JP2002123299A - Processor and method for processing speech, device and method for learning, program and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a high-quality synthetic sound. SOLUTION: In the receiving part 114 of a portable telephone of a CELP (Code Excited Liner Prediction coding) mode, a residual signal and a linear predictor coefficient and decoded from an L cord, G code, I code, and A code. In a speech synthesis filter 29, a synthetic sound is generated from the decoded residual signal and linear predictor coefficient. A classifying part 123 performs classification on the basis of a class tap generated from the L code, G code, I code, A code, the decoded residual signal and the linear predictor coefficient and outputs a corresponding class code to a coefficient memory 124. The coefficient memory 124 outputs a tap coefficient corresponding to the class code. A prediction part 125 determines the predicted value of a high-quality vocal sound by using the tap coefficient and the synthetic sound outputted by the speech synthesis filter 29.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a data processing device and a data processing method, a learning device and a learning method, and a program and a recording medium.
The present invention relates to a data processing device and a data processing method, a learning device and a learning method, and a program and a recording medium that enable a voice encoded by an LP (Code Excited Liner Prediction coding) method to be decoded into a high-quality voice. .

[0002]

2. Description of the Related Art FIGS. 1 and 2 show an example of the configuration of a conventional portable telephone.

[0003] In this portable telephone, a transmission process of encoding a speech into a predetermined code by the CELP method and transmitting the same, and a reception process of receiving a code transmitted from another portable telephone and decoding it into speech are performed. FIG. 1 shows a transmitting unit for performing a transmitting process, and FIG. 2 shows a receiving unit for performing a receiving process.

In the transmitting unit shown in FIG. 1, a voice uttered by a user is input to a microphone (microphone) 1, where it is converted into a voice signal as an electric signal, and A / D (A
(nalog / Digital) conversion unit 2. A / D converter 2
Represents an analog audio signal from the microphone 1, for example, 8
A / D conversion into a digital audio signal is performed by sampling at a sampling frequency such as kHz, and quantization is performed with a predetermined number of bits.
It is supplied to a C (Liner Prediction Coefficient) analysis unit 4.

[0005] The LPC analysis unit 4 performs an LPC analysis of the audio signal from the A / D conversion unit 2 for each frame having a length of, for example, 160 samples, and obtains P-order linear prediction coefficients α ₁ , α ₂ ,
..., determine the α _P. Then, the LPC analysis unit 4 calculates the P-order linear prediction coefficient α _p (p = 1, 2,..., P)
Is a feature vector of speech,
This is supplied to the vector quantization unit 5.

The vector quantization unit 5 stores a code book in which a code is associated with a code vector having a linear prediction coefficient as an element, and based on the code book,
The feature vector α from the LPC analysis unit 4 is vector-quantized, and a code obtained as a result of the vector quantization (hereinafter referred to as a code
A code (A_code), as appropriate),
To supply.

[0007] Further, the vector quantization unit 5 converts the linear prediction coefficients α ₁ ′, α ₂ ′,..., Α _P ′, which constitute the code vector α ′ corresponding to the A code, into speech. It is supplied to the synthesis filter 6.

The speech synthesis filter 6 is, for example, an IIR (I
nfinite Impulse Response) type digital filter, and the linear prediction coefficient α _p ′ (p
= 1, 2,..., P) as the tap coefficients of the IIR filter, and the residual signal e supplied from the arithmetic unit 14.
Is used as an input signal to perform speech synthesis.

That is, the LPC analysis performed by the LPC analysis unit 4 includes (a sample value of) the audio signal s _{n at} the current time n and the past P sample values s _n−1 and s adjacent thereto.
_n-2, ···, the s _nP, linear combination represented by the formula _{s n + α 1 s n-} 1 + α 2 s n-2 + ··· + α P s nP = e n ··· (1) There assuming satisfied, the predicted value of the sample value s _n at the current time n the (linear prediction value) s _n ', past the P sample values _{s n-1, s n-} 2, ···, s nP Using,
Wherein s _n '= - by _{(α 1 s n-1 +} α 2 s n-2 + ··· + α P s nP) ··· (2) when the linear prediction, the actual sample value s _n and linear prediction A linear prediction coefficient α _p that minimizes a square error with the value s _n ′ is obtained.

[0010] Here, in the formula _{(1), {e n}} (··
, E _n−1 , e _n , e _{n + 1} ,...) Are uncorrelated random variables having an average value of 0 and a variance of a predetermined value σ ² .

From the [0011] formula (1), the sample value s _n the formula _{s n = e n - (α} 1 s n-1 + α 2 s n-2 + ··· + α P s nP) ··· (3) Which can be expressed by the following equation.

S = E / (1 + α ₁ z ⁻¹ + α ₂ z ⁻² +... + Α _P z ^−P ) (4) In the equation (4), S and E are expressed by the equation (3) the Z transform of s _n and e _n in), it represents respectively.

Here, from equations (1) and (2), e _n
It is 'can be represented by (5), the actual sample value s _n and linear predicted value s _n' wherein e _n = s _n -s _n called residual signal between.

Therefore, from equation (4), the linear prediction coefficient α _p
With the tap coefficients of the IIR filter, by the residual signal e _n as an input signal of the IIR filter,
It can be obtained audio signal s _n.

Therefore, the speech synthesis filter 6 receives the linear prediction coefficient α _p ′ from the vector quantization unit 5 as described above.
Is used as a tap coefficient, and using the residual signal e supplied from the arithmetic unit 14 as an input signal, the equation (4) is calculated to obtain a speech signal (synthesized sound signal) ss.

In the speech synthesis filter 6, a linear prediction coefficient α obtained as a result of the LPC analysis by the LPC analysis unit 4 is used.
Instead of _p , a linear prediction coefficient α _p ′ as a code vector corresponding to the code obtained as a result of the vector quantization
Is used, the synthesized sound signal output from the sound synthesis filter 6 is not basically the same as the sound signal output from the A / D converter 2.

The synthesized sound signal ss output from the voice synthesis filter 6 is supplied to the arithmetic unit 3. The arithmetic unit 3 converts the synthesized sound signal ss from the speech synthesis filter 6 into an A / D converter 2
Subtracts the output audio signal s, and supplies the subtracted value to the square error calculator 7. The square error calculator 7 is configured to calculate
, The sum of the squares of the subtraction value from the sum (the sum of the squares of the sample values of the k-th frame) is calculated, and the resulting square error is supplied to the square error minimum determination unit 8.

The minimum square error determining section 8 correlates the square error output from the square error calculating section 7 with an L code (L_code) as a code representing a lag, a G code (G_code) as a code representing a gain, And an I code (I_code) as a code representing a code word.
Output the L code, the G code, and the L code corresponding to the squared error output by. The L code is supplied to the adaptive codebook storage unit 9, the G code is supplied to the gain decoder 10, and the I code is supplied to the excitation codebook storage unit 11. Further, the L code, the G code, and the I code are also supplied to the code determination unit 15.

The adaptive codebook storage unit 9 stores an adaptive codebook in which, for example, a 7-bit L code is associated with a predetermined delay time (lag). e is delayed by the delay time associated with the L code supplied from the square error minimum determination unit 8 and output to the arithmetic unit 12.

Here, since the adaptive codebook storage unit 9 outputs the residual signal e with a delay corresponding to the time corresponding to the L code, the output signal is converted into a periodic signal whose cycle is the delay time. It becomes a close signal. This signal is mainly used as a driving signal for generating a synthesized voiced voice in voice synthesis using linear prediction coefficients.

The gain decoder 10 stores a table in which a G code is associated with predetermined gains β and γ, and a gain β associated with the G code supplied from the square error minimum determining unit 8 is stored. And γ are output. The gains β and γ are supplied to computing units 12 and 13, respectively.

The excitation codebook storage unit 11 stores, for example, 9
An excitation codebook in which a bit I code is associated with a predetermined excitation signal is stored.
The excitation signal associated with the I code supplied from
Output to arithmetic unit 13.

Here, the excitation signal stored in the excitation codebook is, for example, a signal close to white noise or the like. In speech synthesis using a linear prediction coefficient, it is mainly used to generate a synthesized voice of unvoiced sound. It becomes a drive signal.

The arithmetic unit 12 stores the adaptive codebook storage unit 9
Is multiplied by the gain β output from the gain decoder 10, and the multiplied value 1 is supplied to the calculator 14. The arithmetic unit 13 multiplies the output signal of the excitation codebook storage unit 11 by the gain γ output by the gain decoder 10,
The multiplied value n is supplied to the arithmetic unit 14. Arithmetic unit 14
Adds the multiplied value 1 from the computing unit 12 and the multiplied value n from the computing unit 13 and uses the sum as a residual signal e as
It is supplied to the voice synthesis filter 6.

As described above, the speech synthesis filter 6 converts the residual signal e supplied from the arithmetic unit 14 into the input signal and the linear prediction coefficient α supplied from the vector quantization unit 5.
Filtering is performed by an IIR filter using _p ′ as a tap coefficient, and the resultant synthesized sound signal is supplied to the arithmetic unit 3. Then, the same processing as described above is performed in the arithmetic unit 3 and the square error calculator 7, and the square error obtained as a result is supplied to the minimum square error determiner 8.

The square error minimum judging section 8 judges whether or not the square error from the square error calculating section 7 has become minimum (minimum). Then, when determining that the square error is not minimized, the square error minimum determination unit 8 outputs an L code, a G code, and an L code corresponding to the square error, as described above. Is repeated.

