JP2013057792A - Speech coding device and speech coding method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To achieve an efficient use of information source codes at the time of coding time-series signals to suppress a calculation amount and improve coding performance in frequency-domain signals.SOLUTION: A first adjustment unit 102 performs a nonlinear amplitude adjustment to a first frequency spectrum. A CELP coding unit 104 generates a decoded signal by performing a CELP coding and a subsequent decoding. A second MDCT unit 105 subjects the decoded signal to an orthogonal transformation to generate a second frequency spectrum. A second adjustment unit 106 gives characteristics opposite to those which have been given to the first frequency spectrum during the amplitude adjustment in the first adjustment unit 102, to the second frequency spectrum. A subtraction unit 107 subtracts the second frequency spectrum to which the opposite characteristics have been given, from the first frequency spectrum to generate a residual spectrum. An FPC coding unit 108 codes the residual spectrum into an FPC code.

Description

  The present invention relates to a speech coding apparatus and speech coding method using a scalable codec technology having a multilayer structure.

  In mobile communication, in order to make effective use of a transmission band, compression encoding of voice or image digital information is essential. In particular, there is a great expectation for speech codec technology widely used in mobile phones, and there is a growing demand for encoding with higher sound quality than conventional high-efficiency encoding with a high compression rate. In addition, since the voice codec is used by the public, standardization is indispensable, and in accordance with this, research and development are actively performed in various companies around the world.

  In recent years, standardization of a scalable codec having a multi-layer structure has been studied by ITU-T (International Telecommunication Union-Telecommunication Standardization Sector) or MPEG (Moving Picture Experts Group). 718 is recommended (see Non-Patent Document 1). In this method, encoding is performed using CELP (Code Excited Linear Prediction) from the first layer to the second layer, and transform encoding using MDCT (Modified Discrete Cosine Transform) is performed on and after the third layer. Also, the transform coding is characterized in that a method called FPC (Factorial Pulse Coding) is used in order to efficiently encode a music signal.

  A speech coding technique whose performance has been greatly improved by the basic method “CELP” that models a speech utterance mechanism established 20 years ago and skillfully applies vector quantization is an algebraic technique described in Non-Patent Document 2. By inventing a fixed sound source with a small number of pulses such as the codebook (Algebraic Codebook), its performance has been further improved.

  G. In 718, the coding residual signal of the layer using CELP is transform-coded by FPC. In this case, there is a tendency that overall performance is improved by attenuating the CELP decoded signal. In 718, the attenuation coefficient is transmitted by 2 bits using this tendency.

  Here, although CELP encoding is performed so as to minimize encoding distortion when encoding the target signal, the CELP decoding signal is attenuated by the FPC method rather than CELP encoding. The overall encoding performance after conversion is improved. This clearly shows the difference in coding capability between CELP and FPC.

  That is, in CELP, encoding is performed locally with high accuracy on the time axis, but in the frequency spectrum (sometimes simply referred to as spectrum), errors are averaged on all frequencies. On the other hand, in FPC, since the frequency spectrum is encoded with a small number of pulses, it is encoded with high accuracy in some spectra, but the error is not reduced in other spectra. Therefore, considering that the coding distortion is uniformly on the frequency spectrum of the synthesized sound of CELP, the power of the CELP decoded signal is reduced rather than the decoding signal of CELP. When the spectrum encoded with the FPC is combined, there are statistically many cases where the coding distortion is totally reduced.

  The fact that there is a statistically large number of cases where coding distortion is reduced overall by attenuating the CELP decoded signal means that it may be better not to attenuate. Therefore, G. In 718, the FPC encoder is moved to obtain an attenuation coefficient that minimizes the coding distortion overall among the four attenuation coefficients.

  Conventionally, there is known a technique that collects a spectrum in a portion where FPC can be easily encoded by noise shaping that controls the amount of encoding distortion (for example, Patent Document 1). Patent Document 1 discloses a process called TNS (Temporal Noise Shaping) used in the standard coding system AAC (Advanced Audio Coding). In Patent Document 1, the MDCT coefficient, which is a frequency axis signal, is regarded as a time axis signal, and noise is concentrated at a large amplitude on the time axis by passing the MDCT coefficient through an LPC (Linear Prediction Coding) filter. Improve the sound quality of signals containing low pitch frequencies, such as male speech.

