JP3186013B2 - Acoustic signal conversion encoding method and decoding method thereof - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention converts an acoustic signal such as a music signal or a voice signal into a signal in a frequency domain,
A high-efficiency audio signal encoding method for normalizing by the square root of the power spectrum envelope to obtain a residual signal, quantizing the residual signal, and quantizing information for obtaining the power spectrum envelope as auxiliary information; It relates to the decoding method.

[0002]

2. Description of the Related Art As a method for encoding an audio signal with high efficiency, an original sound signal is divided into frames, which are called frames, at a fixed interval of about 5 to 50 ms, and MDCT (Modified Discrete Cosine Transform) is applied to one frame signal. Is used to perform time-frequency conversion to obtain a frequency domain signal, which is divided into two information, a power spectrum envelope (envelope of frequency characteristics) and a residual signal obtained by flattening the frequency domain signal with the power spectrum envelope. And encoding each of them.

[0003] The proposed encoding method and its decoding method will be described with reference to FIG. The audio input signal sequence digitized from the input terminal 33 in the encoder 31 is input to the frame dividing means 34, and the input sequence of the past 2 × N samples is extracted for every N input samples, and the length is 2
XN samples are generated in the input frame and the windowing means 35
A time window is applied to the input frame. The window shape is generally a Hanning window. The windowed input signal sequence is subjected to a modified discrete cosine transform by the MDCT means 36 to be converted into a frequency domain signal of N samples.

[0004] The windowed input signal sequence is subjected to linear prediction analysis by a linear prediction analysis means 37, and a P-order prediction coefficient is obtained. This linear prediction analysis is performed after obtaining the autocorrelation. The prediction coefficient is quantized by the quantization means 38. As the quantization method, an LSP quantization method of converting a prediction coefficient into an LSP parameter and quantizing it, or a method of converting a prediction coefficient into a k parameter and then quantizing it can be used. An index 39 indicating the quantized prediction coefficient is transmitted.

The quantized prediction coefficients are used to calculate a power spectrum by a frequency outline calculating means 41 to obtain a frequency characteristic outline signal. Specifically, the quantized output of the quantizing means 38 is inversely quantized, and for example, as shown in FIG. 8, after the P + 1 inversely quantized prediction coefficients (α parameters), 2
An FFT analysis is performed on a 2 × N sample sequence formed by connecting × NP−1 zeros (fast Fourier transform: discrete Fourier transform), and the Nth-order power spectrum is calculated. Each point of the reciprocal of the i-th frequency characteristic outline starting from the 0-th is obtained by averaging the square roots of the i + 1-th and i-th power spectra except for i = N−1, that is, by interpolating. The reciprocal of the (N-1) th approximate frequency characteristic is N-1
Obtained by taking the square root of the power spectrum.

Returning to FIG. 7, in the normalizing means 42, each sample of the frequency domain signal from the MDCT means 36 is multiplied by each sample of the reciprocal of the above-mentioned frequency outline to normalize and flatten. The residual signal. The residual signal is normalized by the power normalizing / gain quantizing means 43 by dividing by an average value of the amplitude or a normalized gain which is a square root of the average value of the power to obtain a normalized residual signal. Further, the normalized gain is quantized, and an index 44 indicating the quantized normalized gain is output.

[0007] The signal of the reciprocal of the approximate frequency characteristic from the approximate frequency calculation means 41 is weighted by the weight calculation means 45 if necessary.
Is subjected to audibility control to obtain a weighted signal. In the normalized residual quantization means 46, the normalized residual signal from the means 43 is adaptively weighted vector-quantized by the weight signal from the means 45. An index 47 indicating the vector value quantized by the quantization means 46 is output. As described above, from the encoder 31, the prediction coefficient quantization index 39,
The gain quantization index 44 and the residual quantization index 47 are output.

