JPH07210198A - Sound information processor - Google Patents

Abstract

(57) [Summary] [Purpose] Interpolation based on the audio power of each frame.
To provide a speech coder / decoder device or the like in which the amount of calculation is reduced by determining whether to perform non-interpolation. [Structure] A means for calculating the difference value between the voice power of the current frame and the power of the previous frame stored in advance, and a calculation based on the difference value and the power of the current frame specified in advance. If the calculated difference value is less than or equal to the threshold value, the interpolated coefficient is used in the current frame, and if the difference value is greater than or equal to the threshold value, the non-interpolation coefficient is used. It is configured to have means.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a code length excitation linear predictive coding (CELP) system speech code decoder and a communication device such as a car telephone system including the decoder for the code.

[0002]

2. Description of the Related Art A digital full-rate speech codec for a car telephone system in Japan
Vector sum excitation linear predictive coding (VSELP: Ve)
center Sum Excite Linear Pr
However, the standard relating to such a system is, for example, “Digital Car Telephone System Standard RCR STD-27B.
(Referred to as RCR standard) ”.

A conventional technique will be described with reference to FIG.

In the VSELP system, the input analog audio signal is sampled at a frequency of 8 (kHz),
The sampled signal is an A / D converter (not shown)
Is converted into a digital signal via. This digital signal is input to the linear prediction analyzer 20.

In the linear prediction analyzer 20, a reflection coefficient (linear prediction) is obtained by a publicly-known or publicly-known technique, for example, a covariance method (for example, described in "Basics of Speech Signal Processing (Ohmsha)"). A coefficient (corresponding to a coefficient) is obtained, and a frame of a predetermined time (20 (ms) is calculated by the Levinson-Durbin recursive method (for example, described in "Digital Speech Processing (Tokai University Press)"). )) For each, the linear prediction coefficient is calculated. Then, in order to improve the voice quality, interpolation processing is performed on the linear prediction coefficient to obtain the linear prediction coefficient for each subframe (5 ms) obtained by subdividing the frame.

The linear prediction coefficient interpolation process used in the conventional VSELP system is disclosed in, for example, Japanese Patent Laid-Open No. 4-23329.
No. 99 publication, "Signal Encoding Method Used for Speech Encoder Using Soft Interpolation Determination of Spectrum Parameter".

In the RCR standard, the non-interpolation coefficient calculation unit 101 shown in FIG.
(k, i) (0 ≦ k <Ns: Ns is the number of subframes, for example, “4”, 0 ≦ i <Np: Np is the order of the prediction coefficient, for example, “10”) 3 is used and calculated for each subframe.

[0008]

[Equation 1]

[0009]

[Equation 2]

[0010]

[Equation 3]

Here, α (i) is the linear prediction coefficient of the current frame, and αpr (i) is the linear prediction coefficient of the previous frame.

Further, the interpolation coefficient αi (k, i) is calculated and obtained by the interpolation coefficient calculation unit 102 for each subframe using the following expressions 4 to 7.

[0013]

[Equation 4]

[0014]

[Equation 5]

[0015]

[Equation 6]

[0016]

[Equation 7]

The interpolation coefficient .alpha.i (k, i) is converted into the reflection coefficient ri.
Converting to (k, i) gives:

First, for a certain sub-frame k, the initial value is set to ri ¹⁰ = αi (k, i) (note that the suffix parenthesis at the upper right of ri is omitted). j = 10, 9, 8,
... 2 may be calculated according to the following equation 8.

[0019]

[Equation 8]

If the absolute value of this reflection coefficient is 1 or less,
The filter using the above-described interpolation coefficient is considered to be in a “stable” state, and the following inverse filter operation is performed. However, if at least one of the reflection coefficients has an absolute value of 1 or more, the filter is considered to be in an “unstable” state, and the non-interpolation coefficient is used, and the soft interpolation flag is set. Set "0".

Here, the soft interpolation flag is a flag expressing whether each frame uses a non-interpolation coefficient or an interpolation coefficient, and is used to perform decoding processing by adopting this method. It is one of the data transmitted to the device including the decryption unit.

Next, regarding the non-interpolation coefficient, the above αn (k,
Using i), the inverse filter unit 103 performs so-called inverse filter processing using the transfer function Hn (z) represented by the following Expression 9.

[0023]

[Equation 9]

Further, the power ENE_N of the output residual signal Z_n (k, i) is calculated by the following equation 10.

[0025]

[Equation 10]

Similarly, in the inverse filter unit 104, the coefficient αi
(k, i) is used to perform the inverse filtering process shown in the following formula 11, and the residual signal power ENE shown in the following formula 12.
Ask for _I.

