). Here, the feature amount extraction unit 102A1 is described as
extracting four feature amounts, but it may be one that extracts one or more feature amounts.
[0048]
Extracting section 102A11 takes input signal x [n] as input, converts the signal in the time
domain into a signal in the frequency domain, calculates the spectrum power for each frequency
bin, and maximizes the spectrum power among the number of frequency bins ω (Maximum
spectrum power) is calculated and output as feature quantity A (1). The extraction unit 102A11
includes a frequency domain conversion unit 102A111, a power calculation unit 102A112, and a
maximum power calculation unit 102A113.
[0049]
Frequency domain conversion section 102A 111 receives input signal x [n] as input, and
performs operation such as FFT (Fast Fourier Transform) to convert input signal x [n] from a
signal in the time domain to a signal X (ω) in the frequency domain. Convert and output.
[0050]
The frequency domain transform unit 102A 111 may perform discrete Fourier transform (DFT),
discrete cosine transform (DCT), Walsh Hadamard transform (WHT), harlet transform (HT), or
slant transform. It is also possible to substitute other orthogonal transform that transforms into a
frequency domain represented by (SLT: Slant Transform) and Karhunen-Loeve Transform (KLT).
[0051]
Specifically, if the input signal x [n] (n = 1, 2, 3,..., N) is an input signal in the time domain of n
samples, then if the order of the FFT is N, the signal in the frequency domain X (ω) (ω = 1, 2, 3...,
N) will have N frequency bin numbers.
08-05-2019
13
Note that the signal of the previous frame may be overlapped with the signal to be subjected to
FFT or zero-padded, the data length may be a power of 2, and the FFT order N may be a power of
2.
[0052]
Power calculator 102A112 receives frequency domain signal X (ω) as input, calculates spectrum
power for each frequency bin, and outputs it as spectrum power pow (ω) (ω = 1, 2, 3,..., N) .
The power calculation unit 102A112 calculates the spectral power of the input signal for each
frequency bin according to Equation (4).
[0053]
[0054]
However, the output of the power calculation unit 102A112 may be one that does not square X
(ω).
Also, as shown in the equation (5), the spectrum power may be calculated using the average
spectrum power pow (ω) using signals from the current to the number of R frames before (R is
an integer).
[0055]
[0056]
By using the average spectral power in this manner, it is possible to suppress transient effects
between adjacent frames.
08-05-2019
14
Here, Xi (ω) shown in the equation (5) is a frequency domain signal X (ω) i frames before (i = 1,
2, 3..., R) from the present. Also, as shown in equation (6), the spectral power pow (ω) may be
weighted by multiplying it with the characteristic change coefficient G0 (ω).
[0057]
[0058]
By weighting in this manner, a more ideal coefficient p can be calculated.
In addition, what is necessary is just to calculate spectrum power from frequency domain signal
X ((omega)), and it is not limited to the above structures.
[0059]
Maximum power calculation unit 102A113 receives spectral power pow (ω), detects maximum
spectral power pow_MAX among all frequency bins, and outputs it as feature quantity A (1). Also,
a value normalized by dividing the spectral power pow_MAX by the sum of all frequency bins
according to the equation (7) may be output as the feature value A (1).
[0060]
[0061]
In this equation (7), A (1) is 1 or less (A (1) ≦ 1), but if it is close to 1, it means that the spectral
power is concentrated in a specific frequency bin.
If the characteristic is changed so as to amplify a specific frequency bin, it can be determined in
advance whether or not a clip occurs in the feature quantity A (1) of this equation (7). The
maximum power calculation unit 102A 113 may calculate the spectrum power that is the
maximum among all the frequency bins, and is not limited to the above example.
08-05-2019
15
[0062]
Extraction section 102A12 takes input signal x [n] as input, converts the signal of time domain to
a signal of frequency domain, calculates spectrum power for each frequency bin, and calculates
its average value (average spectrum power) , And output it as feature quantity A (2). The
extraction unit 102A12 includes a frequency domain conversion unit 102A111, a power
calculation unit 102A112, and an average power calculation unit 102A121.