On the other hand, when the square error minimum judging section 8 judges that the square error has become minimum, it outputs a determination signal to the code determining section 15. The code determination unit 15 latches the A code supplied from the vector quantization unit 5, as well as the L code and G code supplied from the square error minimum determination unit 8.
Code and I code are sequentially latched, and when a decision signal is received from the square error minimum determination unit 8, the A code, L code, G code, and I code latched at that time are converted into channel encoders. 16. The channel encoder 16 includes a code determination unit 15
A code, L code, G code, and I code are multiplexed and output as code data. This code data is transmitted via a transmission path.

In the following, in order to simplify the description,
The A code, L code, G code, and I code are determined for each frame. However, for example, 1
Divide the frame into four subframes, L code,
The G code and the I code can be determined for each subframe.

Here, in FIG. 1 (the same applies to FIGS. 2, 11 and 12 described later), each variable is marked with [k] and is an array variable. Although k represents the number of frames, the description thereof is omitted as appropriate in the specification.

Next, as described above, the code data transmitted from the transmitting section of another portable telephone is received by the channel decoder 21 of the receiving section shown in FIG. The channel decoder 21 converts the code data into an L code, a G code,
The code, the I code, and the A code are separated and supplied to an adaptive codebook storage unit 22, a gain decoder 23, an excitation codebook storage unit 24, and a filter coefficient decoder 25.

The adaptive codebook storage unit 22, the gain decoder 23, the excitation codebook storage unit 24, and the arithmetic units 26 to 28 are the adaptive codebook storage unit 9, the gain decoder 10, and the excitation codebook storage unit 11 of FIG. , And the arithmetic units 12 to 14, respectively, and by performing the same processing as in the case described with reference to FIG.
The G code and the I code are decoded into a residual signal e. The residual signal e is provided to the speech synthesis filter 29 as an input signal.

The filter coefficient decoder 25 stores the same codebook as that stored in the vector quantization unit 5 in FIG. 1, decodes the A code into a linear prediction coefficient α _p ′, It is supplied to the synthesis filter 29.

The speech synthesis filter 29 has the same configuration as the speech synthesis filter 6 shown in FIG. 1, uses the linear prediction coefficient α _p ′ from the filter coefficient decoder 25 as a tap coefficient, and is supplied from the calculator 28. Equation (4) is calculated using the residual signal e as an input signal, thereby generating a synthesized sound signal when the square error is determined to be the minimum in the square error minimum determination unit 8 of FIG. This synthesized sound signal is D /
The signal is supplied to an A (Digital / Analog) converter 30. The D / A converter 30 D / A converts the synthesized sound signal from the voice synthesis filter 29 from a digital signal to an analog signal,
The signal is supplied to the speaker 31 and output.

[0034]

As described above, in the transmitting section of the portable telephone, the residual signal and the linear prediction coefficient as filter data applied to the speech synthesis filter 29 of the receiving section are coded and transmitted. In order to get
The code is decoded into a residual signal and linear prediction coefficients.
However, the decoded residual signal and the linear prediction coefficient (hereinafter, appropriately referred to as a decoded residual signal and a decoded linear prediction coefficient, respectively) include an error such as a quantization error. And the linear prediction coefficient do not match.

For this reason, the speech synthesis filter 29 of the receiving section
Will have a distorted sound quality degraded.

The present invention has been made in view of such a situation, and aims to obtain a high-quality synthesized sound.

[0037]

SUMMARY OF THE INVENTION A speech processing apparatus according to the present invention extracts a high-quality sound for which a predicted value is to be obtained as a target voice, and extracts a prediction tap used for predicting the target voice from the synthesized voice. Predictive tap extracting means, a class tap extracting means for extracting, from a code, a class tap used to classify the target voice into any of several classes, and a class tap extracting means for extracting the target voice based on the class tap. A class classification means for performing class classification for obtaining a class, an acquisition means for acquiring a tap coefficient corresponding to a class of a target voice from among tap coefficients for each class obtained by performing learning, a prediction tap, A prediction unit that calculates a predicted value of the target voice using a tap coefficient corresponding to the class of the target voice.

The voice processing method according to the present invention includes a step of extracting a prediction tap used for predicting a target voice using a high-quality voice for which a predicted value is to be obtained as a target voice from a synthetic voice. A class tap extraction step of extracting, from a code, a class tap used to classify the target voice into one of several classes, and a class classification for obtaining a class of the target voice based on the class tap. A class classification step to be performed; an acquisition step of acquiring a tap coefficient corresponding to a class of a target voice from tap coefficients for each class obtained by performing learning; a prediction tap; corresponding to a class of a target voice And a prediction step of obtaining a predicted value of the target voice using the tap coefficient to be performed.

A first program according to the present invention comprises a predictive tap extracting step of extracting, from a synthesized sound, a predictive tap used for predicting the target voice using a high-quality sound for which a predicted value is to be obtained as a target voice. A class tap extracting step of extracting, from a code, a class tap used to classify the target voice into one of several classes; and a class classification for obtaining a class of the target voice based on the class tap. A class classification step of performing, and an acquisition step of obtaining a tap coefficient corresponding to the class of the target voice from the tap coefficients for each class obtained by performing the learning, a prediction tap, and a class of the target voice. A prediction step of obtaining a predicted value of the target voice using a corresponding tap coefficient.

According to the first recording medium of the present invention, a high-quality sound for which a predicted value is to be obtained is regarded as a target sound, and a prediction tap used for predicting the target sound is extracted from a synthesized sound. A step, a class tap extracting step of extracting, from a code, a class tap used to classify the target voice into one of several classes, and a class for obtaining a class of the target voice based on the class tap. A classification step for performing classification;
An acquisition step of acquiring a tap coefficient corresponding to the class of the target voice from among the tap coefficients for each class obtained by performing the learning, a prediction tap, and a tap coefficient corresponding to the class of the target voice. And a prediction step of obtaining a predicted value of the target voice.

The learning apparatus of the present invention uses a high-quality sound for which a predicted value is to be obtained as a target voice, and uses a class tap used to classify the target voice into one of several classes. Class tap extracting means for extracting from a code, class classifying means for performing class classification for obtaining a class of a target voice based on the class tap,
Learning means for learning so as to statistically minimize a prediction error of a predicted value of high-quality sound obtained by performing a prediction operation using a tap coefficient and a synthesized sound, and obtaining a tap coefficient for each class; It is characterized by having.

According to the learning method of the present invention, a high-quality sound for which a predicted value is to be obtained is regarded as a target voice, and a class tap used for classifying the target voice into one of several classes is used. , A class tap extraction step of extracting from a code, a class classification step of performing a class classification for obtaining a class of a target voice based on a class tap, and a high sound quality obtained by performing a prediction operation using a tap coefficient and a synthesized sound And a learning step of learning and calculating a tap coefficient for each class so that the prediction error of the predicted value of the speech is statistically minimized.

According to a second program of the present invention, a high-quality sound for which a predicted value is to be obtained is regarded as a target sound, and a class used for classifying the target sound into one of several classes is provided. A tap is obtained by performing a class tap extraction step of extracting a tap from a code, a class classification step of performing a class classification for obtaining a class of a target voice based on the class tap, and performing a prediction operation using a tap coefficient and a synthesized sound. A learning step of performing learning so as to statistically minimize a prediction error of a predicted value of a high-quality sound, and obtaining a tap coefficient for each class.

The second recording medium of the present invention uses a high-quality sound for which a predicted value is to be obtained as a target voice, and uses the target voice to classify the target voice into one of several classes. A class tap is obtained by performing a class tap extracting step of extracting a class tap from a code, a class classification step of performing a class classification for obtaining a class of a target voice based on the class tap, and performing a prediction operation using tap coefficients and synthesized sounds. And a learning step of learning and calculating a tap coefficient for each class so that the prediction error of the predicted value of the high-quality sound is statistically minimized.

In the audio processing apparatus and the audio processing method, the first program and the first recording medium of the present invention, the high-quality sound for which the prediction value is to be obtained is set as the target sound, and the target sound is predicted. Is extracted from the synthesized sound, and the class tap used to classify the target voice into one of several classes is extracted from the code. Then, based on the class tap, a class classification for obtaining a class of the target voice is performed, and a predicted value of the target voice is obtained using the prediction tap and a tap coefficient corresponding to the class of the target voice.

In the learning apparatus and the learning method of the present invention, the second program and the second recording medium, a high-quality sound for which a predicted value is to be obtained is regarded as a noticed sound, and the noticed sound is divided into several sounds. A class tap used for classifying the class into one of the classes is extracted from the code, and a class classification for obtaining a class of the target voice is performed based on the class tap. Learning is performed so that the prediction error of the predicted value of the high-quality sound obtained by performing the prediction operation using the tap coefficient and the synthesized sound is statistically minimized, and the tap coefficient for each class is calculated. Can be

[0047]

FIG. 3 shows an example of the configuration of an embodiment of a speech synthesizer to which the present invention is applied.

This speech synthesizer is supplied with code data obtained by multiplexing a residual code and an A code obtained by encoding the residual signal and the linear prediction coefficient to be applied to the speech synthesis filter 44 by vector quantization or the like. The residual signal and the linear prediction coefficient are decoded from the residual code and the A code, respectively, and the decoded signal is supplied to the speech synthesis filter 44 to generate a synthesized sound.
Further, in this speech synthesizer, the speech synthesis filter 44
By performing a prediction operation using the synthesized sound generated in the above and the tap coefficient obtained by learning, a high-quality sound (synthesized sound) with improved sound quality of the synthesized sound is obtained and output. I have.