[0027]

[Equation 11]

[0028]

[Equation 12]

Here, Z_i represents the output residual signal by Hi (z).

The above-mentioned ENE_N and ENE_I are compared in the residual power comparing section 105, and a linear prediction coefficient giving a smaller residual energy is used. That is, when ENE_N> ENE_I ... αi (k, i) is used, and when ENE_N ≦ ENE_I ... αn (k, i) is used.

The determination unit 106 performs such determination processing, outputs the linear prediction coefficient after interpolation to the terminal 90, and when ENE_N> ENE_I ... Soft interpolation flag = 0 When ENE_N ≦ ENE_I ... Soft interpolation flag = 1, and the value of the soft interpolation flag is transmitted to a device equipped with a decoding unit for performing the decoding process using this method.

[0032]

In the above-mentioned conventional technique, arithmetic processing for obtaining both interpolation and non-interpolation coefficients is performed, and further inverse filtering processing is performed to obtain a sum of outputs as processing results. And the comparison process is performed, the product-sum calculation process is performed about 1200 times for each sub-frame in the RCR standard even if only the inverse filter process is performed, resulting in a large calculation amount and processing time. Was needed.

For this reason, it has been extremely difficult to reduce the size and power consumption of the device required as a portable terminal.

[0034]

In order to solve the above problems, the following means are considered.

The linear prediction coefficient calculation means for calculating the linear prediction coefficient of the audio signal for each fixed time interval (frame) and the time interval (subframe) obtained by dividing the frame into a plurality of
At least one interpolation processing unit that performs an interpolation process using the weighted addition value of the linear prediction coefficients in the previous frame and the current frame as the linear prediction coefficient of the subframe, and for each current subframe based on the linear prediction coefficient. ,
Processing means for determining parameters including lag, noise source (codebook), and gain is provided. Furthermore, measuring means for measuring the power of the audio signal, storage means for storing the measured power of the audio signal, power of the audio signal of the previous frame stored in the storage means and the current frame measured by the measuring means. Calculation means for calculating the difference value with the power of the audio signal, comparison means for comparing the difference value with a predetermined threshold value, and if the difference value is greater than or equal to the predetermined threshold value, the interpolation is performed. On the contrary, when the difference value is smaller than a predetermined threshold value without performing the processing, a linear prediction coefficient correction means for performing the interpolation processing is provided.

Further, it may be configured to further include a threshold value updating means for updating the threshold value and a threshold value calculating means for obtaining the threshold value to be updated according to a predetermined rule. In this case, it is preferable that the predetermined rule is a voice information processing device that is an average value of a threshold value determined before updating and the power of the voice signal of the current frame.

Further, instead of the linear prediction coefficient of the audio signal, the reflection coefficient of the audio signal (the degree of reflection of the sound wave in the transmission path when modeling the part from the glottal to the lip on the transmission path through which the sound wave is transmitted) It is also possible to adopt a configuration in which a reflection coefficient calculation means for calculating a parameter that determines the

[0038]

First, when a voice signal is input, the voice power measuring device determines the voice power for one frame. Next, recall the stored voice power of the previous frame,
The difference from the voice power of the current frame is calculated.

If the input voice is at the time of rising, the voice power of the previous frame is relatively small and the difference from the voice power of the current frame is large. If this difference is equal to or greater than a threshold value designated in advance, a non-interpolation determination that non-interpolation processing should be performed is performed, and a linear prediction coefficient composed of non-interpolation coefficients is selected.

If the input voice is a part of the voice being uttered, the voice power of the previous frame and the voice power of the current frame are close to each other, and the power difference is small. Therefore, the power difference value becomes equal to or smaller than the threshold value, the interpolation determination that the interpolation process should be performed is performed, and the interpolation linear prediction coefficient configured by the interpolation coefficient is selected.

Next, when the frame corresponding to the trailing edge of the sound, that is, the end of utterance, is input, the power difference from the previous frame becomes a negative value. Therefore, the power difference value becomes equal to or less than the threshold value, and the interpolation determination that the interpolation process should be performed is performed.

In other cases, for example, even at the time of rising voice, if the voice power is low, that is,
If the user starts speaking in a small voice, the difference between the voice power of the previous frame and the voice power of the previous frame does not exceed the threshold value, and the interpolation determination is performed. On the other hand, even if it is a part of the voice being uttered, no interpolation is determined when a loud voice suddenly becomes loud. However, with the configuration in which the threshold value is updated, even in the above case, if the audio power of the previous frame is sufficiently small, the power difference value becomes equal to or greater than the threshold value, and no interpolation is determined.

