JP3159012B2 - Audio signal encoding device and decoding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio signal encoding and decoding apparatus, and more particularly to a PCM (Pulse Code Modulati).
on) signal is converted to an MDCT (Modified Dis
The present invention relates to a high-efficiency encoding technique that performs orthogonal transformation by a crete cosine transform (Crete Cosine Transform) method and separates an amplitude component and a phase component into vector quantization, and a decoding technology of the encoded data.

[0002]

2. Description of the Related Art Conventionally, high efficiency coding of audio signals has been performed by MPEG (Moving Picture Experts Group) -Aud.
As represented by io, etc., the sample signal in the time domain is mapped to the frequency domain, and the bit allocation related to the uneven distribution of energy in the frequency domain is controlled for the quantization distortion by using the human psychoacoustic characteristics of each sample. Is done by doing.
Such an encoding method is based on a so-called scalar quantization method. However, in encoding at a low bit rate below an average of 1 (bit / sample), quantization distortion becomes extremely large.

Therefore, it is conceivable to introduce vector quantization in which several samples are quantized collectively. here,
"Vector quantization" has been conventionally applied to a moving picture coding method. A pattern of various types of vector data representing a subject is stored in a code book in association with index data, and In transmitting image data, only the index data is transmitted. As a method of applying the vector quantization to an audio signal, a method of vector-quantizing a signal waveform itself, a method of vector-quantizing a linear prediction residual signal, and a method of vector-quantizing an orthogonal transform coefficient are used. Have been tried. However, the method of applying vector quantization to the signal waveform itself or the linear prediction residual signal is not suitable for a wideband (audio band) audio signal having a characteristic that changes greatly and time-series correlation is weak. On the other hand, the method of vector quantizing the orthogonal transform coefficients can utilize the uneven distribution of energy in the frequency domain, and is suitable for the above-described high-efficiency encoding of a wideband acoustic signal.

[0004] Therefore, as for an audio signal coding method combining orthogonal transform and vector quantization, "Introduction of VQ of transform coefficient to transform coding of audio signal" (Kumada et al.,
Techniques such as the Acoustical Society of Japan, 93/10, p.307) are disclosed. FIG. 8 shows a configuration example of the encoding apparatus. In the figure, first, orthogonal transformation by the MDCT method is performed by the MDCT unit 51 on the input PCM data.
The MDCT coefficients are converted into four-dimensional vector data for each of the four coefficient data by a four-dimensional vectorizer 52 along the frequency direction. Next, the normalization calculator 53 obtains an rms (root mean square) value of each vector data, log-quantizes the rms value, and quantizes the rms value into each vector data according to the log-quantized rms value. All the bits
Each vector data is normalized by the rms value. On the other hand, in the codebook 54, index data is registered in association with vector data obtained by performing the same MDCT orthogonal transform, MDCT coefficient four-dimensional vectorization and normalization processing on the learning sample of the audio signal as described above. I have. Therefore, the vector quantizer 55 searches the code book 54 using the normalized vector data as a search key, and the corresponding index data is read. If there is an excess in the number of allocated bits, a quantization error is obtained by inputting the number of bits and the original DCT coefficient data obtained from the four-dimensional vectorizer 52 to a subtractor 56. The quantization error is scalar-quantized. The multiplexer 58 forms a multiplexed bit stream of the rms value, the index data, and the quantization error data, and outputs the multiplexed bit stream to the transmission path.

[0005]

According to the contents reported in the above-mentioned paper, the S / N ratio of the encoding device shown in FIG. 8 is in the range of 90 to 208 kbps compared with the conventional scalar quantization system. About 2 to 3 dB, and good results have been obtained. However, at 90 kbps, the difference from the general scalar quantization method has become smaller, and the tendency is even lower. Application to bit rates is expected to be difficult.

[0006] The cause is considered as follows. First, essentially, vector quantization is a technique that utilizes the bias of the statistical distribution of data. In the above encoding device, the MDCT coefficient x as shown in FIG.
(n) is divided into four units in the frequency direction, converted into four-dimensional vector data, and then stored with data normalized by the rms value. However, the MDCT coefficient x (n) is adjacent due to a change in the phase of the signal. Signs and magnitudes of the combined coefficients greatly change, which limits the efficiency of vector quantization. That is, in order to apply the method of the above encoding apparatus to a bit rate lower than 90 kbps,
It is necessary to use a vector having such a property that the statistical distribution is more biased.