That is, in the speech synthesizer shown in FIG.
Utilizing the classification adaptive processing, the synthesized speech is decoded into (the predicted value of) true high-quality speech.

The class classification adaptation process includes a class classification process and an adaptation process. The class classification process classifies data into classes based on the nature of the data, and performs an adaptation process for each class. Is based on the following method.

That is, in the adaptive processing, for example,
By a linear combination with a predetermined tap coefficient, a predicted value of true high-quality sound is obtained.

More specifically, for example, a true high-quality sound (sample value thereof) is used as teacher data, and the true high-quality sound is converted into an L code, a G code, an I code by the CELP method. Code, and A code,
Using the synthesized sound obtained by decoding these codes by the receiving unit shown in FIG. 2 as student data, the predicted value E [y] of the high-quality sound y as the teacher data is converted into several synthesized sounds ( sample value) x _1, x ₂ of the considered a set of ..., predetermined tap coefficients w _1, w _2, that obtained by linear combination model defined by a linear combination of .... In this case, the predicted value E [y] can be expressed by the following equation.

E [y] = w ₁ x ₁ + w ₂ x ₂ +... (6)

To generalize equation (6), a matrix W composed of a set of tap coefficients w _j , a matrix X composed of a set of student data x _ij , and a matrix Y composed of a set of predicted values E [y _j ] '
To

(Equation 1) Defines the following observation equation.

XW = Y ′ (7) Here, the component x _ij of the matrix X is a set of i-th student data (a set of student data used for predicting the _i-th teacher data y _i ). Means the j-th student data in the matrix W, and the component w _{j of the} matrix W represents a tap coefficient by which a product with the j-th student data in the set of the student data is calculated. Also, y
_i represents the i-th teacher data, and therefore, E [y _i ]
Represents the predicted value of the i-th teacher data. Note that y on the left side of the equation (6) is obtained by omitting the suffix i of the component y _i of the matrix Y. Further, x ₁ , x ₂ ,. The suffix i of the component x _ij is omitted.

Then, it is considered that a least square method is applied to this observation equation to obtain a predicted value E [y] close to a true high-quality sound y. In this case, a matrix Y consisting of a set of true high-quality sound y serving as teacher data and a matrix E consisting of a set of residuals e of predicted values E [y] for high-quality sound y.
To

(Equation 2) From equation (7), the following residual equation is established.

XW = Y + E (8)

In this case, the tap coefficient w _j for obtaining the predicted value E [y] close to the true high-quality sound y is the square error

(Equation 3) Can be obtained by minimizing.

Therefore, the above square error is calculated by using the tap coefficient w _j
, The tap coefficient w _j that satisfies the following equation is equal to the predicted value E close to the true high-quality sound y.
This is an optimum value for obtaining [y].

[0060]

(Equation 4) ... (9)

Therefore, first, the equation (8) is changed to the tap coefficient w
By differentiating with _j , the following equation is established.

[0062]

(Equation 5) ... (10)

From equations (9) and (10), equation (11)
Is obtained.

[0064]

(Equation 6) ... (11)

Further, considering the relationship among the student data x _ij , the tap coefficient w _j , the teacher data y _i , and the error e _i in the residual equation of the equation (8), the following normal equation is obtained from the equation (11). Equation can be obtained.

[0066]

(Equation 7) ... (12)

The normal equation shown in equation (12) is
The matrix (covariance matrix) A and the vector v are

(Equation 8) If the vector W is defined as shown in Expression 1, it can be expressed by the following expression: AW = v (13)

Each normal equation in the equation (12) is prepared by preparing a certain number of sets of the student data x _ij and the teacher data y _i , and forming the same number as the number J of the tap coefficients w _{j to} be obtained. And therefore equation (1)
3) with respect to the vector W (however, equation (1)
To solve 3), the matrix A in equation (13) needs to be non-singular), and the optimal tap coefficient (here, the tap coefficient that minimizes the square error) w _j can be obtained. In solving equation (13), for example, a sweeping method (Gauss-Jordan elimination method) or the like can be used.

As described above, the optimum tap coefficient w _j
, And using the tap coefficient w _j , the prediction value E close to the true high-quality sound y is obtained by Expression (6).
Finding [y] is adaptive processing.

For example, an audio signal sampled at a high sampling frequency or an audio signal to which multiple bits are assigned is used as the teacher data, and the audio signal as the teacher data is thinned out as the student data, When the requantized audio signal is encoded by the CELP method and a synthesized sound obtained by decoding the encoding result is used, an audio signal sampled at a high sampling frequency or a multi-bit is assigned as a tap coefficient. When generating an audio signal, a high-quality audio with a statistically minimum prediction error can be obtained. Therefore, in this case, it is possible to obtain a synthesized sound with higher sound quality.

In the speech synthesizer shown in FIG. 3, the code data composed of the A code and the residual code is decoded into high-quality speech by the above-described class classification adaptive processing.

That is, the demultiplexer (DEMUX) 4
1 is supplied with code data,
The demultiplexer 41 separates the A code and the residual code for each frame from the code data supplied thereto. Then, the demultiplexer supplies the A code to the filter coefficient decoder 42 and the tap generation unit 46, and supplies the residual code to the residual code book storage unit 43 and the tap generation unit 46.

Here, the A code and the residual code included in the code data in FIG. 3 are obtained by converting a linear prediction coefficient and a residual signal obtained by LPC analysis of a speech into a vector quantum code using a predetermined code book. It is the code obtained by the conversion.

The filter coefficient decoder 42 decodes the A code for each frame supplied from the demultiplexer 41 into linear prediction coefficients based on the same codebook used when obtaining the A code. , To the speech synthesis filter 44.

The residual code book storage unit 43 stores the residual code for each frame supplied from the demultiplexer 41 on the basis of the same code book used when obtaining the residual code. The signal is decoded and supplied to the speech synthesis filter 44.

The speech synthesis filter 44 is, for example, an IIR type digital filter, similar to the speech synthesis filter 29 of FIG. 1, and uses the linear prediction coefficient from the filter coefficient decoder 42 as the tap coefficient of the IIR filter and the remaining. By using the residual signal from the difference codebook storage unit 43 as an input signal and filtering the input signal,
A synthesized sound is generated and supplied to the tap generation unit 45.

The tap generation section 45 includes the speech synthesis filter 4
From (the sample value of) the synthesized sound supplied from No. 4, a sample that becomes a prediction tap used in a prediction calculation in a prediction unit 49 described later is extracted. That is, the tap generation unit 45 sets, for example, all the sample values of the synthesized sound of the target frame, which is the frame for which the predicted value of the high-quality sound is to be obtained, as the prediction tap. Then, the tap generation unit 45
The prediction tap is supplied to the prediction unit 49.

The tap generator 46 is provided with the demultiplexer 4
A class tap is extracted from the A code and residual code for each frame (or subframe) supplied from 1. That is, the tap generation unit 46, for example,
All the A codes and residual codes of the frame of interest are class taps. Then, the tap generation unit 46 supplies the class tap to the class classification unit 47.

Here, the configuration patterns of the prediction taps and the class taps are not limited to those described above.

In the tap generation unit 46, in addition to the A code and the residual code, the linear prediction coefficient output from the filter coefficient decoder 42, the residual signal output from the residual code book storage unit 43, and The class tap can also be extracted from the synthesized sound output from the voice synthesis filter 44 or the like.

The class classification section 47 classifies (sample values of) the voice of the focused frame of interest on the basis of the class tap from the tap generation section 46, and classifies the class code corresponding to the class obtained as a result. Coefficient memory 48
Output to

Here, the class classifying section 47 includes, for example,
It is possible to output the A code of the frame of interest as a class tap and the bit sequence itself constituting the residual code as a class code.

The coefficient memory 48 stores tap coefficients for each class obtained by performing a learning process in a learning apparatus shown in FIG.
The tap coefficient stored in the address corresponding to the class code output from 7 is output to the prediction unit 49.

Here, assuming that high-quality sound of N samples is obtained for each frame, to obtain the sound of N samples for the frame of interest by the prediction calculation of equation (6), N sets of tap coefficients are used. is necessary.
Therefore, in this case, the coefficient memory 48 stores N sets of tap coefficients for addresses corresponding to one class code.

The prediction section 49 acquires the prediction tap output from the tap generation section 45 and the tap coefficient output from the coefficient memory 48, and uses the prediction tap and the tap coefficient to obtain
The linear prediction operation (product-sum operation) shown in Expression (6) is performed to obtain (predicted value of) high-quality sound of the frame of interest, and D /
Output to A conversion unit 50.

Here, as described above, the coefficient memory 48 outputs N sets of tap coefficients for obtaining each of the N samples of the audio of the frame of interest.
Is calculated using, for each sample value, a prediction tap and a set of tap coefficients corresponding to the sample value.
Is performed.

The D / A converter 50 D / A converts (predicted value of) the sound from the predictor 49 from a digital signal to an analog signal, and supplies the analog signal to the speaker 51 for output.

Next, FIG. 4 shows the speech synthesis filter 4 shown in FIG.
4 shows a configuration example.

In FIG. 4, the speech synthesis filter 44
The P-order linear prediction coefficient is used, and therefore, is composed of one adder 61, P delay circuits (D) 62 _{1 to} 62 _P , and P multipliers 63 _{1 to} 63 _P. ing.

The P-order linear prediction coefficients α ₁ , α ₂ ,..., Α _P supplied from the filter coefficient decoder 42 are set in the multipliers 63 _{1 to} 63 _P , respectively.
The speech synthesis filter 44 performs an operation according to equation (4) to generate a synthesized sound.