Also, when noise is input, the voice power of the previous frame becomes almost the same value as the voice power of the current frame, and the interpolation determination is performed, unless the noise is very sudden noise.
In such a case, since the volume of the sound itself is small, whether the interpolation coefficient is used or the non-interpolation coefficient is used, the sound quality does not change so much.

As described above, the input voice power is measured for each frame, and when the difference from the input voice power of the previous frame is equal to or more than a predetermined threshold value, the non-interpolation linear prediction coefficient is used, and other than that. In this case, the object of the present invention is achieved by using interpolated linear prediction coefficients.

[0045]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows an embodiment of a means for performing interpolation judgment of a linear prediction coefficient according to the present invention.

In the present embodiment, the linear prediction analyzer 20, the voice power measuring device 130, the memory 40, the voice power difference calculation unit 50, the comparison unit 70, and the linear prediction coefficient correction unit 80.
And is configured. Furthermore, the threshold (E) 60 is
A terminal for inputting to the comparison unit 70 is provided.

Further, each component is connected via a signal line.

Here, each component will be described.

The linear predictive analyzer 20 converts the voice input signal into A
A means for performing linear prediction analysis from the D / D converted (analog / digital converted) digital audio signal 10 to obtain a linear prediction coefficient. For example, CPU, ROM, RA
It is realized by an electronic device such as M. The A / D
An A / D converter having a function of performing conversion is not shown.

The voice power measuring device 130 is a means for measuring the power of the input voice from the digital voice signal 10 and is realized by an electronic device such as a CPU, a ROM, a RAM and the like.

The power may be calculated, for example, by obtaining the square value of the digital audio signal for a predetermined time.

The memory 40 is a means for storing similarly measured audio power in the previous frame, and is realized by an electronic device such as a RAM.

The voice power difference calculation unit 50 is means for calculating a power difference value between two frames from the output of the voice power measuring device 130 and the power value stored in the memory 40. For example, a CPU , ROM, RAM and other electronic devices.

The comparing section 70 is means for comparing the difference between the voice powers and a threshold value E60 which is separately input, and determining whether or not the difference value is equal to or more than E. For example, various CMOs are used.
It is realized by an electronic device such as S.

The linear prediction coefficient correction unit 80 is means for correcting the linear prediction coefficient which is the output of the linear prediction analyzer 20 according to the result of the comparison unit 70. For example, CPU, RO
It is realized by an electronic device such as M and RAM. Note that 90 indicates the linear prediction coefficient output from the linear prediction coefficient correction unit 80.

Next, the operation of this embodiment will be described with reference to FIG.

First, in step 10, the linear prediction analyzer 20 and the audio power measuring device 130 take in the digital audio signal Sij (for example, 0≤j <160) of the i-th frame.

Next, in step 20, the linear prediction analyzer 20 causes the linear prediction coefficient α (i) of the digital input speech.
(0 ≦ i <Np) is calculated by using a publicly known method for obtaining a linear prediction coefficient.

Next, in step 130, the voice power measuring device 130 calculates the voice power Pi by calculating the sum of the squares of the input voices (Equation 13), and outputs the voice power Pi to the voice power difference calculating section 50. To do.

[0061]

[Equation 13]

Next, in step 40, the memory 40
The audio power Pi-1 of the preceding frame is taken into the audio power difference calculation unit 50.

Next, in step 50, the voice power difference calculation unit 50 obtains the difference ε between the voice power of the current frame and the voice power of the previous frame (equation 14).

[0064]

[Equation 14]

Next, in step 70, the comparison section 70
Compares the predetermined threshold value E with the sound power difference ε, and branches to step 71 if ε <E and to step 72 if ε ≧ E.

Next, in step 71, "interpolation determination" is performed, and in step 72, "no interpolation determination" is performed.

Finally, in step 80, from the result of the previous step, the linear prediction coefficient correction unit 80 sets the linear prediction coefficient to the interpolation coefficient αi (k, i) of the k-th subframe, or
The non-interpolation coefficient αn (k, i) (0 ≦ k <Ns, 0 ≦ i <Np) is corrected and output. However, the calculation method of the interpolation / non-interpolation coefficient is the same as the publicly known technique.

According to the present embodiment, the calculation of the inverse filter, which is indispensable in the existing technology, becomes unnecessary, the calculation amount can be greatly reduced, the device can be downsized and the power consumption can be reduced. The same sound quality can be obtained.