[0063]
Average power calculation unit 102A121 receives spectrum power pow (ω), detects an average
value pow_AVG (average spectrum power) of spectrum power for each frequency bin, and
outputs it as feature quantity A (2). Specifically, it is determined according to the number (8).
[0064]
[0065]
The extraction unit 102A13 receives the input signal x [n], obtains the dispersion value of the
amplitude, and outputs it as the feature value A (3).
The extracting unit 102A13 includes an amplitude variance calculating unit 102A131. Amplitude
dispersion calculation unit 102A131 receives input signal x [n], obtains the dispersion value of
the amplitude, and outputs it as feature amount A (3). Specifically, it is determined according to
equation (9). Here, mean in equation (9) is the average value of x (n).
[0066]
[0067]
08-05-2019
16
Extraction unit 102A14 receives input signal x [n], obtains the number of zero crossings, and
outputs it as feature quantity A (4).
The extraction unit 102A14 includes a zero cross number calculation unit 102A141.
[0068]
The zero cross number calculation unit 102A 141 receives the input signal x [n], counts the zero
cross number (when x [n] = 0 and when the positive and negative of x [n] are inverted), Output as
(4). The zero crossing number is used to determine voiced speech and unvoiced speech, so that it
is possible to prevent clipping while suppressing excessive volume reduction of the signal
separately for voiced speech and unvoiced speech.
[0069]
The regression operation unit 102A2 calculates the threshold value of the minimum necessary
level control coefficient for preventing clipping from the feature amount. For example, with the
feature amount A (m) (m = 1, 2, 3, 4) as an input, the coefficient p is calculated based on the
equation of the square error minimum criterion pre-regression analysis such as equation (4).
Output.
[0070]
[0071]
Here, ζ (m) (m = 1, 2, 3, 4) in the equation (10) is a regression analysis of the threshold of the
minimum level control coefficient necessary to prevent clipping and the feature amount A (m) It
is a coefficient and can be set in advance.
If the correlation between the threshold of the minimum necessary level control coefficient for
preventing clipping and the feature amount A (m) is high, a more ideal coefficient p can be
08-05-2019
17
calculated.
[0072]
As described above, in the signal characteristic change device configured as described above, the
coefficient calculation unit 102A determines the coefficient p according to the input signal x [n],
and the balance setting unit 102B determines the digital level correction unit 101A from the
coefficient p. The coefficient for controlling the level of the signal at the time T.sub.1 and the
coefficient for changing the characteristic of the signal at the characteristic correction unit 101B
are determined. Therefore, according to the signal characteristic change device of the above
configuration, the digital level correction unit 101A and the characteristic correction unit 101B
can perform adaptive processing in accordance with the coefficient p of the input signal x [n],
thereby preventing clipping of the digital signal. However, excessive volume reduction can be
prevented.
[0073]
Second Embodiment Next, a second embodiment according to the present invention will be
described. FIG. 5 shows the configuration. In the following description, the same components as
those in the first embodiment described above are denoted by the same reference numerals, and
redundant description will be omitted as necessary to make the description clear.
[0074]
The characteristic changing unit 102 according to the second embodiment receives an output
signal y [n] obtained by changing the characteristic of the signal in addition to the input signal x
[n] as an input, and suppresses an excessive decrease in volume of the signal for each frame.
However, the level control coefficient TH and the characteristic change coefficient G (ω) (ω = 1,
2, 3. The characteristic change unit 102 according to the second embodiment uses a coefficient
calculation unit 102A 'instead of the coefficient calculation unit 102A used in the characteristic
change unit 102 according to the first embodiment.
[0075]
08-05-2019
18
The coefficient calculation unit 102A ′ receives the output signal y [n] obtained by changing the
characteristic of the signal in addition to the input signal x [n], and uses the threshold of the
minimum necessary level control coefficient for preventing clipping as the coefficient p. Output.