The decoder 32 to which these indices 39, 44 and 47 are input decodes as follows, as shown in FIG. That is, the prediction coefficient quantization index 39 is reproduced by dequantizing the corresponding quantized prediction coefficient by the reproducing means 56, and the inverse quantized prediction coefficient is obtained by the frequency rough shape calculating means 57 in the same manner as the frequency rough shape calculating means 41. Calculates the reciprocal of the general frequency characteristic, that is, the reciprocal of the square root of the power spectrum envelope.
7, a quantized normalized residual signal is reproduced. Reproduction means 5
The normalized gain is reproduced from the index 44 input at 9. The quantized normalized residual signal reproduced by the power denormalization means 61 is multiplied by the reproduced normalization gain to perform power denormalization to obtain a quantized residual signal. The quantized residual signal is inversely flattened by the inverse normalizing means 62 for each corresponding sample by the inverse of the frequency outline, that is, the inverse of the square root of the power spectrum envelope from the frequency outline calculating means 57. The inverse-flattened residual signal is subjected to an Nth-order inversely modified discrete cosine transform by an inverse MDCT unit 63 to obtain a time-domain signal, and a time window is applied to the time-domain signal by a windowing unit 64. Here, a Hanning window is used as the window shape. The windowed signal is added to the first half N samples of the frame of length 2 × N samples and the second half N samples of the previous frame by the frame superimposing means 65 and output to the output terminal 66.

[0009]

The encoding shown in FIG.
In the decoding method, the decoder 32 obtains the inverse quantized prediction coefficient from the index 39, obtains the power spectrum as shown in FIG. 8, obtains the square root of each sample, and calculates the reciprocal thereof. , Respectively, but the square root operation for each sample requires a considerable amount of processing, which hinders real-time operation. In addition, the hardware scale becomes large and expensive. For example, when decoding an encoded music signal, it is not preferable that the processing time is long as described above because it is not suitable for practical use, and the hardware scale is large and a large number of expensive decoders are spread. .

Further, a high calculation accuracy is required to directly obtain the power spectrum envelope from the inversely quantized prediction coefficients, which is an obstacle to realizing a decoder having a short calculation word length. SUMMARY OF THE INVENTION It is an object of the present invention to provide a transform coding method that eliminates the need for a decoder to calculate the square root of the power spectrum envelope for each sample. It is another object of the present invention to provide a transform decoding method that does not require a square root operation for each sample.

Still another object of the present invention is to provide a conversion decoding method having a short operation word length as well as a square root operation for each sample.

[0012]

According to the encoding method of the present invention, an envelope representing a square root of a power spectrum envelope is modeled by linear prediction analysis for encoding a spectrum envelope during transform encoding. , Quantize the linear prediction parameters, and use it as coding information to obtain a square root power spectrum envelope, dequantize the quantized parameters, and calculate the reciprocal of the square root power spectrum envelope from the dequantized parameters. This is used for normalization of the signal converted to the frequency domain.

That is, modeling the envelope representing the square root of the power spectrum envelope by linear predictive analysis involves, for example, obtaining the autocorrelation of the input audio signal, performing a Fourier transform on the autocorrelation, and calculating the square root of the power of each sample of the conversion result. , And the series of square roots is subjected to inverse Fourier transform, and linear prediction analysis is performed on the result of the conversion, that is, the square root of the power spectrum. Since the quantized output of the linear prediction analysis parameter is input to the decoding side, the square root power spectrum envelope or its reciprocal can be obtained without performing the square root operation in the decoder.

In order to model the square root of the power spectrum envelope by linear prediction analysis, the input acoustic signal is subjected to linear prediction analysis, an LPC cepstrum is obtained from the linear prediction parameters, and each coefficient of the LPC cepstrum is divided by two. A linear prediction parameter is obtained from the LPC cepstrum divided by 2 by the least square method (claim 2). In order to obtain the reciprocal of the square root power spectrum envelope from the inverse quantized linear prediction parameter, a linear prediction coefficient (α parameter) is obtained from the inversely quantized linear prediction parameter, and the linear prediction coefficient is subjected to a Fourier transform. An absolute value is obtained (claim 3).

To find the reciprocal of the square root power spectrum envelope from the inversely quantized linear prediction coefficient, an LSP parameter is obtained from the inversely quantized linear prediction parameter, and the LSP parameter is obtained.
A spectrum is calculated on the unit circle for the parameter (claim 4). In the above modeling, the input acoustic signal is subjected to linear prediction analysis, and LPC is performed based on the parameters of the linear prediction analysis result.
A cepstrum is obtained, and each coefficient of the LPC cepstrum is divided by 2. The LPC cepstrum divided by the 2 is quantized and output as a code (index) for obtaining a square root of a spectrum envelope. In order to obtain a square root, the quantized LPC cepstrum is inversely quantized, Fourier-transformed, and each Fourier-transformed sample is exponentially transformed (claim 5).