The above is an embodiment for correcting the linear prediction coefficient, and FIG. 9 shows an embodiment for correcting the reflection coefficient. Here, the reflection coefficient is a coefficient that uniquely corresponds to the linear prediction coefficient, and is a constant used in this method (the same applies hereinafter).

The embodiment shown in FIG. 9 is a linear prediction analyzer 20.
The audio power measuring device 130, the memory 40, the audio power difference calculation unit 50, the comparison unit 70, the linear prediction coefficient correction unit 80, and the conversion unit 100. Furthermore, a terminal for inputting the threshold value (E) 60 to the comparison unit 70 is provided. Moreover, each component is connected via a signal line.

Since the constituent elements here designated by the same reference numerals as those in FIG. 1 are the same constituent elements, description thereof will be omitted.
Therefore, as can be seen with reference to FIG.
Only 0 is newly provided.

The conversion section 100 is means for converting the reflection coefficient into a linear prediction coefficient, and can be realized by various CMOSs or the like.

The operation of this embodiment will be described with reference to FIG.

Steps 10, 130, 40, 50, 7
The processes at 0, 71, and 80 are almost the same as the processes at the corresponding steps in FIG. That is, the “linear prediction coefficient” may be replaced with the “reflection coefficient” and similar processing may be performed.

In step 100, based on the reflection coefficient thus obtained, for example, the interpolation coefficient may be obtained by the equation 8 and the linear prediction coefficient may be corrected using the obtained interpolation coefficient.

Also in the embodiment shown in FIG. 9, the device can be downsized and the power consumption can be reduced as in the case of FIG.

Next, FIG. 3 shows another embodiment according to the present invention.

This embodiment comprises a linear prediction analyzer 20, a voice amplitude measuring unit 230, a memory 41, a voice amplitude difference calculating unit 51, a comparing unit 75, and a linear prediction coefficient correcting unit 80. To be done.

Further, a terminal 61 for inputting the threshold value (D) to the comparing section 75 is provided. That is, in this embodiment, a voice amplitude measuring device 230 is used instead of the voice power measuring device 130 in the embodiment shown in FIG. 1, and a voice amplitude difference calculating unit 51 is used instead of the voice power difference calculating unit 50. Is a component. In addition, the memory 40 is a memory 41.
However, the comparison section 7
0 corresponds to the comparison unit 75, but can be considered to be exactly the same. Moreover, each component is connected via a signal line.

Since the constituent elements here designated by the same reference numerals as those in FIG. 1 are the same constituent elements, the description thereof will be omitted.
Therefore, as understood with reference to FIG. 3, the voice amplitude difference calculation unit 51 and the voice amplitude measuring device 230 are newly provided.

The voice amplitude measuring unit 230 is a means for taking the absolute value of the amplitude of the voice signal as an input, and for calculating the total sum of the frames, by inputting the digital voice signal obtained by A / D converting the voice input signal. , CPU, RAM, ROM and other electronic devices. That is, the voice amplitude measuring device 230 has a function of measuring power.

The voice amplitude difference calculation unit 51 is means for obtaining the difference between the total sums of the absolute values of two voices.
It is realized by an electronic device such as U, RAM, or ROM.

Next, the operation of this embodiment will be described with reference to FIG.

First, in step 10, the linear prediction analyzer 20 and the voice amplitude measuring device 230 take in the digital voice signal Sij of the i-th frame (for example, 0≤j <160).

Next, in step 20, the linear prediction analyzer 20 causes the linear prediction coefficient α (i) of the digital input speech.
(0 ≦ i <Np) is calculated according to a publicly known method.

Next, in step 230, the voice amplitude measuring unit 230 calculates the total sum Si of the absolute values of the input voice (Equation 15) and outputs it to the voice amplitude difference calculating section 51.

[0087]

[Equation 15]

Next, in step 41, the memory 41
From the above, the total sum Si−1 of the absolute values of the audio amplitudes of the previous frame stored is fetched into the audio amplitude difference calculation unit 51.

Next, in step 51, the voice amplitude difference calculation unit 51 obtains the difference δ of the total sum of the absolute values of the voice amplitudes of the current frame and the previous frame by the following equation 16.

[0090]

[Equation 16]

Next, in step 75, the comparison section 75
Compares the predetermined threshold value D with the difference δ of the total sum of the absolute values of the voice amplitude. If δ <D, step 71
If δ ≧ D, the process branches to step 72.

Next, in step 71, interpolation judgment is made, and in step 72, no interpolation judgment is made.