That is, the residual signal e output from the residual codebook storage unit 43 is supplied to the delay circuit 62 via the adder 61.
Is supplied to the 1, the delay circuit 62 _p is an input signal thereto, and only one sample delay of the residual signal, and outputs to the delay circuit 62 p _{+ 1} of the subsequent stage, and outputs to the calculator 63 _p . The multiplier 63 _p multiplies the output of the delay circuit 62 _p by the linear prediction coefficient α _p set therein, and outputs the multiplied value to the adder 61.

The adder 61 adds all the outputs of the multipliers 63 _{1 to} 63 _P and the residual signal e, and outputs the addition result as
In addition to supplying to the delay circuit 621, a speech synthesis result (synthesized sound)
Output as

Next, referring to the flowchart of FIG.
The processing (speech synthesis processing) of the speech synthesis device in FIG. 3 will be described.

The demultiplexer 41 sequentially separates the A code and the residual code for each frame from the code data supplied thereto, and separates them into filter coefficient decoders 4.
2 and the residual codebook storage unit 43. further,
The demultiplexer 41 supplies the A code and the residual code to the tap generator 46.

The filter coefficient decoder 42 sequentially decodes the A code for each frame supplied from the demultiplexer 41 into linear prediction coefficients, and supplies the linear prediction coefficients to the speech synthesis filter 44. The residual code book storage unit 43 sequentially decodes the residual code for each frame supplied from the demultiplexer 41 into a residual signal, and supplies the residual signal to the speech synthesis filter 44.

The speech synthesis filter 44 performs the operation of equation (4) using the residual signal and the linear prediction coefficient supplied thereto, thereby generating a synthesized sound of the frame of interest. This synthesized sound is supplied to the tap generation unit 45.

The tap generating section 45 sequentially sets the synthesized sound frames supplied thereto as frames of interest, and in step S 1, predicts taps based on (sample values of) the synthesized sounds supplied from the voice synthesis filter 44. It is generated and output to the prediction unit 49. Further, in step S <b> 1, the tap generation unit 46 outputs the signal A supplied from the demultiplexer 41.
Generate class taps from code and residual code,
Output to the class classification unit 47.

Then, the process proceeds to step S 2, where the class classification unit 47 classifies the class based on the class tap supplied from the tap generation unit 46, and supplies the resulting class code to the coefficient memory 48. , Step S
Proceed to 3.

In step S 3, the coefficient memory 48 reads the tap coefficient from the address corresponding to the class code supplied from the class classification section 47 and supplies the read tap coefficient to the prediction section 49.

Then, the process proceeds to a step S4, wherein the predicting section 49
Obtains the tap coefficient output from the coefficient memory 48, performs the product-sum operation shown in Expression (6) using the tap coefficient and the prediction tap from the tap generation unit 45, and obtains the high sound quality of the frame of interest. (Predicted value of). The high-quality sound is supplied to the speaker 51 from the prediction unit 49 via the D / A conversion unit 50, and is output.

After the predicting section 49 obtains the high-quality sound of the frame of interest, the process proceeds to step S5.
It is determined whether there is a frame to be processed as the frame of interest. If it is determined in step S5 that there is still a frame to be processed as a target frame, the process returns to step S1, and the same process is repeated with a frame to be set as the next target frame newly set as a target frame. If it is determined in step S5 that there is no frame to be processed as the frame of interest, the speech synthesis processing ends.

Next, FIG. 6 shows an example of the configuration of an embodiment of a learning apparatus for performing a learning process of tap coefficients stored in the coefficient memory 48 of FIG.

A learning digital audio signal is supplied to the learning device in a predetermined frame unit. The learning digital audio signal is supplied to the LPC analysis section 71 and the prediction filter 74. You. Further, the learning digital audio signal is also supplied to the normal equation adding circuit 81 as teacher data.

The LPC analysis section 71 sequentially sets the frame of the audio signal supplied thereto as a frame of interest, performs an LPC analysis of the audio signal of the frame of interest, obtains a P-order linear prediction coefficient, It is supplied to the conversion unit 72 and the prediction filter 74.

The vector quantization section 72 stores a code book in which a code vector having linear prediction coefficients as elements is associated with a code. Based on the code book, the LPC analysis section 71 calculates a target frame from the LPC analysis section 71. The feature vector composed of the linear prediction coefficients is vector-quantized, and the A code obtained as a result of the vector quantization is supplied to the filter coefficient decoder 73 and the tap generator 79.

The filter coefficient decoder 73 stores the same codebook as that stored in the vector quantization unit 72, and converts the A code from the vector quantization unit 72 based on the codebook. The signal is decoded into a linear prediction coefficient and supplied to the speech synthesis filter 77. Here, the filter coefficient decoder 42 in FIG. 3 has the same configuration as the filter coefficient decoder 73 in FIG.

The prediction filter 74 uses the audio signal of the frame of interest supplied thereto and the linear prediction coefficient from the LPC analysis unit 71 to perform, for example, an operation according to equation (1), thereby obtaining the frame of interest. Is obtained and supplied to the vector quantization unit 75.

[0108] That is, the Z-transform of s _n and e _n in the formula (1), expressed respectively S and E, equation (1) can be expressed by the following equation.

E = (1 + α ₁ z ⁻¹ + α ₂ z ⁻² +... + Α _P z ^−P ) S (14)

From equation (14), the prediction filter 74 for obtaining the residual signal e can be constituted by a FIR (Finite Impulse Response) type digital filter.

That is, FIG. 7 shows an example of the configuration of the prediction filter 74.

The prediction filter 74 includes an LPC analysis unit 71
, A P-order linear prediction coefficient is supplied. Accordingly, the prediction filter 74 includes P delay circuits (D) 91 _{1 to} 91 _P , P multipliers 92 _{1 to} 92 _P ,
And one adder 93.

Each of the multipliers 92 _{1 to} 92 _P has L
The P-order linear prediction coefficients α ₁ , α ₂ ,..., Α _P supplied from the PC analysis unit 71 are set.

[0114] On the other hand, the audio signal s of the frame of interest is supplied to the delay circuit 91 ₁ and the adder 93. Delay circuit 91
_p delays the input signal thereto by one sample of the residual signal, outputs the delayed signal to the delay circuit 91 _{p + 1} at the subsequent stage, and outputs it to the calculator 92 _p . The multiplier 92 _p multiplies the output of the delay circuit 91 _p by the linear prediction coefficient α _p set therein, and outputs the multiplied value to the adder 93.

The adder 93 adds all the outputs of the multipliers 92 _{1 to} 92 _P and the audio signal s, and
It is output as a residual signal e.

Returning to FIG. 6, the vector quantization unit 75 stores a code book in which a code is associated with a code vector having a sample value of a residual signal as an element, and a prediction is performed based on the code book. The residual vector composed of the sample value of the residual signal of the frame of interest from the filter 74 is vector-quantized, and the residual code obtained as a result of the vector quantization is stored in a residual codebook storage unit 76 and a tap generation unit 79. To supply.

[0117] The residual codebook storage unit 76 stores the same codebook as that stored by the vector quantization unit 75. Based on the codebook, the residual codebook storage unit 76 stores the residual codebook from the vector quantization unit 75. Decoding the code into a residual signal,
It is supplied to the speech synthesis filter 77. Here, the residual codebook storage unit 43 in FIG. 3 is configured similarly to the residual codebook storage unit 76 in FIG.

The speech synthesis filter 77 is an IIR filter constructed in the same manner as the speech synthesis filter 44 in FIG. 3. The linear prediction coefficient from the filter coefficient decoder 73 is used as the tap coefficient of the IIR filter, and the residual codebook is used. By using the residual signal from the storage unit 75 as an input signal and filtering the input signal, a synthesized sound is generated and supplied to the tap generation unit 78.

The tap generation unit 78 forms a prediction tap from the synthesized sound supplied from the speech synthesis filter 77 and supplies the prediction tap to the normal equation addition circuit 81, as in the case of the tap generation unit 45 in FIG. The tap generation unit 79 is configured as shown in FIG.
As in the case of the tap generation unit 46, a class tap is formed from the A code and the residual code supplied from the vector quantization units 72 and 75, respectively.
Supply 0.

The classifying section 80 classifies the class based on the class taps supplied thereto, as in the case of the classifying section 47 of FIG. 3, and classifies the resulting class code with a normal equation adding circuit. 81.

The normal equation addition circuit 81 includes a learning voice, which is a high-quality voice of the frame of interest as teacher data, and a prediction tap (tuple tap) as student data from the tap generator 78. Synthetic sound output)
Is added to the target.

That is, the normal equation adding circuit 81 uses the prediction tap (student data) for each class corresponding to the class code supplied from the class classification unit 80, and calculates the equation (1).
3) are each component in the matrix A,
Multiplication (x _in x _im ) of student data and calculation corresponding to summation (Σ) are performed.

Further, the normal equation adding circuit 81
Class code supplied from the class classification unit 80.
For each class that responds, the student data (
Of the synthesized sound output from the voice synthesis filter 77.
Pull value) and teacher data (high-quality sound of the frame of interest)
Using the voice sample value), the vector v in equation (13) is
Student data and teachers
Data multiplication (x_iny _i) And summation (Σ)
Perform the appropriate operation.

The normal equation addition circuit 81 performs the above-mentioned addition using all the frames of the learning speech supplied thereto as a frame of interest, whereby the normal equation shown in the equation (13) is obtained for each class. To build.