In the modeling according to claim 5, the division by 2 is omitted, and instead, each sample of the inversely quantized LPC cepstrum obtained by Fourier transform is divided by 2.
Exponential conversion may be performed (claim 6). According to the decoding method of the seventh aspect, a linear prediction coefficient or a parameter equivalent thereto (for example, LSP) is inversely quantized from the input code,
From this, the reciprocal of the square root power spectrum envelope is obtained, and the residual signal obtained by the inverse quantization of the input code is divided by the reciprocal of the square root power spectrum envelope to obtain a denormalized frequency domain signal. Here, the square root power spectrum envelope is a sequence in the frequency domain corresponding to the square root for each sample of the power spectrum envelope of the original system. The reciprocal of the square root power spectrum envelope is obtained by obtaining a linear prediction coefficient (α parameter) from the reproduced linear prediction parameter, performing a Fourier transform on the linear prediction coefficient, and taking an absolute value of each sample (claim 8). . Alternatively, an LSP parameter is obtained from the reproduced linear prediction parameter, and a spectrum is calculated on the unit circle for the LSP parameter (claim 9).

According to the decoding method of the tenth aspect, a linear prediction coefficient or a parameter equivalent thereto (for example, LSP) is inversely quantized from the input code, a corresponding impulse response is calculated from this, and the impulse response is calculated. Fourier transform is performed to obtain a square root power spectrum envelope, and the square root power spectrum envelope is multiplied by a residual signal obtained by inverse quantization of an input code to obtain a denormalized frequency domain signal.

In the decoding method according to the eleventh aspect, the square root power spectrum envelope is obtained from an LPC cepstrum from a reconstructed linear prediction parameter or a parameter equivalent thereto.
The LPC cepstrum is Fourier-transformed, and each Fourier-transformed sample is obtained by exponential conversion. Claim 1
In the decoding method of 2, the square root power spectrum envelope is
The LPC cepstrum is obtained from the reconstructed linear prediction parameter or a parameter equivalent thereto, the LPC cepstrum is subjected to Fourier transform, and each sample of the Fourier transform is halved and then subjected to exponential transform.

[0019]

FIG. 1 shows an embodiment of the invention of claim 1, and FIG.
The same reference numerals are given to portions corresponding to. In this embodiment, the output signal sequence of the windowing means 35 is branched and supplied to means 71 for modeling the envelope representing the square root of the power spectrum envelope by linear prediction analysis. The means 71 first obtains the autocorrelation function of the input signal up to N points in one frame by the correlation function means 72, for example. Instead, as shown in FIG. 2A, the auto-correlation function is performed by the correlation function means 7 up to P + 1 points.
2a, a P-order linear prediction analysis is performed on the result by the linear prediction analysis means 72b, and the obtained linear prediction coefficient is calculated as a correlation function from the P + 2 order to the Nth point with respect to the autocorrelation function up to the P + 1 point. The extrapolation means 72c may extrapolate. Next, this autocorrelation function in N points is added to this series by N
Add a zero to the point or symmetrically substitute the N point to get 2N
The real Fourier transform of the point is performed by the Fourier transform means 73. If the correlation function is obtained by the autocorrelation method, the real part after the conversion is the power spectrum, and all take positive values except for the error in the calculation accuracy. It is well known that a power spectrum is obtained by obtaining the autocorrelation of an input signal and performing Fourier transform on the autocorrelation.

The square root of each point of the power spectrum is obtained by a square root means 74. At this time, after all the imaginary parts are set to zero (or the imaginary parts are all zero when the symmetry is substituted), the inverse Fourier transform is performed by the inverse Fourier transform means 75 to obtain an autocorrelation function corresponding to the square root power spectrum. Finally, a linear prediction analysis is performed by the linear prediction analysis means 76 based on the autocorrelation function to obtain prediction parameters,
That is, a model obtained by linear prediction analysis representing the square root of the power spectrum envelope is obtained. This is quantized by the prediction coefficient quantization means 38 to obtain an index 39. The index 39 is not obtained by the linear prediction analysis of the input signal sequence, but is obtained by quantizing a frequency characteristic of the input signal sequence, that is, a signal obtained by performing a linear prediction analysis on the signal sequence corresponding to the square root of the power spectrum envelope.