Finally, in step 80, from the result of the previous step, the linear prediction coefficient correction unit 80 sets the linear prediction coefficient to the interpolation coefficient αi (k, i) of the m-th subframe, or
The non-interpolation coefficient αn (k, i) (0 ≦ k <Ns, 0 ≦ i <Np) is corrected and output. However, as a method of calculating the interpolation / non-interpolation coefficient, a publicly known and publicly known technique for calculating the interpolation / non-interpolation coefficient may be used.

According to the present embodiment, the calculation of the inverse filter, which is indispensable in the existing technology, becomes unnecessary, the calculation amount can be reduced, the device can be downsized, and the power consumption can be reduced. Further, it is equivalent to the known technology. The sound quality of is obtained.

Further, by deleting the square calculation in the power calculation in the above-mentioned embodiment, it is possible to improve the processing accuracy in the comparison section.

The above is an embodiment for correcting the linear prediction coefficient, and FIG. 11 shows an embodiment for correcting the reflection coefficient.

In the embodiment shown in FIG. 11, the linear prediction analyzer 2 is used.
0, a voice amplitude measuring unit 230, a memory 41, a voice power difference calculation unit 51, a comparison unit 75, a linear prediction coefficient correction unit 80, and a conversion unit 100. Further, the terminal 6 for inputting the threshold value (D) to the comparison unit 75.
1 is provided. Moreover, each component is connected via a signal line.

Since the constituent elements here designated by the same reference numerals as those in FIG. 3 are the same constituent elements, the description thereof will be omitted.
Therefore, as can be seen with reference to FIG.
Only 00 is newly provided.

The conversion unit 100 is means for converting the reflection coefficient into a linear prediction coefficient, and can be realized by various CMOSs or the like.

The operation of this embodiment will be described with reference to FIG.

Steps 10, 230, 41, 51, 7
Since the processes in 5, 71, 72 and 80 are almost the same as the processes in the corresponding steps in FIG. 3, the description thereof will be omitted. That is, "linear prediction coefficient" is replaced by "reflection coefficient"
, And similar processing may be performed.

In step 100, based on the reflection coefficient thus obtained, for example, the interpolation coefficient may be obtained by the equation 8 and the linear prediction coefficient may be corrected using the obtained interpolation coefficient.

Also in the embodiment shown in FIG. 11, the device can be downsized and the power consumption can be reduced as in the case of FIG.

Next, FIG. 5 shows another embodiment according to the present invention.

In this embodiment, the linear prediction analyzer 20, the voice power measuring device 130, the memory 40, the voice power difference calculating section 50, the comparing section 70, and the linear prediction coefficient correcting section 80 are used.
And a threshold calculation unit 61.

Since the constituent elements here designated by the same reference numerals as those in FIG. 1 are the same constituent elements, the description thereof will be omitted.

That is, the present embodiment is configured by including the threshold value calculation unit 61 in the configuration of the first embodiment.

The threshold calculation unit 61 is means for storing the threshold E_pr used in the previous frame, further accepting the voice power Pi in the current frame, and setting the average value of these two values as the new threshold E. For example, CPU, RAM, RO
It is realized by an electronic device such as M.

Next, the operation of this embodiment will be described with reference to FIG.

First, in step 10, the linear prediction analyzer 20 and the audio power measuring device 130 take in the digital audio signal Sij of the i-th frame (for example, 0≤j <160).

Next, in step 20, the linear prediction analyzer 20 causes the linear prediction coefficient α (i) of the digital input speech.
(0 ≦ i <Np) is calculated according to a known and publicly known method for obtaining a linear prediction coefficient.

Next, in step 130, the voice power measuring unit 130 calculates the voice power Pi by calculating the sum of the squares of the input voices (Equation 13), and the voice power difference calculator 50 and the threshold value are calculated. Output to the calculator 61.

Next, in step 40, the memory 40
The audio power Pi-1 of the previous frame is taken into the audio power difference calculation unit 50.

Next, in step 50, the voice power difference calculation unit 50 obtains the difference ε between the voice power of the current frame and the voice power of the previous frame (according to equation 14).

Next, in step 61, the threshold value calculation unit 61 calculates the average of the threshold value E_pr of the previous frame and the voice power Pi of the current frame according to the following expression 17, and sets this as the threshold value E of the current frame. .

[0116]

[Equation 17]

Next, in step 70, the comparison unit 70
Compares the threshold value E with the voice power difference ε, and branches to step 71 if ε <E and branches to step 72 if ε ≧ E.

Next, in step 71, interpolation determination is performed, and in step 72, no interpolation determination is performed.