JP 2002-23798 A

Recommendation ITU-T G.718, 2008/6 Salami, Laflamme, Adoul, “8kbit / s ACELP Coding of Speech with 10ms Speech-Frame: aCandidate for CCITT Standardization”, IEEE Proc. ICASSP94, pp.II-97-II-100

  However, in the conventional apparatus, since the coding distortion is reduced by reducing the power of the spectrum of the decoded signal of CELP together with the coding distortion, there is a problem that the information source code of CELP coding is wasted. . Further, the conventional apparatus has a problem that a large amount of calculation is required because the encoder is moved a plurality of times for all four attenuation coefficients. In addition, although Patent Document 1 discloses noise shaping in spectrum coding, there is no description of noise shaping in time axis coding such as CELP, which has an effect of improving coding performance by FPC. There is a problem that how noise shaping is performed in CELP has not been clarified.

  An object of the present invention is to avoid wasting an information source code when encoding a time-series signal, to reduce the amount of calculation, and to improve encoding performance in a frequency domain signal. To provide a speech encoding apparatus and speech encoding method that can be improved.

  The speech encoding apparatus according to the present invention includes a first orthogonal transform unit that performs orthogonal transform on a target signal to be encoded to generate a first frequency spectrum, and a first non-linear function with respect to the first frequency spectrum. A first adjusting means for performing amplitude adjustment; an inverse orthogonal transforming means for generating a time signal by performing inverse transform of the orthogonal transform on the first frequency spectrum subjected to the first amplitude adjustment; and the time signal. First encoding means for decoding after decoding to generate a decoded signal, second orthogonal transform means for performing orthogonal transformation on the decoded signal to generate a second frequency spectrum, and the first From the second adjustment means for giving the second frequency spectrum a characteristic opposite to the characteristic given to the first frequency spectrum during amplitude adjustment, and the first frequency spectrum generated by the first orthogonal transform means, Subtracting means for generating a residual spectrum by subtracting the second frequency spectrum to which the reverse characteristic is given by the second adjusting means; and second encoding means for transform-coding the residual spectrum. Take the configuration.

  The speech coding method according to the present invention includes a first orthogonal transform step of performing orthogonal transform on a target signal to be encoded to generate a first frequency spectrum, and non-linear amplitude adjustment with respect to the first frequency spectrum. A first adjustment step to be performed; an inverse orthogonal transformation step for performing an inverse transformation of the orthogonal transformation on the first frequency spectrum whose amplitude has been adjusted in the first adjustment step to generate a time signal; and encoding the time signal A first encoding step that generates a decoded signal after decoding, a second orthogonal transformation step that performs orthogonal transformation on the decoded signal to generate a second frequency spectrum, and the first adjustment step. A second adjustment step for giving the second frequency spectrum a characteristic opposite to the characteristic given to the first frequency spectrum; and the first orthogonal transformation step. A subtracting step for generating a residual spectrum by subtracting the second frequency spectrum to which the reverse characteristic is given by the second adjusting step from the generated first frequency spectrum, and transform encoding the residual spectrum A second encoding step.

  According to the present invention, it is possible to eliminate the waste of an information source code when encoding a time-series signal, it is possible to reduce the amount of calculation, and to improve the encoding performance of a frequency domain signal. Can be improved.

The block diagram which shows the structure of the audio | voice coding apparatus which concerns on embodiment of this invention. The block diagram which shows the structural example 1 of the 1st adjustment part in embodiment of this invention. The block diagram which shows the structural example 2 of the 1st adjustment part in embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

(Embodiment)
<Configuration of speech encoding apparatus>
FIG. 1 is a block diagram showing a configuration of speech encoding apparatus 100 according to an embodiment of the present invention. In FIG. 1, CELP is used as an example of time-series encoding, and FPC is used as an example of subsequent transform encoding.

  The first MDCT unit 101 obtains a first frequency spectrum by performing orthogonal transform by MDCT on the input target signal (target vector) to be encoded. Specifically, the first MDCT unit 101 obtains the first frequency spectrum by multiplying the target signal by a sine window having a length twice as long as the subframe length and by multiplying the target signal by an MDCT matrix. The target signal is typically a voice signal, but may be a music signal or an acoustic signal including both a voice signal and a music signal, or before being input to the first MDCT unit 101. These signals may be signals obtained by performing some kind of preprocessing.