The quantized output of the quantizing means 38 is inversely quantized by an inverse quantizing means 77, and the inversely quantized linear prediction coefficient is subjected to a Fourier transform and an absolute value means 78 to perform a Fourier transform.
The absolute value of the complex number of each sample is taken to obtain the reciprocal of the square root power spectrum envelope, and the frequency domain signal from the MDCT means 36 is multiplied by the multiplier 42 for each sample for normalization. Other processes are the same as those in FIG.

As shown in FIG. 2B, the power spectrum obtained by the Fourier transform means 73 of the correlation function is converted to the square root means 74.
May take the square root. Further, as shown in FIG. 2C, the autocorrelation function of the input signal sequence at point P is obtained by means 72a,
This may be subjected to linear prediction analysis by means 72b, and an impulse response corresponding to the analysis result obtained by means 72d to obtain an N-point autocorrelation function. That is, the autocorrelation function at N points can be obtained by obtaining the impulse response of the filter using the linear prediction analysis result of the means 72b as the filter coefficient. The Fourier transform of the linear prediction coefficient results in the reciprocal of the power spectrum envelope. The impulse response obtained by the means 72d is Fourier transformed by the means 73 to obtain an absolute value to obtain a power spectrum envelope. Instead of the Fourier transform means 73, a correlation function of the impulse response obtained by the means 72d may be obtained, and the correlation function may be Fourier-transformed to obtain a power spectrum envelope.

In any case, the order of the correlation function means 72 is increased by, for example, about 50% with respect to the order of the linear prediction analysis by the means 76, and a correlation function as long as possible is obtained without depending on the frequency component. Faithful information is input to the Fourier transform means 73. FIG. 3A shows still another method of the modeling method in the linear prediction of the square root power spectrum envelope (Claim 2). In this example, L
A linear prediction coefficient corresponding to the square root spectrum is obtained via the PC cepstrum. That is, the autocorrelation function of the input sound signal is obtained by the correlation function means 72a,
Perform linear prediction analysis with b. An LPC cepstrum is obtained from the linear prediction parameters by the recurrence formula calculation means 91.
That is, the calculation procedure for calculating the j-th order LPC cepstrum coefficient c _j from the prediction parameter α _k (k = 1,..., P) in the p-order linear prediction of the input acoustic signal is known as follows. Usually, p is about 10 to 20, but the order n of the LPC cepstrum needs to be 2 to 3 times p.

[0024]

(Equation 1)

The square root of the power spectrum in the region of the LPC cepstrum simply multiplies all the coefficients by 0.5, and the LPC cepstrum from the means 91 is divided by the means 92
, The square root of the power spectrum envelope is obtained, and this is calculated by the normal equation calculation means 93 to obtain a linear prediction parameter, and further, if necessary, quantized by the LSP parameter in the quantization means 38 and index 39
Output as Means 93 may use the above relationship between the LPC cepstrum and the linear prediction parameter in reverse,
The number of LPC cepstrum coefficients is determined by the linear prediction parameter α
, There is generally no linear prediction parameter that satisfies all LPC cepstrum constraints. Therefore, the above relationship is regarded as a regression equation, and a linear prediction parameter is determined so as to minimize the square of the regression error e _i of the LPC cepstrum. In this case, since the stability of the obtained linear prediction parameter is not guaranteed, for example, PA
A stability check such as conversion into RCOR coefficients is necessary. The relationship between the linear prediction parameters and the LPC cepstrum is represented by a matrix as follows.

[0026]

(Equation 2)

[0027] In the above relation, it is possible to obtain a linear prediction parameters by solving the following normal equation by means 93 in order to minimize the regression error energy d = E ^T E of the LPC cepstrum. D ^T DA = −D ^T C In the above description, a linear prediction parameter is obtained as a quantization output of a linear prediction model of the square root of the power spectrum envelope,
Alternatively, although the LSP parameter or the like is used, the output of the 1/2 means 92 may be quantized by the quantization means 38 and output as an index 39 as shown in FIG. 3B. In this case, 1 which is inversely quantized by the inverse quantization means 77 on the encoder side is used.
The / 2 LPC cepstrum is Fourier-transformed by means 78 into a logarithmic power spectrum envelope, and further exponentially transformed by means 94 to obtain a square root power spectrum envelope, and the reciprocal is taken by means 95 to the multiplier 42 of FIG. Output.