Finally, in step 80, from the result of the previous step, the linear prediction coefficient correction unit 80 uses the linear prediction coefficient as the interpolation coefficient αi (k, i) of the k-th subframe, or
The non-interpolation coefficient αn (k, i) (0 ≦ k <Ns, 0 ≦ i <Np) is corrected and output. However, the calculation of the interpolation / non-interpolation coefficient is performed according to a publicly known method.

According to the present embodiment, the calculation of the inverse filter, which is indispensable in the existing technology, becomes unnecessary, the calculation amount can be reduced, the device can be downsized and the power consumption can be reduced, and the sound quality can be maintained. Further, as compared with the first embodiment, since the threshold value is variable, clearer voice can be obtained.

The above is an embodiment for correcting the linear prediction coefficient, and FIG. 13 shows an embodiment for correcting the reflection coefficient.

The embodiment shown in FIG. 13 is a linear prediction analyzer 2.
0, an audio power measuring unit 130, a memory 40, an audio power difference calculating unit 50, a comparing unit 70, a linear prediction coefficient correcting unit 80, a threshold value calculating unit 61, and a converting unit 100. It Moreover, each component is connected via a signal line.

Since the constituent elements here designated by the same reference numerals as those in FIG. 5 are the same constituent elements, description thereof will be omitted.
Therefore, as can be seen with reference to FIG.
Only 00 is newly provided.

The conversion section 100 is means for converting the reflection coefficient into a linear prediction coefficient, and can be realized by various CMOSs, for example.

The operation of this embodiment will be described with reference to FIG.

Steps 10, 130, 40, 61 and 5
Since the processing at 0, 70, 71, 72, 80 is almost the same as the processing at the corresponding steps in FIG. 6,
The description is omitted. That is, the “linear prediction coefficient” may be replaced with the “reflection coefficient” and similar processing may be performed.

In step 100, based on the reflection coefficient thus obtained, for example, the interpolation coefficient may be obtained by the equation 8 and the linear prediction coefficient may be corrected using the obtained interpolation coefficient.

Also in the embodiment shown in FIG. 14, as in the case of FIG. 5, it is possible to downsize the device, reduce the power consumption, maintain the sound quality, and the like.

Next, FIG. 7 shows another embodiment according to the present invention.

In this embodiment, the linear prediction analyzer 20, the voice amplitude measuring unit 230, the memory 41, the voice amplitude difference calculating unit 51, the comparing unit 75, the linear prediction coefficient correcting unit 80, and the threshold calculator 62 are used. And is configured. Moreover, each component is connected via a signal line.

Since the constituent elements here designated by the same reference numerals as those in FIG. 3 are the same constituent elements, the description thereof will be omitted.
Therefore, as can be seen with reference to FIG. 3, the threshold calculator 62 is newly provided.

The threshold calculator 62 is a means for taking the average of the sum of the absolute values of the amplitudes of the digital audio signals in the current frame and the threshold in the previous frame and setting it as the threshold in the current frame. , CPU, RAM, ROM and other electronic devices.

Next, the operation of this embodiment will be described with reference to FIG.

First, in step 10, the linear prediction analyzer 20 and the voice amplitude measuring device 230 take in the digital voice signal Sij of the i-th frame (for example, 0≤j <160).

Next, in step 20, the linear prediction analyzer 20 causes the linear prediction coefficient α (i) of the digital input speech.
(0 ≦ i <Np) is calculated according to a known and publicly known method for obtaining a linear prediction coefficient.

Next, in step 230, the voice amplitude measuring unit 230 calculates the total sum Si of the absolute values of the input voice (Equation 15), and the voice amplitude difference calculating unit 51 and the threshold value calculating unit 62.
Output to.

Next, in step 41, the memory 41
From the above, the stored sum total Si-1 of the absolute values of the voice amplitudes of the previous frame is fetched into the voice amplitude difference calculation unit 51.

Next, in step 51, the voice amplitude difference calculation unit 51 obtains the difference δ of the total sum of the absolute values of the voice amplitudes of the current frame and the previous frame according to the equation (16).

Next, in step 62, the threshold value calculator 62 obtains the average of the sum Si of the absolute values of the audio signals of the current frame and the threshold value D_pr of the previous frame according to the following expression 18 The threshold value D of is calculated.

[0140]

[Equation 18]

Next, in step 75, the comparison unit 75
Compares the threshold value D with the difference δ of the total sum of the absolute values of the voice amplitude, and when δ <D, the process proceeds to step 71, where δ ≧
If D, branch to step 72.

Next, in step 71, an interpolation judgment is made, and in step 72, no interpolation judgment is made.