  The first MDCT unit 101 outputs the acquired first frequency spectrum to the first adjustment unit 102 and the subtraction unit 107. As for DCT (Discrete Cosine Transform), various methods for reducing the amount of calculation have been studied, but the first MDCT unit 101 may use any method.

  The first adjustment unit 102 adjusts the amplitude of the first frequency spectrum input from the first MDCT unit 101, and outputs the amplitude-adjusted first frequency spectrum to the IMDCT unit 103. The first adjustment unit 102 sends the code of a predetermined parameter used for amplitude adjustment as encoded data to a speech decoding apparatus (not shown). A specific method of amplitude adjustment will be described later.

  The IMDCT unit 103 performs inverse transformation of the orthogonal transformation performed in the first MDCT unit 101 on the first frequency spectrum input from the first adjustment unit 102, and describes a time-series signal (hereinafter referred to as “time signal”). ) To get. The IMDCT unit 103 outputs the acquired time signal to the CELP encoding unit 104.

  CELP encoding section 104 encodes and decodes the time signal input from IMDCT section 103 to acquire a decoded signal, and outputs the acquired decoded signal to second MDCT section 105. CELP encoding section 104 sends a code acquired by CELP encoding to a speech decoding apparatus (not shown) as encoded data.

  Second MDCT section 105 obtains the second frequency spectrum by performing orthogonal transform by MDCT on the decoded signal input from CELP encoding section 104. The second MDCT unit 105 outputs the acquired second frequency spectrum to the second adjustment unit 106.

  The second adjustment unit 106 gives a characteristic opposite to the characteristic given to the first frequency spectrum in the first adjustment unit 102 to the second frequency spectrum input from the second MDCT unit 105. Specifically, the second adjustment unit 106 extends the portion attenuated in the first adjustment unit 102. The second adjustment unit 106 outputs the second frequency spectrum to which the reverse characteristic is given to the subtraction unit 107. A specific method for providing the opposite characteristic will be described later.

  The subtraction unit 107 subtracts the second frequency spectrum input from the second adjustment unit 106 from the first frequency spectrum input from the first MDCT unit 101, and obtains a residual spectrum that can be easily encoded by FPC. The subtraction unit 107 outputs the acquired residual spectrum to the FPC encoding unit 108.

  The FPC encoding unit 108 transform-encodes the residual spectrum input from the subtracting unit 107 by the FPC method and outputs encoded data. This encoded data is sent to a speech decoding apparatus (not shown). The FPC encoding / decoding algorithm is described in ITU-T standard G.264. Since detailed description is given in the 718 standard, this description is omitted.

<Configuration of first adjustment unit>
As configurations of the first adjustment unit 102, two configurations of Configuration Example 1 and Configuration Example 2 will be described below. In the present embodiment, any one of Configuration Example 1 and Configuration Example 2 may be used.

(Configuration example 1)
FIG. 2 is a block diagram illustrating a configuration example 1 of the first adjustment unit 102.

  In Configuration Example 1, as a policy of amplitude adjustment in the first adjustment unit 102, in each sample of the first frequency spectrum, a sample having a relatively small spectral value is left without being attenuated, and a sample having a relatively large spectral value is increased. Attenuate. For example, the first adjustment unit 102 does not attenuate a sample having a spectral value less than the threshold value, and attenuates the sample having a spectral value greater than or equal to the threshold value as the spectral value increases. Adjust the amplitude. Therefore, in the configuration example 1, the amplitude adjustment in the first adjustment unit 102 is non-linear.

The average value calculation unit 201 receives the first frequency spectrum output from the first MDCT unit 101, calculates the following equation (1), and obtains the average value m.

  The average value calculation unit 201 outputs the obtained average value m to the amplitude adjustment unit 202.

The amplitude adjustment unit 202 uses the average value m input from the average value calculation unit 201 and a constant to obtain a threshold value according to the following equation (2).

The amplitude adjustment unit 202 performs amplitude adjustment on the first frequency spectrum input from the first MDCT unit 101 according to the following equation (3) using the threshold value t.

  Equation (3) is a continuous function up to the first derivative, and a linear function and a logarithmic function are interchanged at a threshold. That is, the amplitude adjusting unit 202 keeps the spectrum having the amplitude equal to or smaller than the threshold value, and attenuates the spectrum having the amplitude larger than the threshold value by a logarithmic function.