As shown in FIG. 6, the coefficient calculation unit 102A 'includes a feature quantity extraction
unit 102A'1 and a regression operation unit 102A'2.
[0076]
The feature quantity extraction unit 102A'1 includes the extraction unit 102A'11, the extraction
unit 102A'12, the extraction unit 102A'13, and the extraction unit 102A'14, and the input signal
x In addition to [n], an output signal y [n] in which the characteristics of the signal are changed is
input, and a plurality of feature quantities are extracted for each frame, feature quantity A (m)
and feature quantity B (m) (m = 1 , 2, 3, 4).
[0077]
Here, although four feature quantities are extracted by the feature quantity extraction unit
102A'1, the feature quantities include spectrum power absolute value, spectrum power average
value, spectrum power dispersion value, spectrum power standard deviation value, and
amplitude average One or more of an amplitude maximum value, a zero cross number, an
amplitude dispersion, an amplitude standard deviation, an amplitude difference dispersion
between samples, an amplitude difference standard deviation between samples, and the like may
be calculated.
[0078]
The extraction unit 102A'11 receives the input signal x [n] and the output signal y [n] obtained
by changing the characteristics of the signal, converts the signal in the time domain into a signal
in the frequency domain, and converts the spectrum for each frequency bin The power is
calculated, and the spectrum power (maximum spectrum power) that is the largest among the
frequency bin numbers ω is output as the feature amount A (1) and the feature amount B (1).
The extraction unit 102A'11 includes a frequency domain conversion unit 102A'111, a power
calculation unit 102A'112, and a maximum power calculation unit 102A'113.
[0079]
08-05-2019
19
The frequency domain transform unit 102A '111 receives an output signal y [n] obtained by
changing the characteristic of the signal in addition to the input signal x [n] as an input and
performs an operation such as FFT (Fast Fourier Transform) to obtain the input signal x [ n] is
converted from the time domain signal to the frequency domain signal X (ω), and the output
signal y [n] is converted from the time domain signal to the frequency domain signal Y (ω).
[0080]
The power calculator 102A'112 receives the frequency domain signal Y (ω) in addition to the
frequency domain signal X (ω), calculates the spectral power for each frequency bin, and
calculates the spectral powers xpow (ω) and ypow Output as ω) (ω = 1, 2, 3..., N).
The power calculator 102A'112 calculates the spectral power of the signal for each frequency
bin according to the equations (11) and (12).
[0081]
[0082]
[0083]
Maximum power calculation unit 102A '113 receives spectral powers xpow (ω) and ypow (ω),
detects maximum spectral powers xpow_MAX and ypow_MAX among all frequency bins,
respectively, and detects them as feature amount A (1) And the feature amount B (1).
[0084]
The extraction unit 102A'12 receives the input signal x [n] and the output signal y [n] obtained
by changing the characteristics of the signal, converts the signal in the time domain into a signal
in the frequency domain, and converts the spectrum for each frequency bin The power is
calculated, the average value (average spectrum power) is calculated, and it is output as the
feature amount A (2) and the feature amount B (2).
08-05-2019
20
The extraction unit 102A12 includes a frequency domain conversion unit 102A111, a power
calculation unit 102A112, and an average power calculation unit 102A'121.
[0085]
Average power calculation unit 102A '121 receives spectrum powers xpow (ω) and ypow (ω),
detects average values xpow_AVG and ypow_AVG (average spectrum power) of spectrum power
for each frequency bin, and detects them as feature amount A Output as (2) and the feature
amount B (2).
Specifically, it is calculated according to the equations (13) and (14).
[0086]
[0087]
[0088]
The extraction unit 102A'13 receives the input signal x [n] and the output signal y [n] obtained
by changing the characteristics of the signal as input, and obtains the dispersion value of the
amplitude, and uses it to calculate the feature amount A (3) Output as B (3).
The extracting unit 102A'13 includes an amplitude variance calculating unit 102A'131.