On the other hand, when the amplitude component data and the phase component data are obtained from the MDCT coefficient x (n) in FIG. 5, the results are as shown in FIGS. 6 and 7, respectively. In particular, the bias of the statistical distribution of the amplitude component data is large. Understood. Therefore, the present invention
Instead of vector quantizing the MDCT coefficient x (n) itself, a special method of vector quantizing the amplitude component data and the phase component data obtained from the MDCT coefficient x (n) is adopted, and the overall efficiency is improved. It has been created for the purpose of providing an encoding device that enables efficient encoding and a decoding device corresponding thereto.

[0008]

According to a first aspect of the present invention, an acoustic signal modulated by a PCM method is orthogonally transformed by an MDCT method,
In an audio signal encoding apparatus for encoding MDCT coefficients obtained by the conversion, data separating means for separating the MDCT coefficients into amplitude component data and phase component data, and each data separated by the data separating means as vector data. Vectorization means for converting the MDCT coefficients obtained from the training sample of the acoustic signal into amplitude components obtained by performing substantially the same data processing as the data processing by the data separation means and the vectorization means. A code book in which mutually different index data is stored in correspondence with each vector data related to the phase component, and the code book is searched using each vector data related to the amplitude component and the phase component obtained by the vectorization unit. , The vector data of the code book which is the same as or approximate to the respective vector data Encoding a sound signal, comprising: vector quantization means for reading index data relating to the associated amplitude component and phase component; and outputting the index data read by the vector quantization means to a transmission path. Related to the device.

According to the present invention, M
The DCT coefficient is separated into amplitude component data and phase component data, and each separated data is converted into vector data by vectorization means. The codebook corresponds to each vector data relating to the amplitude component and the phase component obtained by performing substantially the same processing as the MDCT / data separation / vectorization according to the present invention on the learning sample of the acoustic signal. Different index data are tabulated. Therefore, if the vector quantization means refers to the codebook as a search key using each vector data related to the amplitude component and the phase component converted by the vectorization means, and reads and transmits the corresponding index data, the transmission efficiency can be improved. A significant improvement can be achieved, and the capacity of the storage medium on the transmission path can be reduced.

As is apparent from the comparison between FIG. 5 and FIG. 6, the statistical deviation of the distribution state of the amplitude component data a (m) is far greater than that of the MDCT coefficient x (n). large. Therefore, it is sufficient that the vector data relating to the amplitude component stored in the code book has a large number of conversion dimensions and a small number of data.
On the other hand, for the phase component data θ (m), since the bias of the statistical distribution is small as shown in FIG. 7, it is necessary to prepare a large number of vector data having a smaller conversion dimension than the amplitude component data. No. However, even if the codebooks have the same memory capacity, as described above, a smaller amount of vector data relating to the amplitude component can be allocated to the vector data relating to the phase component. Therefore, as compared with the case where the MDCT coefficient x (n) is vector-quantized as it is and stored in the codebook, the effective allocation of the bit amount becomes possible. Can be reduced.

According to a second aspect of the present invention, there is provided a decoding apparatus corresponding to the encoding apparatus according to the first aspect of the present invention, wherein the codebook stores the same data as data stored in a codebook provided in the encoding apparatus. Vector inverse quantization means for searching the code book using the input index data related to the amplitude component and the phase component, and reading out vector data related to the amplitude component and the phase component corresponding to each index data; Vector synthesizing means for synthesizing each vector data read by the vector dequantizing means to create amplitude component data and phase component data relating to MDCT coefficients;
A data synthesizing unit for synthesizing the amplitude component data and the phase component data obtained by the vector synthesizing unit to obtain an MDCT coefficient; and reproducing a sound signal modulated by the PCM method based on the MDCT coefficient obtained by the data synthesizing unit. Reverse MD
The present invention relates to an audio signal decoding device including CT means.

It is natural that the codebook of this decoding device must have the same data content as the data stored in the codebook of the first invention, and the index book is searched using the index data as a search key. Only differ. A series of data processing by the vector synthesizing unit, the data synthesizing unit, and the inverse MDCT unit after the vector dequantizing unit reads out the vector data related to the amplitude component and the phase component is performed by the orthogonal transform by the MDCT method according to the first invention. This corresponds to the reverse processing of the processing executed by the data separating means and the vectorizing means. Therefore, since the encoding apparatus and the decoding apparatus have the codebook having the above characteristics, the decoding apparatus can reproduce an audio signal having a high SN ratio even at a low bit rate.