Further, as shown in FIG. 3C, the means 92 may be omitted, and each sample of the logarithmic power spectrum envelope may be supplied from the Fourier transform means 78 to the exponential transform means 94 by halving the means 96. As described above, by using, as the index 39, a value obtained by performing linear prediction analysis on the square root of the power spectrum envelope (square root power spectrum) and quantizing it, the processing in the decoder is performed as shown in FIG. Is also easier.

FIG. 4 shows an embodiment of the decoding method according to the seventh aspect of the present invention.
The same reference numerals are given to the portions corresponding to those in FIG.
In this embodiment as well, the index 39 is inversely quantized by the reproducing means 56 to obtain a linear prediction coefficient. The linear prediction coefficient is obtained by performing linear prediction analysis on the square root power spectrum, as is apparent from the description of FIG. Is Fourier-transformed by the Fourier transform means 82, and the reciprocal of the square root power spectrum envelope is obtained by obtaining its absolute value. The reciprocal of the square root power spectrum envelope is used to generate the residual signal reproduced from the multiplier 61. The frequency domain signal is reproduced in the divider 62. Others are the same as in FIG.

FIG. 5A shows an embodiment of the tenth aspect of the present invention, in which parts corresponding to those in FIG. 7 are denoted by the same reference numerals. In this embodiment, the index 39 obtained in FIG. 1 is inversely quantized by the reproducing means 56, the impulse response of the inversely quantized linear prediction coefficient is obtained by the means 84, and the impulse response is Fourier transformed by the Fourier transforming means 85. Is transformed, its absolute value is taken and the square root power spectrum envelope is obtained,
This is multiplied by the residual signal from the multiplier 61 by the multiplier 86 to reproduce the frequency domain signal.

The Fourier transform means 82 performs a Fourier transform by adding zero to the inverse quantized linear prediction coefficient. It is convenient that the quantization by the quantization means 38 is performed by converting the linear prediction coefficients into LSP coefficients. At this point, an index indicating an LSP coefficient may be used as the index 39. In this case, instead of the Fourier transform means 82 and 85 in the decoder, the value of the spectrum is calculated on the unit circle to obtain the power transfer function (power spectrum). This calculation method is, for example, 19
This is shown on page 91 of "Digital Audio Processing" by Sadahiro Furui, published by Tokai University Press in 1985.

FIG. 5B shows a part of a decoding method corresponding to the encoding method of FIG. 3B. In this case, the prediction parameter reproducing means 56 is a Ｌ LPC cepstrum, which is converted into a logarithmic power spectrum envelope by a Fourier transform means 82, which is exponentially converted for each sample by means 97 to obtain a square root power spectrum envelope. This is supplied to the multiplier 86. In the decoding method corresponding to the encoding method shown in FIG. 3C, as shown in FIG. 6A, the logarithmic power spectrum envelope from the Fourier transform means 82 in FIG.
It is sufficient to supply the data to the exponential conversion means 97 after halving with 8.

Instead of obtaining the impulse response in FIG. 5A, as shown in FIG. 6B, the P-order linear prediction coefficient reproduced from the reproducing means 56 is converted into a PARIOR coefficient, and a correlation function up to the P-order is obtained. At 99, the P-th order or higher may be extrapolated by the recurrence formula and supplied to the Fourier transform means 85. In the above description, the linear prediction parameter means a linear prediction coefficient (α parameter), an LSP parameter, a Percoll coefficient, and the like. Therefore, the prediction index 39 is quantized in any of these various forms. Also in the encoding method and the decoding method, the prediction index 39 may be inversely quantized, and the inversely quantized one may be used as it is, or may be converted to another linear prediction parameter and used.