Amplitude dispersion calculation unit 102A '131 receives output signal y [n] obtained by
changing the characteristic of the signal in addition to input signal x [n], and obtains the
dispersion value of the amplitude, and calculates it as feature amount A (3). And the feature
amount B (3).
[0089]
08-05-2019
21
Specifically, it is determined according to the following equations (15) and (16).
Here, xmean in the following equation (15) is an average value of x [n], and ymean in the
following equation (16) is an average value of y [n].
[0090]
[0091]
[0092]
The extraction unit 102A'14 receives the output signal y [n] obtained by changing the
characteristic of the signal in addition to the input signal x [n], obtains the number of zero
crossings, and calculates the number of feature points A (4) and the feature amount Output as B
(4).
The extracting unit 102A'14 includes a zero crossing number calculating unit 102A'141.
[0093]
The zero cross number calculation unit 102A '141 receives the output signal y [n] obtained by
changing the characteristic of the signal in addition to the input signal x [n], and receives the zero
cross number (x [n] = 0, y [n] = In the case of 0, and when the positive and negative of x [n] and y
[n] are inverted, the result is output as the feature amount A (4) and the feature amount B (4).
The zero crossing number is used to determine voiced speech and unvoiced speech, so that it is
possible to prevent clipping while suppressing excessive volume reduction of the signal
separately for voiced speech and unvoiced speech.
[0094]
08-05-2019
22
The regression operation unit 102A'2 receives the feature amount A (m) and the feature amount
B (m) (m = 1, 2, 3, 4) as an input, and uses the equation of the square error minimum criterion
which is regressed in advance. Based on the factor p is output.
Specifically, it is expressed by equation (17).
[0095]
[0096]
Here, ζ ′ (m) and ξ ′ (m) (m = 1, 2, 3..., S) in the equation (17) are threshold values and
characteristics of minimum necessary level control coefficients for preventing clipping in
advance It is the coefficient which carried out regression analysis of quantity A (m) and featurevalue B (m), and can be preset.
Here, since the feature amount B (m) is a feature amount of the output signal obtained by
changing the characteristics of the signal, it is easy to determine whether the signal is clipped.
This means that the correlation with the threshold of the minimum level control coefficient that
prevents clipping is higher, and the ideal coefficient p can be calculated.
[0097]
As described above, even with such a configuration, the same effect as that of the first
embodiment can be obtained. Moreover, according to such a configuration, since the feature
amount B (m) is used, it is easy to determine whether the signal is clipping. This makes it possible
to calculate a more ideal coefficient p because the correlation with the threshold of the minimum
level control coefficient that prevents clipping is high. That is, according to the coefficient p, the
digital level correction unit 101A and the characteristic correction unit 101B can perform
adaptive processing with high accuracy. Therefore, while the digital signal is effectively
prevented from being clipped, excessive volume reduction is prevented. it can.
08-05-2019
23
[0098]
Third Embodiment Next, a third embodiment according to the present invention will be
described. FIG. 7 shows the configuration. In the following description, the same components as
those in the above-described embodiment are denoted by the same reference numerals, and
redundant description will be omitted as appropriate to simplify the description.
[0099]
As shown in FIG. 7, the signal correction unit 101 in the third embodiment includes an automatic
volume correction unit 101C in addition to the digital level correction unit 101A and the
characteristic correction unit 101B. In the following description, the automatic volume correction
unit 101C is located downstream of the digital level correction unit 101A. However, the
automatic volume correction unit 101C may be upstream of the digital level correction unit 101A
or downstream of the characteristic correction unit 101B.
[0100]
Automatic volume correction unit 101C receives signal d [n], performs automatic volume control,
and outputs it as signal z [n]. The volume that is automatically set is determined based on a
volume control coefficient GAIN [n] (details will be described later) output from the characteristic
change unit 102 described later. That is, the signal d [n] is amplified or attenuated by the volume
control coefficient GAIN [n] to obtain an output signal z [n] (z [n] = GAIN [n] · d [n]). By doing this,
it is possible to automatically adjust to an appropriate sound pressure (such as a sound pressure
that human feels comfortable).