[0013]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an embodiment of an audio signal encoding and decoding apparatus according to the present invention; FIG. First, FIG. 1 shows a functional block circuit diagram of an encoding device. In the figure, 1 is the input P
MDC that performs orthogonal transform on CM signal by MDCT method
T unit, 2 is an amplitude / phase separator for obtaining amplitude component data a (m) and phase component data θ (m) from the MDCT coefficients obtained by MDCT unit 1 and separating and outputting them. 3 is amplitude component data a.
(m) and the phase component data θ (m) are divided for each band, and the two data a (m) and θ (m) are changed in the number of conversion dimensions for each band to obtain vector data av (m) and θv ( m), 4 is a normalization calculator that normalizes vector data av (m) related to the amplitude component by a coefficient a (k), 5 is a codebook for the amplitude component, and 6 is a codebook for the amplitude component. A codebook for phase component, 7 is a vector quantizer that searches the codebook 5 for amplitude component using the vector data avn (m) relating to the normalized amplitude component as a search key and reads out index data IDa, 8 Searches the codebook 5 for the phase component using the vector data θv (m) related to the phase component as a search key,
A vector quantizer 9 for reading out Dp, and a multiplexer 9 for forming a bit string obtained by multiplexing the coefficient a (k) used for normalization and the index data IDa and IDp and outputting the bit stream to the transmission line.

Next, the operation of the encoding apparatus will be described with a more detailed description of each functional block element.
The input PCM signal is cut out frame by frame by the MDCT unit 1, and an orthogonal transformation process is performed on each frame to obtain an MDCT coefficient x (n) in the frequency axis direction. In the amplitude / phase separator 2, the obtained MDC
Since the T coefficient x (n) has both amplitude information and phase information, the following equation 1 and equation 2 are applied to the amplitude component data a
(m) and the phase component data θ (m) are calculated to separate the amplitude component and the phase component.

(Equation 1)

Where n and m are spectrum numbers, and n = 0, 2, 4, 6,..., N−1, and m = 0, 1, 2, 3,.
.., (N-1) / 2, where N is the number of spectra in one frame.

Then, the obtained amplitude component data a (m)
And the phase component data θ (m) are input to the band division / vectorization unit 3. In the band splitter / vectorizer 3, first, the frequency band of the audio signal is set to, for example, 0 to 2.25 kHz and 2.2.
5 to 4.5 kHz, 4.5 to 9.0 kHz and 9.0 to 16.0
The frequency band is divided into four bands of kHz and the amplitude component data a (m) and the phase component data θ (m) are divided for each of these bands. Next, each of the divided data a (m) and θ (m) is converted into vector data av (m) and θv (m) using the conversion dimension set for each band. That is, the sampling number S is set for each band, and vector data of the conversion dimension number S is created for each of the sampling data groups.

In this case, for example, in the case of the above-mentioned band division, at 0 to 2.25 kHz, the dimension number of av (m) is 6, θv (m)
, The dimension number of av (m) is 12, and the dimension number of θv (m) is 4 in the range of 2.25 to 4.5 kHz, and 4.5 to 9.0.
In kHz, the number of dimensions of av (m) is 24, and the number of dimensions of θv (m) is 6.
In 9.0 to 16.0 kHz, the dimension number of av (m) is 32,
The conversion dimension in a low frequency band is made smaller than the conversion dimension in a high frequency band such that the dimension of θv (m) is set to 20. This is in consideration of the fact that the higher the frequency is in human auditory psychology, the worse the frequency discrimination ability is, and the more effective data amount is allocated using the characteristic of being an acoustic signal.

Further, vector data θv relating to the phase component
When creating (m), the difference between the first value in the dimension and the subsequent data value is used as vector data instead of the absolute value of the phase component data θ (m). This is because, as shown in FIG. 7, since the bias of the statistical distribution of the phase component data θ (m) is generally small, it is better to vectorize the difference data than to use the absolute value to obtain the vector data θv (m). This is because the amount becomes smaller as a whole.