Finally, in step 80, from the result of the previous step, the linear prediction coefficient correction unit 80 sets the linear prediction coefficient to the interpolation coefficient αi (k, i) of the k-th subframe, or
The non-interpolation coefficient αn (k, i) (0 ≦ k <Ns, 0 ≦ i <Np) is corrected and output. However, as a method of calculating the interpolation / non-interpolation coefficient, a publicly known or publicly used method may be used.

According to this embodiment, the calculation of the inverse filter, which is indispensable in the existing technology, becomes unnecessary, the amount of calculation can be reduced, the device can be downsized and the power consumption can be reduced, and the sound quality can be maintained. . Further, by deleting the squared calculation existing in the power calculation in the above-mentioned embodiment, the calculation processing accuracy in the comparison section can be improved. Furthermore, a clearer voice can be obtained by the configuration in which the threshold value is variable.

The above is the embodiment for correcting the linear prediction coefficient, and FIG. 11 shows an embodiment for correcting the reflection coefficient.

The embodiment shown in FIG. 15 is a linear prediction analyzer 2.
0, a voice amplitude measuring unit 230, a memory 41, a voice power difference calculating unit 51, a comparing unit 75, a linear prediction coefficient correcting unit 80, a threshold calculator 62, and a converting unit 100. It Moreover, each component is connected via a signal line.

Since the constituent elements here designated by the same reference numerals as those in FIG. 7 are the same constituent elements, the description thereof will be omitted.
Therefore, as can be seen with reference to FIG.
Only 00 is newly provided.

The conversion section 100 is means for converting the reflection coefficient into a linear prediction coefficient, and can be realized by various CMOSs, for example.

The operation of this embodiment will be described with reference to FIG.

Steps 10, 230, 41, 62 and 5
Since the processes at 1, 75, 71, 72, 80 are almost the same as the processes at the corresponding steps in FIG.
The description is omitted. That is, the “linear prediction coefficient” may be replaced with the “reflection coefficient” and similar processing may be performed.

In step 100, based on the reflection coefficient thus obtained, for example, the interpolation coefficient may be obtained by the equation 8 and the linear prediction coefficient may be corrected using the obtained interpolation coefficient.

Also in the embodiment shown in FIG. 15, the device can be downsized and the power consumption can be reduced as in the case of FIG.

As described above, the linear prediction coefficient or the reflection coefficient instead of the linear prediction coefficient is corrected to reduce the amount of calculation and downsize the device.
It is possible to provide a voice processing device suitable for the ELP method.

[0154]

According to the present invention, it is possible to determine whether to perform interpolation or no interpolation before calculating an interpolation or non-interpolation linear prediction coefficient. Therefore, one of interpolation and non-interpolation linear prediction coefficient is calculated. Moreover, since the calculation of residual energy is not necessary, the calculation amount can be significantly reduced, and the program scale and the memory amount can be reduced. As a result, the size and weight of the device can be reduced and the power consumption can be reduced.

Further, by making the threshold variable based on the history of the audio power, it is possible to maintain the sound quality of the audio signal regardless of the size of the input audio signal.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an embodiment of the present invention.

FIG. 2 is a flowchart showing a process in one embodiment of the present invention.

FIG. 3 is a configuration diagram of another embodiment of the present invention.

FIG. 4 is a flowchart showing processing in another embodiment of the present invention.

FIG. 5 is a configuration diagram of another embodiment of the present invention.

FIG. 6 is a flowchart showing a process in another embodiment of the present invention.

FIG. 7 is a configuration diagram of another embodiment of the present invention.

FIG. 8 is a flowchart showing a process in another embodiment of the present invention.

FIG. 9 is a configuration diagram of another embodiment of the present invention.

FIG. 10 is a flowchart showing a process in another embodiment of the present invention.

FIG. 11 is a configuration diagram of another embodiment of the present invention.

FIG. 12 is a flowchart showing a process in another embodiment of the present invention.

FIG. 13 is a configuration diagram of another embodiment of the present invention.

FIG. 14 is a flowchart showing a process in another embodiment of the present invention.

FIG. 15 is a configuration diagram of another embodiment of the present invention.

FIG. 16 is a flowchart showing a process in another embodiment of the present invention.

FIG. 17 is a configuration diagram showing an example of a conventional technique.