The tap coefficient determination circuit 82 obtains a tap coefficient for each class by solving the normal equation generated for each class in the normal equation addition circuit 81.
The coefficients are supplied to the addresses of the coefficient memory 83 corresponding to the respective classes.

Depending on the audio signal prepared as the audio signal for learning, the normal equation adding circuit 81
In some cases, a class may not be obtained in which the required number of normal equations for obtaining the tap coefficient is obtained. The tap coefficient determination circuit 82 outputs, for example, a default tap coefficient for such a class.

The coefficient memory 83 stores the tap coefficient determination circuit 8
The tap coefficient for each class supplied from 2 is stored in the address corresponding to the class.

Next, referring to the flowchart of FIG.
The processing (learning processing) of the learning device in FIG. 6 will be described.

The learning apparatus is supplied with a learning audio signal. The learning audio signal is supplied to the LPC analysis unit 71 and the prediction filter 74, and is supplied to the normal equation adding circuit 81 as teacher data. Is done. And
In step S11, student data is generated from the learning audio signal.

That is, the LPC analysis section 71 sequentially sets the frames of the audio signal for learning as a target frame, and performs an LPC analysis on the audio signal of the target frame to obtain a P-order linear prediction coefficient, and To the conversion unit 72.
The vector quantization unit 72 vector-quantizes the feature vector composed of the linear prediction coefficient of the frame of interest from the LPC analysis unit 71, and A is obtained as a result of the vector quantization.
The code is supplied to the filter coefficient decoder 73 and the tap generator 79. The filter coefficient decoder 73 decodes the A code from the vector quantization unit 72 into a linear prediction coefficient, and supplies the linear prediction coefficient to the speech synthesis filter 77.

On the other hand, the prediction filter 74 that has received the linear prediction coefficient of the frame of interest from the LPC analysis unit 71 uses the linear prediction coefficient and the speech signal for learning of the frame of interest and follows the equation (1). By performing the calculation, the residual signal of the frame of interest is obtained and supplied to the vector quantization unit 75. The vector quantization unit 75 includes a prediction filter 74
Is quantized, and the residual code obtained as a result of the vector quantization is supplied to the residual code book storage unit 76 and the tap generation unit 79. I do. The residual codebook storage unit 76 decodes the residual code from the vector quantization unit 75 into a residual signal, and supplies the residual signal to the speech synthesis filter 77.

As described above, the speech synthesis filter 77
Receives the linear prediction coefficient and the residual signal, performs speech synthesis using the linear prediction coefficient and the residual signal, and uses the resultant synthesized sound as student data as the tap generation unit 78.
Output to

Then, the process proceeds to step S12, where the tap generation section 78 generates a prediction tap from the synthesized sound supplied from the speech synthesis filter 77, and the tap generation section 79 outputs the A code from the vector quantization section 72. And a class tap from the residual code from the vector quantization unit 75. The prediction tap is supplied to a normal equation addition circuit 81, and the class tap is supplied to a class classification unit 80.

Thereafter, in step S13, the class classifying section 80 classifies the class based on the class tap from the tap generating section 79, and supplies the resulting class code to the normal equation adding circuit 81.

Then, the process proceeds to step S14, where the normal equation adding circuit 81 determines, for the class supplied from the classifying section 80, the sample value of the high-quality sound of the frame of interest as the teacher data supplied thereto, and the tap. Expression (13) for a prediction tap (sample value of a synthetic sound constituting the prediction tap) as student data from the generation unit 78
Is added to the matrix A and the vector v as described above, and the process proceeds to step S15.

In step S15, it is determined whether there is still a speech signal for learning a frame to be processed as the frame of interest. In step S15,
If it is determined that there is an audio signal for learning of a frame to be processed as the frame of interest, the process returns to step S11,
With the next frame as a new frame of interest, the same processing is repeated thereafter.

If it is determined in step S15 that there is no audio signal for learning the frame to be processed as the frame of interest, that is, if the normal equation adding circuit 81 obtains a normal equation for each class , To step S16, where the tap coefficient determination circuit 82
Solves the normal equation generated for each class, finds a tap coefficient for each class, supplies the tap coefficients to the address corresponding to each class in the coefficient memory 83, stores the tap coefficients, and ends the processing.

As described above, the tap coefficients for each class stored in the coefficient memory 83 are stored in the coefficient memory 48 of FIG.

Accordingly, the tap coefficients stored in the coefficient memory 48 of FIG. 3 are statistically different from the prediction error (here, the square error) of the prediction value of the high-quality sound obtained by performing the linear prediction operation. Since the sound is obtained by performing learning so as to minimize the sound, the sound output by the prediction unit 49 in FIG. 3 has the distortion of the synthesized sound generated by the sound synthesis filter 44 reduced (eliminated). It will be of high sound quality.

In the speech synthesizer shown in FIG. 3, as described above, for example, when the tap generation unit 46 is to cause the tap generation unit 46 to extract class taps from among linear prediction coefficients, residual signals, and the like, Also, the tap generation unit 79 in FIG. 6 also calculates the linear prediction coefficient output from the filter coefficient decoder 73 and the residual signal output from the residual codebook storage unit 76 from the residual signal.
It is necessary to extract similar class taps. However, when class taps are also extracted from the linear prediction coefficients and the like, the number of taps increases. Therefore, it is desirable to perform the class classification by compressing the class taps by, for example, vector quantization. When class classification is performed only from the residual code and the A code, the bit sequence of the residual code and the A code can be used as the class code as it is, so that the load required for the class classification processing is reduced. Can be.

Next, FIG. 9 shows a transmission system to which the present invention is applied (a system refers to a device in which a plurality of devices are logically aggregated, and it is determined whether or not the devices of each configuration are in the same housing. (Regardless of the present invention).

In this transmission system, the portable telephone 101
₁ and 101 _2, between the base station 102 ₁ and 102 _2, respectively, performs transmission and reception by radio, the base station 102 ₁
And 102 ₂ transmit and receive data to and from the exchange 103, and eventually, the mobile phones 101 ₁ and 101 1
In between the 01 _2, the base station 102 ₁ and 102 _2,
In addition, voice transmission and reception can be performed via the exchange 103. Note that base stations 102 ₁ and 1
02 ₂ may be the same base station, or may be a different base station.

Here, the mobile phones 101 ₁ and 101 ₂ will be referred to as the mobile phone 101 unless otherwise specified.
It is described.

FIG. 10 shows a configuration example of the mobile phone 101 shown in FIG.

The antenna 111 receives a radio wave from the base station 102 ₁ or 102 ₂ and supplies the received signal to the modulation / demodulation unit 112, and also transmits a signal from the modulation / demodulation unit 112 by radio wave to the base station 102 ₁ or 102 _2. It sends to 102 _2. The modulation and demodulation unit 112 demodulates the signal from the antenna 111 and supplies the resulting code data as described in FIG. Also, the modulation / demodulation unit 1
12 modulates the code data supplied from the transmission unit 113 as described with reference to FIG. 1, and supplies the resulting modulated signal to the antenna 111. The transmitting unit 113 is configured similarly to the transmitting unit shown in FIG. 1, encodes the user's voice input thereto into code data, and supplies the code data to the modem unit 112. The reception unit 114 receives the code data from the modulation / demodulation unit 112, decodes the code data, and decodes and outputs the same high-quality sound as in the speech synthesis device in FIG.

That is, FIG. 11 shows a configuration example of the receiving section 114 of FIG. In the figure, portions corresponding to those in FIG. 2 are denoted by the same reference numerals, and a description thereof will be omitted as appropriate below.

The synthesized sound output from the speech synthesis filter 29 is supplied to the tap generation section 121, and the tap generation section 121 determines, from the synthesized sound, what is to be a prediction tap (sample value). The extracted data is supplied to the prediction unit 125.

The L code, G code, I code, and A code for each frame (or subframe) output from the channel decoder 21 are supplied to the tap generation unit 122. Furthermore, tap generation unit 1
To 22, a residual signal is supplied from a calculator 28 and a linear prediction coefficient is supplied from a filter coefficient decoder 25. The tap generation unit 122 extracts what is to be a class tap from the L code, G code, I code, and A code supplied thereto, as well as the residual signal and the linear prediction coefficient. Supply.

The classifying section 123 includes the tap generating section 12
Classification is performed based on the class tap supplied from 2, and the class code as the classification result is
The coefficient is supplied to the coefficient memory 124.

Here, L code, G code, I code,
When a class tap is formed from the A and A codes, the residual signal and the linear prediction coefficient, and the class is classified based on the class tap, the number of classes obtained as a result of the class classification may be enormous. . Therefore, the classifying unit 123 uses, for example, an L code, a G code, an I code, and an A code, and a code obtained by performing vector quantization on a vector having elements of a residual signal and a linear prediction coefficient as a class classification result. It can be output.

The coefficient memory 124 stores tap coefficients for each class obtained by performing a learning process in the learning apparatus shown in FIG. 12, which will be described later, and stores addresses corresponding to the class codes output from the class classifying unit 123. Is supplied to the prediction unit 125.

The prediction section 125 acquires the prediction tap output from the tap generation section 121 and the tap coefficient output from the coefficient memory 124, as in the prediction section 49 of FIG. To perform the linear prediction operation shown in equation (6). Accordingly, the prediction unit 125 obtains (predicted value of) the high-quality sound of the frame of interest and calculates the D / A
It is supplied to the converter 30.

In the receiving section 114 configured as described above, basically, the same processing as the processing according to the flowchart shown in FIG. 5 is performed, so that the synthesized sound of high sound quality is decoded. Output as a result.