  In addition, the amplitude adjustment unit 202 outputs the threshold value to the second adjustment unit 106. In addition, since the amplitude adjusting unit 202 needs to perform inverse transform even in a speech decoding device (not shown), the threshold t that is a parameter used for amplitude adjustment is encoded, and the code of the threshold t is converted to the speech decoding device. Send to. In this case, the encoding method includes scalar quantization by integerization. In order to avoid an adverse effect of the threshold value t due to coding distortion, after obtaining the threshold value t by the equation (2), encoding and decoding are performed, and the decoded threshold value t is used ( 3) The process of the formula and the process of the second adjustment unit 106 are performed.

  Then, the amplitude adjustment unit 202 outputs the amplitude-adjusted first frequency spectrum to the IMDCT unit 103.

(Configuration example 2)
FIG. 3 is a block diagram illustrating a configuration example 2 of the first adjustment unit 102.

  In the configuration example 2, as a policy of amplitude adjustment in the first adjustment unit 102, in each sample of the first frequency spectrum, a sample with a relatively gentle change in the spectrum value is left as it is without attenuating the spectrum value. Samples that are pulsed in the change of the value attenuate the spectral value. For example, the first adjustment unit 102 does not attenuate a sample whose spectral value change amount between adjacent samples of the first frequency spectrum is less than a threshold value, and the change amount is equal to or greater than the threshold value. Amplitude adjustment is performed on the sample so that the larger the amount of change, the greater the attenuation. Therefore, also in the configuration example 2, the amplitude adjustment in the first adjustment unit 102 is non-linear.

The pulse analysis unit 301 uses the first frequency spectrum input from the first MDCT unit 101 to obtain an attenuation coefficient vector. Specifically, the pulse property analysis unit 301 obtains the pulse property by analyzing the relationship between the spectral value of the sample at the center and the spectral values of the samples around the sample. At this time, for example, as described above, the pulse property is obtained by comparing the threshold value with the change amount of the spectrum value. When the pulse characteristic analysis unit 301 determines that there is a pulse characteristic, the pulse characteristic analysis unit 301 stores a constant value less than 1, and stores “1.0” as a value otherwise, thereby obtaining an attenuation coefficient vector. That is, the pulse characteristic analysis unit 301 obtains an attenuation coefficient vector according to the following equation (4).

  The pulse property analysis unit 301 outputs the obtained attenuation coefficient vector to the amplitude adjustment unit 302 and the second adjustment unit 106. Here, the attenuation coefficient vector is a function for attenuating a pulsed spectral value.

  In addition, since it is necessary for inverse analysis to be performed in the speech decoding apparatus, the pulse analysis unit 301 encodes the attenuation coefficient vector, which is a parameter used for amplitude adjustment, and sends the sign of the attenuation coefficient vector to the speech decoding apparatus. At this time, as an encoding method, a method in which the attenuated sample is set to “1” and a non-attenuated sample is set to “0” and the vector is transmitted as it is, or a method in which vector quantization is performed by dividing a section into several parts is given. . A method of limiting the number of parts to be attenuated in advance and encoding and transmitting only a part of the attenuation coefficient vector is also conceivable. As an extreme example, a method may be used in which the attenuation coefficient vector shown in the equation (4) is created from the spectrum obtained by the second adjustment unit 106 and no information is sent. In this case, the sign of the attenuation coefficient vector is not sent.

The amplitude adjustment unit 302 uses the attenuation coefficient vector input from the pulse property analysis unit 301 to adjust the amplitude of the first frequency spectrum according to the following equation (5).

  Then, the amplitude adjustment unit 302 outputs the first frequency spectrum whose amplitude has been adjusted to the IMDCT unit 103.

<Operation in Second Adjustment Unit>
The operation of the second adjustment unit 106 in each case where the first adjustment unit 102 has the configuration of the above configuration example 1 or the configuration of the configuration example 2 will be described below.

(Operation when the configuration of the first adjustment unit is the configuration example 1)
The second adjustment unit 106 adaptively adjusts the amplitude according to the overall tendency of the amplitude of the second frequency spectrum. The adjustment of the second adjustment unit 106 corresponds to the inverse transformation of the first adjustment unit 102, and is characterized in that the amplitude is increased non-linearly when it is larger than the threshold value t obtained by the equation (2).