In the above description, the correlation function means 72 (72a)
As an alternative, the absolute value of the output of the MDCT means 36, that is, the frequency domain signal may be obtained, and this may be subjected to inverse Fourier transform to obtain the autocorrelation function. Further, the conversion from the time domain to the frequency domain may be performed by another method such as a discrete Fourier transform (fast Fourier transform), a discrete cosine transform, or the like.

[0035]

As described above, according to the encoding method of the present invention, the envelope of the square root power spectrum of the frequency domain signal is modeled by the linear prediction parameter, and the linear prediction parameter is quantized to produce an encoded output. Because
As shown in the decoding method of the present invention, the square root operation is not required for decoding, and the processing of the decoder is reduced.

Further, when calculating a power spectrum envelope composed of linear prediction parameters and parameters equivalent thereto, generally high calculation accuracy is required. However, the variation of the square root power spectrum is smaller than that of a normal power spectrum. Less accuracy is required. For the same reason, the square root power spectrum can reduce the number of bits when quantizing linear prediction parameters from the power spectrum, particularly when vector quantization is used.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of an encoder to which an embodiment of the present invention is applied.

FIG. 2 is a block diagram showing various other methods of the linear prediction modeling unit 71 of the square root power spectrum envelope in FIG.

FIG. 3A is a block diagram showing still another example of the modeling unit 71, and FIGS. 3B and 3C are diagrams showing still another example and a unit for obtaining the reciprocal of the square root spectral envelope for normalization.

FIG. 4 is a block diagram showing an example of a decoder to which the embodiment of the invention of claim 6 is applied.

FIG. 5A is a block diagram showing an example of a decoder to which the invention of claim 8 is applied, and FIG. 5B is a block diagram showing a part of another decoding method.

FIG. 6 is a block diagram showing another example of decoding a square root power spectrum envelope.

FIG. 7 is a block diagram showing a previously proposed acoustic signal transform coding / decoding device.

FIG. 8 is a graph showing the powers obtained by Fourier transform from linear prediction coefficients .
The figure which shows a mode that a vector envelope value is calculated | required.

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁷ Identification code Fig10L 9/18 E

Claims (12)