Next, the band dividing / vectoring unit 3 transfers the created vector data av (m) relating to the amplitude component together with the band identification data B to the normalization computing unit 4, and also generates the vector data θv relating to the phase component. (m) and the band identification data B are transferred to the vector quantizer 8. In the normalization arithmetic unit 4, the transferred vector data av (m) is logarithmized, and the maximum power value in the vector data av (m) is normalized as a normalization coefficient a (k). m) and transfers it to the vector quantizer 7 together with the band identification data B. Here, the logarithmization / normalization processing is performed to reduce the data amount of the vector data av (m) related to the amplitude component.

Thus, the vector quantizer 7 receives the band identification data B and the vector data avn (m) relating to the normalized amplitude component, and the vector quantizer 8 receives the band identification data B and the phase component. Are input to the vector data θv (m) according to
Using [B and avn (m)] and [B and θv (m)] as search keys, the corresponding codebooks 5 and 6 are searched.

At this stage, the contents of the codebooks 5 and 6 will be described with reference to FIGS. Both codebooks 5 and 6 contain the above-mentioned “MDCT unit 1 → amplitude / phase separator 2 → band division / vectorizer 3 → (normalization calculator 4 ) >>, the data patterns of vector data avn (m) and θv (m) obtained by performing substantially the same processing are stored in association with mutually different index data IDa and IDp. That is, each of the data a is obtained by a known method such as an LBG algorithm using a learning sample of an acoustic signal.
All data patterns of vn (m) and θv (m) are registered. However, in the present embodiment, band division is performed in the processing stage of the band division / vectorization unit 3 as described above, and the number of dimensions to be converted into vector data is changed for each band. Therefore, regarding the codebook 5 for the amplitude component, as shown in FIGS.
Inside the codebook [ACB; 1], [ACB; 2], [AC
B; 3] and [ACB; 4], each of which is obtained by processing and normalizing a learning sample, from 0 to 2.25 kHz and 2.25 to
4.5 kHz, 4.5 to 9.0 kHz, and 9.0 to 16.0 kHz
The data pattern of the vector data avn (m) is registered. 2 (B) to 2 (E) are shown in an illustrative pattern for easy understanding. Then, according to the vectorization method corresponding to the example of the present embodiment,
The number of conversion dimensions of each vector data pattern is 0 to 2.2
To 5 at 5 kHz, to 12 at 2.25 to 4.5 kHz
In the case of 4.5 to 9.0 kHz, it is set to 24, and in the case of 9.0 to 16.0 kHz, it is set to 32, and 1024 data patterns are provided for each band. These patterns are shown in FIG.
Frequency after orthogonal transform by MDCT method shown in (A)-
When associated with the graph of the amplitude logarithmic power, the pattern of the vector data avn (m) in the minute frequency band Δf is shown. The higher the frequency band, the larger the minute frequency band Δf and the larger the number of conversion dimensions. It is a data pattern.

As shown in FIGS. 1 and 3 (B)-(E), the codebook 6 for the phase component also includes a codebook [PCB; ],
[PCB; 2], [PCB; 3], and [PCB; 4] are divided into four parts.
25 kHz, 2.25 to 4.5 kHz and 4.5 to 9.0 kHz
And vector data θv (m) at 9.0 to 16.0 kHz
Has been registered. Further, corresponding to the vectorization method of the present embodiment, the data pattern of the vector data θv (m) having a larger conversion dimension than the band-based codebook of the lower frequency band for the band-based codebook of the higher frequency band is used. The same is true for the registration. For 0 to 2.25 kHz, the value is changed to 2, 2.25 to 4.5 kHz.
At 4 Hz, at 4.5-9.0 kHz, 6 and 9.0-1
At 6.0 kHz, it is 20. By the way, the conversion dimension of the data pattern of the vector data θv (m) in the codebook 6 for the phase component is smaller than that in the codebook 5 for the amplitude component throughout the entire band. Codebook for each band [PCB; 1], [P.
Vector data θ in CB; 2], [PCB; 3], [PCB; 4]
Codebook for amplitude component with 2048 total v (m)
It is twice as large as in the case of 5. This is because the bias of the statistical distribution of the phase is small, as is clear from the graph of the frequency-phase value after the orthogonal transform by the MDCT method shown in FIG. Otherwise, the phase change mode cannot be appropriately expressed, and the error rate in the vector quantization stage increases.