[0101]
The characteristic change unit 102 uses a coefficient calculation unit 102Aa instead of the
coefficient calculation unit 102A used in the characteristic change unit 102 described in the
above-described embodiment. Also, a balance setting unit 102Ba is used instead of the balance
setting unit 102B. Furthermore, an automatic volume adjustment unit 102E is used.
08-05-2019
24
[0102]
As shown in FIG. 8, the coefficient calculation unit 102Aa receives the input signal x [n], outputs
the threshold of the minimum level control coefficient necessary to prevent clipping as the
coefficient p1, and further performs automatic volume control. The coefficient is output as the
coefficient p2. The coefficient calculation unit 102Aa includes a feature quantity extraction unit
102A and a regression operation unit 102A2a.
[0103]
The regression operation unit 102A2a receives the feature amount A (m) (m = 1, 2, 3, 4) as an
input, and calculates the coefficient p1 and the coefficient p2 based on the equation of the
square error minimum criterion which has been subjected to regression analysis in advance.
Output. Specifically, the coefficients p1 and p2 are calculated by the equations (18) and (19).
[0104]
[0105]
[0106]
Here, ζa (m) (m = 1, 2, 3, 4) in the equation (18) represents the threshold value of the minimum
level control coefficient required to prevent clipping in advance and the feature value A (m) It is
an analyzed coefficient and can be set in advance.
[0107]
Further, ξa (m) (m = 1, 2, 3, 4) in the equation (19) is a coefficient obtained by regression
analysis of the coefficient for automatically adjusting to the appropriate sound pressure in
advance and the feature quantity A (m) And can be set in advance.
If correlation between feature value A (m) and threshold value for minimum necessary level
control coefficient to prevent clipping and coefficient for automatic adjustment to appropriate
08-05-2019
25
sound pressure is high, more ideal coefficient p1 and coefficient p2 are calculated can do.
[0108]
The balance setting unit 102Ba receives the coefficient p1 and the coefficient p2 and makes a
condition judgment so that sound quality deterioration (for example, distortion due to level
control, unnaturalness in volume due to automatic volume control, sense of noise, etc.) does not
occur. The temporary level control coefficient THp, the temporary characteristic change
coefficient Gp (ω), and the temporary sound volume control coefficient GAINp are output.
[0109]
Here, the provisional level control coefficient THp and the provisional characteristic change
coefficient Gp (ω) are determined by the same processing as in the first embodiment according
to the coefficient p1.
The temporary volume control coefficient GAINp obtains an absolute value | GAINp1-p2 | of the
difference between the temporary volume control coefficient GAINp1 and the coefficient p2
determined one frame earlier than the present, and is smaller than the upper limit value γ (|
GAINP1-p2 | <γ ), The provisional volume control coefficient GAINp is output as the coefficient
p2 as it is.
[0110]
On the other hand, when it is larger than the upper limit value γ (| GAINp1-p2 |> γ), the
temporary volume control coefficient GAINp is limited and output to γ.
Here, the upper limit value γ is a volume difference that does not cause a hearing problem, and
is generally about 3 dB.
However, the upper limit value γ is not limited to this. By performing such processing, it is
possible to prevent the sound quality deterioration (unnaturalness of the volume due to the
automatic volume control).
08-05-2019
26
[0111]
Automatic volume adjustment unit 102E receives the temporary volume control coefficient
GAINp, performs smoothing processing (smoothing processing) on a sample basis to suppress
transient influence between adjacent frames, and is set in automatic volume correction unit
101C. Output the volume control coefficient GAIN [n]. Specifically, it is determined according to
equation (20).
[0112]
[0113]
Here, the volume control coefficient GAIN [n] can be calculated using the volume control
coefficients GAIN [n−L] to GAIN [n−1] calculated up to S samples before the present.