(57) [Claims] 1. An input audio signal is converted from a time domain to a signal in a frequency domain, a square root of a power spectrum envelope of the converted frequency domain signal is obtained, and the square root of the power spectrum envelope is normalized with the frequency domain signal. In the audio signal transform encoding method for encoding the residual signal and the signal for obtaining the power spectrum envelope, respectively, a process of modeling an envelope representing a square root of the power spectrum envelope by linear predictive analysis, A process of quantizing the linear prediction parameters to obtain coding information for obtaining the power spectrum envelope; a process of inversely quantizing the quantized linear prediction parameters; and a process of calculating the power from the dequantized linear prediction parameters. Obtaining the reciprocal of the square root of the spectral envelope, and using the reciprocal of the square root, and Acoustic signal transform coding method that. 2. The step of modeling the square root of the power spectrum envelope by linear prediction analysis comprises the steps of: performing a linear prediction analysis of the input audio signal; and obtaining an LPC cepstrum from the linear prediction parameters obtained thereby. 2. The audio signal transform coding according to claim 1, further comprising a step of dividing each coefficient of the LPC cepstrum by 2, and a step of obtaining a linear prediction parameter from the LPC cepstrum divided by 2 by a least square method. Method. 3. The step of obtaining the reciprocal of the square root of the power spectrum envelope from the inversely quantized linear prediction parameters includes performing a Fourier transform on the linear prediction coefficients obtained from the inversely quantized linear prediction parameters, and calculating an absolute value of the linear prediction coefficient. 3. The acoustic signal conversion encoding method according to claim 1, further comprising: 4. The step of obtaining the reciprocal of the square root of the power spectrum envelope from the inversely quantized linear prediction parameters includes the step of obtaining L based on the inversely quantized linear prediction parameters.
3. The acoustic signal conversion method according to claim 1, wherein a step of obtaining a spectrum on a unit circle for the SP parameter is performed. 5. The step of modeling the envelope representing the square root of the power spectrum envelope by linear predictive analysis comprises the steps of linear predictive analysis of the input audio signal and LPC based on linear predictive parameters obtained by the linear predictive analysis. A process of obtaining a cepstrum, a process of dividing each coefficient of the obtained LPC cepstrum by 2, and a process of dividing the same by 2 as encoding information for quantizing the LPC cepstrum to obtain the power spectrum envelope. The process of obtaining the reciprocal of the square root power spectrum envelope from the inverse quantized linear prediction parameters includes a process of inversely quantizing the quantized LPC cepstrum, a process of performing a Fourier transform on the dequantized LPC cepstrum, and a process of performing a Fourier transform thereof. Converting the converted output to exponential conversion for each sample. 2. The acoustic signal conversion encoding method according to claim 1. 6. The step of modeling the envelope representing the square root of the power spectrum envelope by linear predictive analysis, the step of performing linear predictive analysis on the input audio signal and the LPC cepstrum based on linear predictive parameters obtained by the linear predictive analysis. The process of quantizing the LPC cepstrum as coding information for obtaining the power spectrum envelope, and calculating the reciprocal of the square root power spectrum envelope from the inverse quantized linear prediction parameter is performed by the quantization LPC Dequantizing the cepstrum, Fourier transforming the dequantized LPC cepstrum, dividing the Fourier transform output by 2 for each sample, and exponentially transforming each value divided by 2 2. The method according to claim 1, further comprising the steps of: 7. A residual signal is obtained by dequantizing a first index in an input code, a square root of a power spectrum envelope is obtained from a second index in the input code, and the residual signal is denormalized to obtain a frequency. An audio signal conversion decoding method for obtaining a domain signal and converting the frequency domain signal into a time domain signal to obtain an audio signal, wherein a square root power spectrum envelope is obtained from the linear prediction parameter indicated by the second index or a parameter equivalent thereto. Obtaining the reciprocal of the above, and performing the inverse normalization by dividing the residual signal by the square root power spectrum envelope. 8. The process of obtaining the reciprocal of the square root power spectrum envelope is a process of obtaining a linear prediction coefficient from the second index, performing a Fourier transform on the linear prediction coefficient, and obtaining an absolute value of each sample. The method for transforming and decoding an audio signal according to claim 7, characterized in that: 9. The method according to claim 7, wherein the step of obtaining the reciprocal of the square root power spectrum envelope is a step of obtaining an LSP parameter from the second index and obtaining a spectrum on the unit circle for the LSP parameter. Audio signal conversion decoding method. 10. A residual signal is obtained by dequantizing a first index in an input code, a square root of a power spectrum envelope is obtained from a second index in the input code, and the residual signal is denormalized to obtain a frequency. A sound signal conversion decoding method for obtaining a domain signal and converting the frequency domain signal into a time domain signal to obtain a sound signal, wherein a corresponding impulse response is obtained from a linear prediction parameter indicated by the second index or a parameter equivalent thereto. Calculating the Fourier transform of the calculated impulse response to obtain a square root power spectrum, and multiplying the residual signal by the square root power spectrum to perform the denormalization. Characteristic audio signal conversion decoding method. 11. A residual signal is obtained by dequantizing a first index in an input code, a square root of a power spectrum envelope is obtained from a second index in the input code, and the residual signal is denormalized to obtain a frequency. In an audio signal conversion decoding method for obtaining an area signal and converting the frequency domain signal into a time domain signal to obtain an audio signal, an LPC cepstrum is obtained from a linear prediction parameter indicated by the second index or a parameter equivalent thereto. A Fourier transform of the LPC cepstrum, an exponential transform of each of the Fourier transformed samples to obtain an envelope of the square root power spectrum, and a step of multiplying the residual signal by the square root power spectrum. Performing the inverse normalization. 12. A residual signal is obtained by dequantizing a first index in an input code, a square root of a power spectrum envelope is obtained from a second index in the input code, and the residual signal is denormalized. In an audio signal transform decoding method for obtaining a frequency domain signal and converting the frequency domain signal into a time domain signal to obtain an audio signal, an LPC cepstrum is obtained from a linear prediction parameter indicated by the second index or a parameter equivalent thereto. Obtaining, Fourier-transforming the LPC cepstrum, dividing each of the Fourier-transformed samples by 2, exponentially transforming each of the samples divided by 2 to obtain an envelope of the square root power spectrum. And a step of multiplying the residual signal by the square root power spectrum to perform the denormalization. Acoustic signal conversion decoding method comprising.