Returning to FIG. 1, the vector quantizers 7 and 8 respectively search the codebooks 5 and 6 described above. Books [ACB; 1] ~ [ACB; 4] and [PCB; 1] ~ [PC
B; 4] is selected. Each of the vector quantizers 7 and 8 uses the vector data avn (m) related to the amplitude component and the vector data θv (m) related to the phase component included in the band as search keys, and Codebook [ACB; i] by band selected in
[PCB; i] is searched and each vector data avn (m) and θv (m)
The index data IDa and IDp associated with the same or the closest registered data are read out.

Thereafter, the normalization coefficient a (k) used by the normalization operation unit 4 and the index data IDa and IDp read by the vector quantizers 7 and 8 are transferred to the multiplexer 9, and the multiplexer 9 multiplexes them. And outputs the converted bit string to the transmission path. In the encoding device according to the present embodiment, the band division / vectorization unit 3 separates the MDCT coefficient X (n) into the amplitude component data a (m) and the phase component data θ (m) as described above. Codebooks 5 and 6 with a small data size by allocating the number of conversion dimensions of vectorization, vectorizing by the difference of phase component data θ (m), and normalizing vector data av (m) related to the amplitude component. Can be configured. In the codebooks 5 and 6, the conversion dimension of the vector data avn (m) related to the amplitude component is larger than the conversion dimension of the vector data θv (m) related to the phase component. From the viewpoint, it is necessary to reduce the amount of data on the amplitude information side that is sufficient for conversion with a relatively coarse pattern because the bias of the statistical distribution is large, and conversely, it is necessary to convert with a fine pattern where the bias of the statistical distribution is small The data amount on the phase information side can be increased, and rational and highly efficient encoding can be realized.

Next, FIG. 4 shows a functional block circuit diagram of a decoding device corresponding to the above-mentioned coding device. In the figure, reference numeral 11 denotes a demultiplexer for individually separating a normalization coefficient a (k) and index data IDa and IDp received from a transmission path, 12 a codebook for an amplitude component, 13 a codebook for a phase component, Reference numeral 14 denotes a vector dequantizer for searching the amplitude component codebook 12 using the index data IDa as a search key and reading out the vector data avn (m) and the band identification data B relating to the normalized amplitude component. A vector inverse quantizer that searches the phase component codebook 13 using the index data IDp as a search key and reads out the vector data θv (m) and the band identification data B relating to the phase component, and 16 is a normalization coefficient a (k ) Using vector data avn
(m) is inversely normalized by using an inverse normalization calculator 17 using vector data av (m) related to the inversely normalized amplitude component, vector data θv (m) related to the phase component, and band identification data B. Each of the vector data av (m) and θv (m) is arranged in the corresponding band, and the vector data av (m) and θv (m) are synthesized to combine the amplitude component data a (m) and the phase component data θv ( m) is a band combining / vector combining unit, and 18 is both component data a
(m), an amplitude / phase synthesizer that obtains an MDCT coefficient x (n) using θ (m). 19 is an inverse MDCT for the MDCT coefficient x (n).
IMDCT (Inver
s MDCT) device.

Each functional block of the decoding device corresponds to each functional block of the above-described encoding device in an inverse relationship with respect to the flow of data conversion and processing, and performs each of the inverse conversion and processing. Since the operation is executed, a detailed description of the operation is omitted here, but the decoding device has the following features.

First, the codebooks 12 and 13 have the same data contents as the codebooks 5 and 6 on the encoder side, and the vector dequantizers 14 and 15 transmit the demultiplexer 11 from the transmission path side. Using the index data IDa and IDp received via this as search keys, the vector data avn associated with the search keys in each of the codebooks 12 and 13 is used.
(m) and θv (m) are read. In this case, although the band identification data B is not transmitted from the transmission path side, since each index data IDa, IDp is given with a different code individually, the index data IDa hit in the search of the code books 12, 13 is obtained. , Band identification data B can be detected based on the address of IDp.

The amplitude / phase combiner 18 uses the amplitude component data a (m) and the phase component data θ (m) created by the band combiner / vector combiner 18 to obtain the following formulas 3 and 4. The MDCT coefficients x (n) and x (n +
Ask for 1).

(Equation 3)

Where n and m are spectrum numbers, and n = 0, 2,
4, 6, ..., N-1, m = 0, 1, 2, 3, ..., (N-1) / 2
And N is the number of spectra in one frame.