That is, the channel decoder 21 separates the L code, G code, I code, and A code from the code data supplied thereto, and separates them into the adaptive codebook storage unit 22, the gain decoder 23, It is supplied to the codebook storage unit 24 and the filter coefficient decoder 25. Further, the L code, the G code, the I code, and the A code are also supplied to the tap generation unit 122.

The adaptive codebook storage unit 22, the gain decoder 23, the excitation codebook storage unit 24, and the arithmetic units 26 to 28 include the adaptive codebook storage unit 9 shown in FIG.
The same processing as in the gain decoder 10, the excitation codebook storage unit 11, and the arithmetic units 12 to 14 is performed, whereby the L code, the G code, and the I code are decoded into the residual signal e. This residual signal is supplied to the speech synthesis filter 29 and the tap generator 122.

Further, the filter coefficient decoder 25 has the configuration shown in FIG.
As described above, the A code supplied thereto is decoded into linear prediction coefficients and supplied to the speech synthesis filter 29 and the tap generation unit 122. The speech synthesis filter 29 performs speech synthesis using the residual signal from the arithmetic unit 28 and the linear prediction coefficient from the filter coefficient decoder 25, and supplies the resultant synthesized sound to the tap generation unit 121.

The tap generation unit 121 sets the synthesized sound frame output from the voice synthesis filter 29 as the frame of interest,
In step S <b> 1, a prediction tap is generated from the synthesized sound of the frame of interest and supplied to the prediction unit 125. Further, in step S1, the tap generation unit 122 supplies the L code, the G code, the I code, and the A code supplied thereto.
A class tap is generated from the code, the residual signal and the linear prediction coefficient, and supplied to the classifying unit 123.

Then, the process proceeds to step S 2, where the class classification unit 123 performs a class classification based on the class tap supplied from the tap generation unit 122, and supplies the resulting class code to the coefficient memory 124. The process proceeds to step S3.

In step S3, the coefficient memory 124 stores
The tap coefficient is read from the address corresponding to the class code supplied from the class classification unit 123, and the prediction unit 12
5

Then, the process proceeds to a step S4, wherein the prediction section 12
5 obtains a tap coefficient output from the coefficient memory 124, performs a product-sum operation shown in Expression (6) using the tap coefficient and the prediction tap from the tap generation unit 121,
(Predicted value of) high-quality sound of the frame of interest.

The high-quality sound obtained as described above is supplied from the prediction unit 125 to the speaker 31 via the D / A conversion unit 30, and the high-quality sound is output from the speaker 31. Is output.

After the process in step S4, the process proceeds to step S5, where it is determined whether or not there is still a frame to be processed as the frame of interest. The same process is repeated hereafter with the target frame as a new frame of interest. If it is determined in step S5 that there is no frame to be processed as the frame of interest, the process ends.

FIG. 12 shows the coefficient memory 12 of FIG.
4 illustrates a configuration example of an embodiment of a learning device that performs a learning process of a tap coefficient to be stored in No. 4.

The microphone 201 to the code determination unit 215
The configuration is the same as that of the microphone 1 to the code determination unit 15 in FIG. The microphone 1 receives a learning audio signal. Therefore, the microphone 201 to the code determination unit 215 output the learning audio signal to the microphone 1.
The same processing as in the case of FIG. 1 is performed.

The tap generator 131 is supplied with the synthesized sound output from the voice synthesis filter 206 when the square error minimum judging section 208 judges that the square error has become minimum. Also, the tap generation unit 132 includes an L code, a G code, and an L code, which are output when the code determination unit 215 receives the determination signal from the square error minimum determination unit 208.
An I code and an A code are supplied. Further, the tap generation unit 132 outputs an element of a code vector (centroid vector) corresponding to the A code as a vector quantization result of the linear prediction coefficient obtained by the LPC analysis unit 204, which is output by the vector quantization unit 205. , And the residual signal output by the arithmetic unit 214 when the square error has been minimized by the square error minimum determination unit 208. Also, the normal equation addition circuit 134 outputs the audio output from the A / D conversion unit 202,
Supplied as teacher data.

[0166] The tap generation section 131 converts the synthesized sound output from the speech synthesis filter 206 into the tap generation section 1 shown in FIG.
The same prediction tap as that of 21 is configured, and as student data,
It is supplied to the normal equation adding circuit 134.

The tap generation section 132
11 from the L code, the G code, the I code, and the A code, the linear prediction coefficient supplied from the vector quantization unit 205, and the residual signal supplied from the arithmetic unit 214. The same class tap as that of the generation unit 122 is configured and supplied to the class classification unit 133.

The classifying section 133 includes the tap generating section 13
Based on the class tap from No. 2, the same class classification as in the class classification unit 123 in FIG. 11 is performed, and the resulting class code is converted into a normal equation addition circuit 13.
4

The normal equation adding circuit 134 receives the audio from the A / D converter 202 as teacher data, receives the predicted tap from the tap generator 131 as student data, and receives the teacher data and student data. For each class, the same normal addition as in the normal equation addition circuit 81 of FIG. 6 is performed for each class code from the class classification unit 133, so that the normal equation shown in Expression (13) is obtained for each class. Te

The tap coefficient determination circuit 135 obtains tap coefficients for each class by solving the normal equation generated for each class in the normal equation adding circuit 134, and stores the tap coefficients in an address corresponding to each class in the coefficient memory 136. Supply.

Depending on the audio signal prepared as the audio signal for learning, the normal equation adding circuit 134 may have a class in which the number of normal equations required for obtaining the tap coefficients cannot be obtained. For such a class, the tap coefficient determination circuit 135 outputs, for example, a default tap coefficient.

The coefficient memory 136 stores the linear prediction coefficients for each class and the tap coefficients for the residual signal supplied from the tap coefficient determination circuit 135.

In the learning apparatus configured as described above, basically, the same processing as the processing according to the flowchart shown in FIG. 8 is performed, so that the tap coefficient for obtaining a high-quality synthesized sound is obtained. Is required.

A learning audio signal is supplied to the learning device. In step S11, the learning audio signal is
Teacher data and student data are generated.

That is, the audio signal for learning is transmitted to the microphone 201
And the microphone 201 through the code determination unit 215
The same processing as in the case of the microphone 1 to the code determination unit 15 in FIG. 1 is performed.

As a result, the audio of the digital signal obtained by the A / D converter 202 is supplied to the normal equation adding circuit 134 as teacher data. Further, when the squared error minimum determination unit 208 determines that the squared error is minimized, the synthesized sound output from the voice synthesis filter 206 is supplied to the tap generation unit 131 as student data.

Further, the linear prediction coefficient output from the vector quantization section 205 and the L code, G code, and L code output by the code determination section 215 when the square error minimum determination section 208 determines that the square error has become minimum. The I code, the A code, and the residual signal output from the arithmetic unit 214 are supplied to the tap generation unit 132.

Thereafter, the flow proceeds to step S12, where the tap generation section 131 sets a frame of the synthesized sound supplied as the student data from the voice synthesis filter 206 as a frame of interest, and generates a prediction tap from the synthesized sound of the frame of interest. It is supplied to the normal equation adding circuit 134. Further, in step S12, the tap generation unit 132 generates a class tap from the L code, G code, I code, A code, linear prediction coefficient, and residual signal supplied thereto, and the class classification unit 133 Supply.

After the processing in step S12, step S1
3, the classifying unit 133 sets the tap generating unit 132
Classification is performed based on the class tap from
34.

Then, the process proceeds to a step S 14, where the normal equation adding circuit 134 outputs the learning voice, which is the high-quality voice of the target frame as the teacher data from the A / D converter 202, and the learning voice from the tap generator 132. For the prediction tap as the student data, the above-described addition of the matrix A and the vector v of Expression (13) is performed for each class code from the class classification unit 133, and step S1 is performed.
Go to 5.

In step S15, it is determined whether or not there is still a frame to be processed as the frame of interest. If it is determined in step S15 that there is still a frame to be processed as the target frame, the process returns to step S11, and the same process is repeated with the next frame as a new target frame.

If it is determined in step S15 that there is no frame to be processed as the frame of interest, that is, if the normal equation adding circuit 134 obtains a normal equation for each class, step S1
Proceeding to 6, the tap coefficient determination circuit 135 obtains a tap coefficient for each class by solving the normal equation generated for each class, and supplies the tap coefficient to the address of the coefficient memory 136 corresponding to each class. Then, the process is terminated.

As described above, the tap coefficients for each class stored in the coefficient memory 136 are stored in the coefficient memory 124 of FIG.

Accordingly, the tap coefficients stored in the coefficient memory 124 of FIG. 11 are obtained by calculating the prediction error (square error) of the high-quality sound predicted value obtained by performing the linear prediction operation.
Since the sound is obtained by performing learning so as to be statistically minimized, the sound output by the prediction unit 125 in FIG. 11 has high sound quality.

Next, the above-described series of processing can be performed by hardware or can be performed by software. When a series of processing is performed by software, a program constituting the software is
Installed on a general-purpose computer.

FIG. 13 shows an example of the configuration of an embodiment of a computer in which a program for executing the above-described series of processing is installed.

The program is stored in a hard disk 305 or a ROM 3 as a recording medium built in the computer.
03 can be recorded in advance.

Alternatively, the program includes a flexible disk, a CD-ROM (Compact Disc Read Only Memory),
MO (Magneto optical) disc, DVD (Digital Versatile)
Disc), a magnetic disk, a semiconductor memory or the like, and can be temporarily (permanently) stored (recorded) in a removable recording medium 311. Such a removable recording medium 311 can be provided as so-called package software.