Also, φLj and φp are smoothing coefficients, and weighting for smoothing can be changed. It
should be noted that this automatic volume adjustment unit 102E only needs to suppress
transient effects between adjacent frames, and may use smoothing processing as shown in, for
example, Japanese Patent Application Laid-Open No. P2007-93827.
[0114]
As mentioned above, even if it is such composition, the same effect as an example mentioned
above is exhibited. Moreover, according to such a configuration, it is possible to automatically
adjust to an appropriate sound pressure in addition to preventing an excessive decrease in sound
volume while preventing clipping of the digital signal.
[0115]
The present invention is not limited to the above embodiment as it is, and at the implementation
stage, the constituent elements can be modified and embodied without departing from the scope
08-05-2019
27
of the invention. Further, various inventions can be formed by appropriately combining a
plurality of constituent elements disclosed in the above embodiment. Further, for example, a
configuration in which some components are deleted from all the components shown in the
embodiment is also conceivable.
[0116]
Furthermore, the components described in different embodiments may be combined as
appropriate. For example, in the above embodiment, the characteristic change unit 102 is
described as performing signal processing with a plurality of functional blocks as shown in FIG. It
is possible to realize with one chip.
[0117]
In this case, the signal processing may be performed according to the flowchart shown in FIG.
That is, after acquiring the input signal x [n] in step 2a, in step 2b, processing similar to the
feature quantity extraction unit 102A1 is performed, and in step 2c, processing similar to the
regression operation unit 102A2 is performed. Then, in step 2d, the same process as the balance
setting unit 102B is performed, and in step 2e, the same process as the digital level adjustment
unit 102C and the characteristic adjustment unit 102D is performed.
[0118]
It goes without saying that the present invention can be implemented in various ways without
departing from the scope of the present invention.
[0119]
FIG. 1 is a circuit block diagram showing a configuration of a signal characteristic change device
according to an embodiment of the present invention.
FIG. 7 is a view showing an example of a provisional level control coefficient THp set by the
balance setting unit of the signal characteristic change device according to the embodiment of
the present invention. The figure showing an example of the temporary characteristic change
08-05-2019
28
coefficient Gp ((omega)) set by the balance setting part of the signal characteristic change
apparatus concerning one Embodiment of this invention. The structural example of 102 A of
coefficient calculation parts of the signal characteristic change apparatus concerning one
Embodiment of this invention. The circuit block diagram which shows the structure of the signal
characteristic change apparatus concerning the 2nd Embodiment of this invention. The structural
example of coefficient calculation part 102A 'of the signal characteristic change apparatus
concerning the 2nd Embodiment of this invention. The circuit block diagram which shows the
structure of the signal characteristic change apparatus concerning the 3rd Embodiment of this
invention. The structural example of the coefficient calculation part 102Aa of the signal
characteristic change apparatus concerning the 3rd Embodiment of this invention. The flowchart
for demonstrating the process of the signal correction apparatus concerning one Embodiment of
this invention.
[0120]
101 Signal correction unit 102 Characteristic change unit 101A Digital level correction unit
101B Characteristic correction unit 102A Coefficient calculation unit 102B Balance setting unit
102C Digital level adjustment unit 102D Characteristic adjustment unit 102A1 ... feature amount
extraction unit 102A2 ... regression operation unit 102A2, 102A11 ... extraction 1 unit, 102A12
... extraction 2 unit, 102A13 ... extraction 3 unit, 102A14 ... extraction 4 unit, 102A1S ...
extraction S unit, 102A 111 ... frequency domain conversion unit 102A112 ... power calculation
unit, 102A 113 ... maximum power calculation unit, 102A 121 ... average power calculation unit,
102A 131 ... amplitude dispersion calculation unit, 102A 141 ... zero crossing number calculation
unit.
08-05-2019
29
Документ
Категория
Без категории
Просмотров
0
Размер файла
41 Кб
Теги
jp2010212887
1/--страниц
Пожаловаться на содержимое документа