Based on the threshold value t input from the amplitude adjustment unit 202, the second adjustment unit 106 performs adjustment by performing inverse conversion according to the following equation (6).

  From the equation (6), the second adjustment unit 106 determines that the sample having the spectrum value less than the threshold value t out of the samples of the second frequency spectrum is not stretched, and the spectrum value is equal to or greater than the threshold value t. Amplitude adjustment is performed so that the larger the spectral value is, the larger the spectral value is.

(Operation when the configuration of the first adjustment unit is the configuration example 2)
The second adjustment unit 106 adaptively adjusts the amplitude according to the overall tendency of the amplitude of the second frequency spectrum. The adjustment of the second adjustment unit 106 corresponds to the inverse transformation of the first adjustment unit 102, and is characterized in that the amplitude is increased nonlinearly when the amount of change in the spectral value between adjacent samples is larger than the threshold value. Yes.

The second adjustment unit 106 performs inverse conversion according to the following equation (7) based on the attenuation coefficient vector input from the pulse property analysis unit 301.

  From the equation (7), the second adjustment unit 106 determines that the change amount of the spectrum value between the adjacent samples of the second frequency spectrum is less than the threshold value, and does not expand it, and the change amount is the threshold value. Amplitude adjustment is performed so that the larger the amount of change is, the larger the amount of change is for a sample equal to or greater than the value.

  Note that the speech decoding apparatus corresponding to the speech encoding apparatus of the present embodiment decodes the threshold value t or the attenuation coefficient vector based on the transmitted threshold code or attenuation coefficient vector code. This speech decoding apparatus performs the same operation as that performed by the CELP encoding unit 104 of the speech encoding device, the second MDCT unit 105, and the second adjustment unit 106, thereby decoding the synthesized sound of CELP. Generate a frequency spectrum. Further, the speech decoding apparatus adds the decoded FPC frequency spectrum generated by decoding the FPC code to the decoded frequency spectrum of the synthesized sound of CELP. Since the description of each block is as described above, the detailed description is omitted.

<Effects of the present embodiment>
According to the present invention, it is possible to eliminate waste of an information source code at the time of CELP encoding, reduce the amount of calculation, and realize noise shaping processing in CELP encoding. FPC encoding performance can be improved.

<Modification of the present embodiment>
In the above embodiment, the configuration example 1 and the configuration example 2 are shown as the configuration of the first adjustment unit. However, the present invention is not limited to this, and other configurations and operations are possible as long as the configuration and operation attenuate the spectrum value. It may be used. This is because the present invention does not directly depend on the configuration and operation of the adjustment unit before and after the CELP encoding unit.

  Moreover, in the said embodiment, although the average value was calculated | required by the arithmetic mean in the 1st adjustment part, this invention is not restricted to this, You may use a mode or a median. In the case of using the median, the same effect can be obtained by searching for the amplitude value of the rank specified in advance and using it.

  In the above embodiment, the subsequent encoder is an FPC encoding unit. However, the present invention is not limited to this, and TCX (transform coded excitation), AMR-WB + (Extended Adaptive Multi-Rate Wideband), or AAC An encoding unit that performs transform encoding such as (Advanced Audio Coding) may be provided. This is because any encoding method may be used to efficiently encode a residual spectrum having a pulse property, and the present invention is effective for other than FPC.

  In the above embodiment, the CELP encoding unit is used. However, the present invention is not limited to this, and an encoding unit that performs time-series encoding such as MPC (Multiple Pulse Coding) or ADPCM may be provided. These are because time series signals can be efficiently encoded, but the effect of the attenuation of the decoded signal in the subsequent spectral encoding is the same as CELP.

  In the above embodiment, MDCT is used as the orthogonal transform method. However, the present invention is not limited to this, and an orthogonal transform method such as DCT (Discrete Cosine Transform) or FFT may be used. The reason is that any method capable of obtaining a frequency spectrum can be applied to the above embodiment.

  In the above-described embodiment, the device is referred to as a speech encoding device / speech decoding device, but “speech” here indicates speech in a broad sense. That is, the input signal in the speech coding apparatus and the decoded signal in the speech decoding apparatus indicate any signal such as a speech signal, a music signal, or an acoustic signal including both a speech signal and a music signal. .

  Further, although cases have been described with the above embodiment as examples where the present invention is configured by hardware, the present invention can also be realized by software in cooperation with hardware.