The program can be installed in the computer from the removable recording medium 311 as described above, can be wirelessly transferred from a download site to the computer via an artificial satellite for digital satellite broadcasting, or can be connected to a LAN (Local Area). Network) or the Internet, and the program can be transferred to the computer by wire. The computer can receive the transferred program by the communication unit 308 and install the program on the built-in hard disk 305.

The computer has a CPU (Central Processing).
Unit 302). The CPU 302 has a bus 3
01 is connected to the input / output interface 310 via the input / output interface 310, and the user operates the input unit 307 including a keyboard, a mouse, a microphone, and the like via the input / output interface 310. When a command is input, the ROM (Read O
nly Memory) 303 is executed. Alternatively, the CPU 302 may execute a program stored in the hard disk 305, a program transferred from a satellite or a network and received by the communication unit 308 and installed in the hard disk 305, or a removable recording medium 311 mounted in the drive 309. The program read and installed on the hard disk 305 is stored in a RAM (Random Access Memory).
y) Load into 304 and execute. Thereby, the CPU 30
2 performs processing according to the above-described flowchart or processing performed by the configuration of the above-described block diagram. Then, the CPU 302 transmits the processing result as necessary, for example, via the input / output interface 310.
Output is performed from an output unit 306 including an LCD (Liquid Crystal Display), a speaker, or the like, or transmitted from the communication unit 308, and further recorded on the hard disk 305.

Here, in this specification, processing steps for writing a program for causing a computer to perform various processing do not necessarily have to be processed in a time series in the order described in the flowchart, and may be performed in parallel. Alternatively, it also includes processing executed individually (for example, parallel processing or processing by an object).

The program may be processed by one computer, or may be processed in a distributed manner by a plurality of computers. Further, the program may be transferred to a remote computer and executed.

In the present embodiment, what kind of signal is used as the learning speech signal is not particularly mentioned. Alternatively, for example, a song (music) or the like can be adopted. According to the above-described learning process, when a human utterance is used as the learning voice signal, a tap coefficient that improves the sound quality of the voice of such a human utterance is obtained, When a song is used, a tap coefficient that improves the sound quality of the song is obtained.

In the embodiment of FIG. 11, the tap coefficients are stored in advance in the coefficient memory 124. However, the tap coefficients stored in the coefficient memory 124 are the same as those in FIG. Base station 102
(Or the exchange 103) or a WWW (World
It can be downloaded from a Wide Web server or the like. That is, as described above, a tap coefficient suitable for a certain type of audio signal, such as for a human utterance or music, can be obtained by learning. Further, depending on teacher data and student data used for learning, it is possible to obtain a tap coefficient that causes a difference in sound quality of a synthesized sound. Therefore, such various tap coefficients are assigned to the base station 1
02 and the like, and the user can download the tap coefficient desired by the user. Such tap coefficient download service can be performed free of charge or can be performed for a fee. Furthermore, when the tap coefficient download service is performed for a fee, the price for the tap coefficient download can be charged together with, for example, the telephone charge of the mobile phone 101.

Further, the coefficient memory 124 stores the data of the portable telephone 1
01 can be configured with a memory card or the like that is removable. In this case, if a different memory card storing the above-described various tap coefficients is provided, the user can replace the memory card storing the desired tap coefficient with a mobile phone as necessary. It becomes possible to use it by attaching it to 101.

Further, the present invention relates to, for example, VSELP (V
ector Sum Excited Liner Prediction), PSI-CE
LP (Pitch Synchronous Innovation CELP), CS-A
C such as CELP (Conjugate Structure Algebraic CELP)
The present invention is widely applicable to a case where a synthesized sound is generated from a code obtained as a result of encoding by the ELP method.

The present invention is not limited to the case where a synthesized speech is generated from a code obtained as a result of encoding according to the CELP method, but generates a synthesized speech by obtaining a residual signal and a linear prediction coefficient from a certain code. Widely applicable in cases.

Further, in the present embodiment, the prediction values of the residual signal and the linear prediction coefficient are obtained by the linear primary prediction operation using the tap coefficients. Can be obtained by a higher-order prediction calculation of

Also, for example, in the embodiments of FIGS. 11 and 12, class taps are designated by L code, G code, I code
In addition to the code and the A code, the linear prediction coefficient obtained from the A code and the residual signal obtained from the L code, the G code, and the I code are generated based on the
The class tap can also be generated from, for example, only the L code, the G code, the I code, and the A code. Also, the class tap can be generated from only one (or a plurality) of the four types of L code, G code, I code, and A code, that is, for example, from only the I code. For example, a class tap
In the case where only the I code is used, the I code itself can be used as the class code. here,
In the VSELP method, 9 bits are assigned to the I code. Therefore, when the I code is used as a class code as it is, the number of classes is 512 (= 2 ⁹ ). In the VSELP method, each bit of the 9-bit I code has two kinds of code polarities of 1 or -1. Therefore, when such an I code is used as a class code, for example, -1 The bit that is
What should be considered as.

Furthermore, in the CELP system, code data may include list interpolation bits and frame energy. In this case, class taps can be configured using soft interpolation bits and frame energy. .

For example, Japanese Patent Application Laid-Open No. Hei 8-202399 discloses a method of improving the sound quality of a synthesized sound by passing the sound through a high-frequency emphasizing filter. It differs from the invention described in Japanese Patent Application Laid-Open No. 8-202339 in that the points obtained by learning and the tap coefficients to be used are determined by the result of class classification by codes.

[0202]

According to the audio processing apparatus and the audio processing method, the first program and the first recording medium of the present invention, the high-quality sound for which the prediction value is to be obtained is regarded as the target sound, The prediction tap used to predict is extracted from the synthesized speech, and the target voice is
The class taps used to classify into any of several classes are extracted from the code. Then, based on the class tap, a class classification for obtaining a class of the target voice is performed, and a predicted value of the target voice is obtained using the prediction tap and a tap coefficient corresponding to the class of the target voice. Therefore, it is possible to generate a high-quality synthesized sound.

According to the learning apparatus and the learning method of the present invention, the second program and the second recording medium, the high-quality sound for which the prediction value is to be obtained is regarded as the noticed sound, Class taps used for classifying into any of the classes are extracted from the code, and the class classification for finding the class of the target voice is performed based on the class tap. Learning is performed so that the prediction error of the predicted value of the high-quality sound obtained by performing the prediction operation using the tap coefficient and the synthesized sound is statistically minimized, and the tap coefficient for each class is calculated. Can be Therefore, it is possible to generate a high-quality synthesized sound by the tap coefficient.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an example of a transmission unit of a conventional mobile phone.

FIG. 2 is a block diagram illustrating a configuration of an example of a receiving unit of a conventional mobile phone.

FIG. 3 is a block diagram illustrating a configuration example of an embodiment of a speech synthesis device to which the present invention has been applied;

4 is a block diagram illustrating a configuration example of a speech synthesis filter 44. FIG.

FIG. 5 is a flowchart illustrating a process of the speech synthesizer of FIG. 3;

FIG. 6 is a block diagram illustrating a configuration example of an embodiment of a learning device to which the present invention has been applied.

FIG. 7 is a block diagram illustrating a configuration example of a prediction filter 74.

FIG. 8 is a flowchart illustrating a process of the learning device in FIG. 6;

FIG. 9 is a diagram illustrating a configuration example of an embodiment of a transmission system to which the present invention has been applied.

FIG. 10 is a block diagram illustrating a configuration example of a mobile phone 101.

11 is a block diagram illustrating a configuration example of a receiving unit 114. FIG.

FIG. 12 is a block diagram illustrating a configuration example of another embodiment of a learning device to which the present invention has been applied.

FIG. 13 is a block diagram illustrating a configuration example of a computer according to an embodiment of the present invention.

[Explanation of symbols]

Reference Signs List 21 channel decoder, 22 adaptive codebook storage unit, 23 gain decoder, 24 excitation codebook storage unit, 25 filter coefficient decoder, 26 to 2
8 arithmetic unit, 29 speech synthesis filter, 30 D /
A conversion unit, 31 speaker, 41 demultiplexer, 42 filter coefficient decoder, 43 residual codebook storage unit, 44 speech synthesis filter, 45, 46
Tap generation unit, 47 class classification unit, 48 coefficient memory, 49 prediction unit, 50 D / A conversion unit, 51
Speaker, 61 adder, 62 _{1 to} 62 _P delay circuit, 63 _{1 to} 63 _P multiplier, 71 LPC analysis unit, 72 vector quantization unit, 73 filter coefficient decoder, 74 prediction filter, 75 vector quantization unit, 76 Residual codebook storage unit, 77 speech synthesis filter, 78, 79 tap generation unit, 80
Classification unit, 81 Normal equation addition circuit, 82
Tap coefficient determination circuit, 83 coefficient memory, 91 _{1 to} 91 _P delay circuit, 92 _{1 to} 72 _P multiplier, 93
Adder, 101 ₁ , 101 ₂ mobile phone, 10
2 ₁ , 102 ₂ base station, 103 exchange, 111
Antenna, 112 modulator / demodulator, 113 transmitter,
114 receiving unit, 121, 122 tap generating unit,
123 Classifier, 124 Coefficient memory, 125
Prediction unit, 131, 132 tap generation unit, 133
Classifier, 134 Normal equation adder, 13
5 tap coefficient determination circuit, 136 coefficient memory, 2
01 microphone, 202 A / D conversion unit, 203 arithmetic unit, 204 LPC analysis unit, 205 vector quantization unit, 206 speech synthesis filter, 207 square error calculation unit, 208 minimum square error determination unit, 209 adaptive codebook storage unit, 210 gain decoder, 2
11 excitation codebook storage unit, 212 to 214
Arithmetic unit, 215 code decision unit, 301 bus,
302 CPU, 303 ROM, 304 RAM, 30
5 hard disk, 306 output unit, 307 input unit, 308 communication unit, 309 drive, 310
I / O interface, 311 Removable recording medium