[Explanation of symbols]

10 ... Voice signal, 20 ... Linear prediction analyzer, 40 ... Memory, 41 ... Memory, 50 ... Voice power difference calculation unit, 60
... Threshold value, 61 ... Threshold value calculation section, 70 ... Comparison section, 100 ... Conversion section, 101 ... Non-interpolation coefficient calculation section, 102 ... Interpolation coefficient calculation section, 103 ... Inverse filter section (non-interpolation coefficient), 104 ...
Inverse filter unit (interpolation coefficient) 105 ... Residual power comparison unit 106 ... Interpolation / non-interpolation determination unit 130 ... Voice power measuring device 230 ... Voice amplitude measuring device

Claims (6)

[Claims] 1. A linear prediction coefficient calculation means for calculating a linear prediction coefficient of an audio signal for each fixed time interval (frame), and a time interval (subframe) obtained by dividing a frame into a plurality of parts.
At least one interpolation processing unit that performs an interpolation process using the weighted addition value of the linear prediction coefficients in the previous frame and the current frame as the linear prediction coefficient of the subframe, and for each current subframe based on the linear prediction coefficient. ,
A processing means for determining parameters including a lag, a noise source (codebook), and a gain is provided, and further, measuring means for measuring the power of the audio signal, storage means for storing the measured power of the audio signal, and the storage means. Comparing the difference value between the power of the audio signal of the previous frame and the power of the audio signal of the current frame measured by the measuring device, the magnitude relationship between the difference value and a predetermined threshold value. When comparing means for comparing and the difference value is equal to or more than a predetermined threshold value, the interpolation process is not performed, and conversely,
An audio information processing apparatus comprising: a linear prediction coefficient correction unit that performs the interpolation process when the difference value is smaller than a predetermined threshold value. 2. A linear prediction coefficient calculation means for calculating a linear prediction coefficient of an audio signal for each fixed time interval (frame), and a time interval (subframe) obtained by dividing a frame into a plurality of parts.
At least one interpolation processing unit that performs an interpolation process using the weighted addition value of the linear prediction coefficients in the previous frame and the current frame as the linear prediction coefficient of the subframe, and for each current subframe based on the linear prediction coefficient. ,
A processing means for obtaining parameters including a lag, a noise source (codebook), and a gain is provided, and further, a sampling means for sampling a voice signal in a frame and a sum of absolute values of amplitudes of the sampled voice signals are obtained. Amplitude absolute value calculation means, storage means for storing the calculated sum of absolute values of the amplitude of the audio signal, sum of absolute values of the amplitude of the audio signal of the previous frame stored in the storage means, and the amplitude absolute value calculation Calculating means for calculating the difference value between the sum of the absolute values of the amplitudes of the audio signals of the current frame obtained by the means, comparing means for comparing the difference between the difference value and a predetermined threshold value, and the difference value If the difference value is smaller than a predetermined threshold value, the interpolation process is not performed, and conversely, if the difference value is smaller than the predetermined threshold value, a linear prediction coefficient correction means is provided. A voice information processing device characterized by the following. 3. The method according to claim 1, further comprising threshold updating means for updating the threshold, and threshold calculating means for obtaining a threshold to be updated according to a predetermined rule. Voice information processing device. 4. The audio information processing apparatus according to claim 3, wherein the predetermined rule is an average value of a threshold value set before updating and the power of the audio signal of the current frame. 5. A reflection coefficient of an audio signal at a constant time interval (frame) (when the model from the glottal to the lip is modeled on a transmission path for transmitting sound waves, the degree of reflection of the sound waves on the transmission path is determined. Parameter) and at least one of the time intervals (subframes) obtained by dividing the frame into a plurality of frames, the weighted addition value of the reflection coefficients in the previous frame and the current frame is used as the reflection coefficient of the subframe. And interpolation processing means for performing an interpolation processing, and processing means for obtaining parameters including a lag, a noise source (codebook), and a gain for each current subframe based on a linear prediction coefficient uniquely corresponding to the reflection coefficient. Further, a measuring unit for measuring the power of the audio signal, a storage unit for storing the measured power of the audio signal, and a storage unit for storing the measured power of the audio signal. Comparing means for computing the difference value between the power of the voice signal of the previous frame and the power of the voice signal of the current frame measured by the measuring means, and comparing the magnitude relationship between the difference value and a predetermined threshold value. And the difference value is equal to or greater than a predetermined threshold value, the interpolation processing is not performed, and conversely,
An audio information processing apparatus comprising reflection coefficient correction means for performing the interpolation process when the difference value is smaller than a predetermined threshold value. 6. The threshold value updating means for updating the threshold value according to claim 5, and the threshold value to be updated is an average value of the threshold value determined before the update and the power of the audio signal of the current frame. A voice information processing apparatus, comprising: a threshold value calculating unit for obtaining.