  Each functional block used in the description of the above embodiment is typically realized as an LSI which is an integrated circuit. These may be individually made into one chip, or may be made into one chip so as to include a part or all of them. The name used here is LSI, but it may also be called IC, system LSI, super LSI, or ultra LSI depending on the degree of integration.

  Further, the method of circuit integration is not limited to LSI's, and implementation using dedicated circuitry or general purpose processors is also possible. An FPGA (Field Programmable Gate Array) that can be programmed after manufacturing the LSI or a reconfigurable processor that can reconfigure the connection or setting of circuit cells inside the LSI may be used.

  Further, if integrated circuit technology comes out to replace LSI's as a result of the advancement of semiconductor technology or a derivative other technology, it is naturally also possible to carry out function block integration using this technology. Biotechnology can be applied.

  The speech coding apparatus and speech coding method according to the present invention are suitable for using a scalable codec technology having a multilayer structure.

DESCRIPTION OF SYMBOLS 100 Speech coding apparatus 101 1st MDCT part 102 1st adjustment part 103 IMDCT part 104 CELP encoding part 105 2nd MDCT part 106 2nd adjustment part 107 Subtraction part 108 FPC encoding part

Claims (6)

First orthogonal transform means for performing orthogonal transform on a target signal to be encoded to generate a first frequency spectrum;
First adjusting means for performing a first amplitude adjustment that is nonlinear with respect to the first frequency spectrum;
An inverse orthogonal transform means for performing an inverse transform of the orthogonal transform on the first frequency spectrum subjected to the first amplitude adjustment to generate a time signal;
First encoding means for encoding the time signal and then decoding to generate a decoded signal;
Second orthogonal transform means for performing orthogonal transform on the decoded signal to generate a second frequency spectrum;
Second adjusting means for giving the second frequency spectrum a characteristic opposite to the characteristic given to the first frequency spectrum during the first amplitude adjustment;
Subtracting means for generating a residual spectrum by subtracting the second frequency spectrum given the reverse characteristic by the second adjusting means from the first frequency spectrum generated by the first orthogonal transforming means;
Second encoding means for transform encoding the residual spectrum;
A speech encoding apparatus.   The first adjustment means performs the first amplitude adjustment by attenuating the sample having a spectrum value greater than or equal to a threshold value among the samples of the first frequency spectrum as the spectrum value increases. The speech encoding apparatus according to claim 1, wherein   The second adjustment means performs a second amplitude adjustment for expanding a sample with a spectrum value greater than or equal to a threshold value among samples of the second frequency spectrum, as the spectrum value increases. The speech coding apparatus according to claim 2, wherein the inverse characteristic is given to the second frequency spectrum by performing adaptively according to the amplitude of the two frequency spectrum.   The first adjustment means performs the first amplitude adjustment by attenuating the first frequency spectrum in a sample where the change amount of the spectrum value between adjacent samples is greater than or equal to a threshold value, as the change amount is larger. The speech encoding apparatus according to claim 1, wherein:   The second adjustment means performs a second amplitude adjustment for increasing the amount of change in the spectrum value between adjacent samples of the second frequency spectrum that is larger as the amount of change is larger than a threshold value. The speech coding apparatus according to claim 4, wherein the inverse characteristic is given to the second frequency spectrum by performing adaptively according to the amplitude of the second frequency spectrum. A first orthogonal transform step of performing orthogonal transform on the target signal to be encoded to generate a first frequency spectrum;
A first adjustment step for performing non-linear amplitude adjustment on the first frequency spectrum;
An inverse orthogonal transform step of generating a time signal by performing an inverse transform of the orthogonal transform on the first frequency spectrum whose amplitude has been adjusted by the first adjustment step;
A first encoding step of encoding the time signal and then decoding to generate a decoded signal;
A second orthogonal transform step of performing an orthogonal transform on the decoded signal to generate a second frequency spectrum;
A second adjustment step for giving the second frequency spectrum a characteristic opposite to the characteristic given to the first frequency spectrum by the first adjustment step;
A subtracting step of generating a residual spectrum by subtracting the second frequency spectrum given the inverse characteristic by the second adjusting step from the first frequency spectrum generated by the first orthogonal transforming step;
A second encoding step for transform encoding the residual spectrum;
A speech encoding method comprising:

Images