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Masaaki Hattori 6-7-35 Kita-Shinagawa, Shinagawa-ku, Tokyo Inside Sony Corporation (72) Inventor Hiroto Kimura 6-35-35 Kita-Shinagawa, Shinagawa-ku, Tokyo Inside Sony Corporation (72) Inventor Yasuhiro Fujimori 6-35 Kitashinagawa, Shinagawa-ku, Tokyo Sony Corporation F-term (reference) 5D045 CA01 5J064 AA01 BB01 BB03 BC01 BC06 BC09 BC12 BD02

Claims (17)

[Claims] 1. A predicted value of a high-quality sound having improved sound quality is predicted from a synthesized sound obtained by applying a linear prediction coefficient and a residual signal generated from a predetermined code to a voice synthesis filter. A prediction tap for extracting the prediction value for the high-quality sound by performing a predetermined prediction operation using the prediction tap and a predetermined tap coefficient. The high-quality sound to be obtained is regarded as a target sound, the prediction tap used for predicting the target sound is extracted from the synthetic sound, a prediction tap extracting unit, and the target sound is classified into several classes. A class tap extracting means for extracting, from the code, a class tap used to classify the class into one of the classes; and a class of the target voice based on the class tap. Classifying means for performing class classification for obtaining, from among the tap coefficients for each class obtained by performing learning, obtaining means for obtaining the tap coefficient corresponding to the class of the target voice, A data processing apparatus comprising: a prediction tap; and a prediction unit that calculates a predicted value of the target voice using the tap coefficient corresponding to the class of the target voice. 2. The data processing according to claim 1, wherein the prediction unit obtains a predicted value of the target voice by performing a linear primary prediction operation using the prediction tap and the tap coefficient. apparatus. 3. The apparatus according to claim 1, wherein the obtaining unit obtains the tap coefficients of a class corresponding to the target voice from a storage unit that stores the tap coefficients for each class. Data processing device. 4. The method according to claim 1, wherein the class tap extracting unit extracts the class tap from the code and the linear prediction coefficient or the residual signal obtained by decoding the code. 2. The data processing device according to 1. 5. The tap coefficient is such that a prediction error of a prediction value of the high-quality sound obtained by performing a predetermined prediction operation using the prediction tap and the tap coefficient is statistically minimized. The data processing apparatus according to claim 1, wherein the data processing apparatus is obtained by performing learning. 6. The data processing apparatus according to claim 1, further comprising the speech synthesis filter. 7. The code according to claim 1, wherein the code is a CELP (Code
2. The data processing apparatus according to claim 1, wherein the data is obtained by encoding according to an Excited Liner Prediction coding) method. 8. A predicted value of a high-quality sound having improved sound quality is predicted from a synthesized sound obtained by applying a linear prediction coefficient and a residual signal generated from a predetermined code to a sound synthesis filter. A prediction tap for extracting a prediction value of the high-quality sound by performing a predetermined prediction operation using the prediction tap and a predetermined tap coefficient. A prediction tap extraction step of extracting, from the synthesized sound, the prediction tap used for predicting the high-quality sound to be obtained as the high-quality sound as the high-quality sound; A class tap extracting step of extracting, from the code, a class tap used to classify the class into one of the class taps; A class classification step of performing a class classification for obtaining a class of; and an obtaining step of obtaining the tap coefficient corresponding to the class of the target voice from among the tap coefficients for each class, obtained by performing learning. A data processing method comprising: a prediction step of obtaining a predicted value of the target voice using the prediction tap and the tap coefficient corresponding to the class of the target voice. 9. Predicting a predicted value of a high-quality sound with improved sound quality from a synthesized sound obtained by applying a linear prediction coefficient and a residual signal generated from a predetermined code to a sound synthesis filter. A prediction tap for extracting the predicted value of the high-quality sound by performing a predetermined prediction operation using the predicted tap and a predetermined tap coefficient. A prediction tap extraction step of extracting, from the synthesized sound, the prediction tap used to predict the high-quality sound for which the high-quality sound for which the prediction value is to be obtained is used as the target voice; and Extracting, from the code, a class tap used to classify the class into any one of several classes; A class classification step of performing a class classification for obtaining the class of the target voice based on the tap, and the tap coefficient corresponding to the class of the target voice from the tap coefficients for each class obtained by performing learning. A program comprising: an acquisition step of acquiring a tap coefficient; and a prediction step of obtaining a predicted value of the target voice using the prediction tap and the tap coefficient corresponding to the class of the target voice. 10. A predicted value of a high-quality sound with improved sound quality is predicted from a synthesized sound obtained by applying a linear prediction coefficient and a residual signal generated from a predetermined code to a sound synthesis filter. Extract prediction taps for
A recording medium storing a program for causing a computer to perform a sound process for obtaining a predicted value of the high-quality sound by performing a predetermined prediction operation using the prediction tap and a predetermined tap coefficient. The high-quality sound for which the predicted value is to be obtained as the target voice, the prediction tap used to predict the target voice, a prediction tap extraction step of extracting from the synthesized sound, the target voice, A class tap extracting step of extracting, from the code, a class tap used for classifying the class tap into any one of several classes; and a class for performing a class classification for obtaining the class of the target voice based on the class tap. Classifying step, from among the tap coefficients for each class obtained by performing learning, A program comprising: an acquisition step of acquiring the tap coefficient corresponding to a class; a prediction step of obtaining the predicted value of the target voice using the prediction tap and the tap coefficient corresponding to the class of the target voice. A recording medium characterized by being recorded. 11. A predictive value of a high-quality sound having improved sound quality is determined from a synthesized sound obtained by applying a linear prediction coefficient and a residual signal generated from a predetermined code to a voice synthesis filter. A learning device for learning a predetermined tap coefficient used for obtaining by the prediction operation of the above, wherein the high-quality sound for which the predicted value is to be obtained is a target voice, and the target voice is selected from among several classes. A class tap extracting unit for extracting a class tap used for classifying into any one of the following from the code: a class tap unit for performing a class classification for obtaining a class of the target voice based on the class tap; As the prediction error of the predicted value of the high-quality sound obtained by performing the prediction operation using the coefficient and the synthesized sound is statistically minimized, A learning unit that performs learning and obtains a tap coefficient for each class. 12. The learning unit according to claim 1, wherein a prediction error of a predicted value of the high-quality sound obtained by performing a linear primary prediction operation using the tap coefficient and the synthesized sound is statistically minimized. The learning device according to claim 11, wherein learning is performed. 13. The method according to claim 12, wherein the class tap extracting unit extracts the class tap from the code and the linear prediction coefficient or the residual signal obtained by decoding the code. 12. The learning device according to 11. 14. The code according to claim 1, wherein said code is a CELP (Cod
12. The learning device according to claim 11, wherein the learning device is obtained by performing encoding using an e Excited Liner Prediction coding) method. 15. A synthesized speech obtained by applying a linear prediction coefficient and a residual signal generated from a predetermined code to a speech synthesis filter to obtain a predicted value of a high-quality sound whose sound quality has been improved by a predetermined value. A learning method for learning a predetermined tap coefficient used to obtain by the prediction operation of the above, wherein the high-quality sound for which the predicted value is to be obtained is a target voice, and the target voice is selected from several classes. A class tap extraction step of extracting, from the code, a class tap used for classifying into any one of the following: a class tapping step of performing a class classification for obtaining a class of the target voice based on the class tap; The prediction error of the predicted value of the high-quality sound obtained by performing the prediction operation using the coefficient and the synthesized sound is statistically minimized. A learning step of performing learning and obtaining a tap coefficient for each class. 16. A synthesized speech obtained by applying a linear prediction coefficient and a residual signal generated from a predetermined code to a speech synthesis filter, and obtains a predicted value of a high-quality sound having an improved sound quality by a predetermined value. A learning process for learning a predetermined tap coefficient used to obtain by the prediction calculation of the above, a program that causes a computer to perform the high-quality sound for which the predicted value is to be obtained as a target voice, A class tap extracting step of extracting, from the code, a class tap used for classifying the class into one of several classes; and performing a class classification for obtaining the class of the target voice based on the class tap. Classifying step; and performing a prediction operation using the tap coefficient and the synthesized sound. A learning step of performing learning so that a prediction error of a predicted value is statistically minimized, and calculating a tap coefficient for each class. 17. A synthesized speech obtained by applying a linear prediction coefficient and a residual signal generated from a predetermined code to a speech synthesis filter to obtain a predicted value of a high-quality sound whose sound quality has been improved by a predetermined value. On a recording medium in which a program for causing a computer to perform a learning process of learning a predetermined tap coefficient used for obtaining a predicted value is obtained by paying attention to the high-quality sound for which the predicted value is to be obtained. A class tap extraction step of extracting, from the code, a class tap used to classify the target voice into any of several classes as the voice, based on the class tap, A class classification step of performing a class classification for obtaining a class; and performing a prediction operation using the tap coefficient and the synthesized sound. And a learning step of learning and calculating a tap coefficient for each class so that a prediction error of a predicted value of the high-quality sound is statistically minimized. recoding media.