The encoding apparatus and the decoding apparatus according to the present embodiment have a basic effect of realizing high transmission efficiency and saving of storage medium capacity by the vector quantization method. Vector data avn (m), θv related to amplitude and phase components for 6, 14, and 15
By devising the number of conversion dimensions and the number of data in the method of assigning (m), and applying means to reduce the amount of data during vectorization, audio data can be transmitted more efficiently and accurately, and decoded. It is possible to reproduce high-quality sound at the output of the conversion device. In the present embodiment, the encoding device and the decoding device are provided with a normalization operation unit 4 and a denormalization operation unit 16, respectively, to reduce the data amount of the vector data av (m) related to the amplitude component. But,
Even without these functional blocks, the above effects can be obtained only by slightly increasing the size of the codebooks 5, 14.

[0029]

According to the "audio signal encoding apparatus and decoding apparatus" of the present invention, the following effects can be obtained by having the above configuration. A first aspect of the present invention is a code for orthogonally transforming a PCM signal by an MDCT method, encoding an MDCT coefficient obtained by the transform, and applying a vector quantization method to achieve highly efficient compression and improvement of transmission efficiency. A code book in which the MDCT coefficients are separated into amplitude component data and phase component data and the vector data is made to correspond to the index data by the vectorization device. It is possible to perform more efficient encoding by utilizing the characteristic of the acoustic signal that the distribution has a large bias. According to the second aspect of the present invention, it is possible to perform an efficient calculation when obtaining the amplitude component data and the phase component data from the MDCT coefficients by the data separating means. According to a third aspect of the present invention, as the data conversion method in the vectorization means, the number of conversion dimensions is increased in a high frequency band,
A reduction in the amount of transmitted data and a reduction in the size of the codebook are realized in consideration of psychoacoustic characteristics. The invention according to claim 4 suppresses the data amount of the vector data related to the phase component to a small value, realizes a small codebook, and includes more data for vector quantization in a codebook having the same capacity. Make it possible. The invention according to claims 5 and 6 is characterized in that the vectorization conversion dimension of phase component data having a small statistical distribution bias is smaller than the vectorization conversion dimension of amplitude component data having a large statistical distribution bias, and By increasing the number of vector data related to components, efficient and rational encoding can be performed without increasing the amount of transmission data and the codebook capacity. According to the seventh aspect of the present invention, it is possible to reduce the size of the codebook by normalizing the vector data related to the amplitude component, and to provide a codebook having the same capacity with more data for vector quantization. To The invention of claim 8 is
A decoding device corresponding to the encoding device according to the first to sixth aspects of the present invention is realized. According to a ninth aspect of the present invention, a decoding apparatus corresponding to the encoding apparatus according to the seventh aspect of the present invention is realized.
According to a tenth aspect of the present invention, in the decoding apparatus according to the eighth or ninth aspect, the data synthesizing unit enables efficient computation when obtaining the MDCT coefficient from the amplitude component data and the phase data.

[Brief description of the drawings]

FIG. 1 is a functional block circuit diagram of an audio signal encoding device according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram regarding registered contents of a code book for an amplitude component.

FIG. 3 is an explanatory diagram relating to registered contents of a code book for a phase component.

FIG. 4 is a functional block circuit diagram of the decoding device according to the embodiment.

FIG. 5 is a graph showing an actual example of a variation mode of an MDCT coefficient.

FIG. 6 is a graph showing how the amplitude component data relating to the MDCT coefficient in FIG. 5 changes.

FIG. 7 is a graph showing a change aspect of phase component data related to the MDCT coefficient in FIG. 5;

FIG. 8 is a functional block circuit diagram of a conventional encoding device employing a vector quantization method.

[Explanation of symbols]

1 ... MDCT unit, 2 ... amplitude / phase separator (data separation unit), 3 ... band splitting / vectorization unit (vectorization unit), 4 ...
Normalization arithmetic unit (normalization means), 5,12 codebook for amplitude component, 6,13 codebook for phase component, 7,8 vector quantizer (vector quantization means), 9 multiplexer , 11 ... demultiplexer, 14, 15 ... vector inverse quantizer (vector inverse quantizer), 16 ... inverse normalizer (inverse normalizer), 17 ... band synthesis / vector synthesizer (vector synthesizer) , 18 ... amplitude / phase synthesizer (data synthesis means), 19 ...
IMDCT unit (inverse MDCT means).

──────────────────────────────────────────────────続 き Continuation of front page (56) References JP-A-59-94797 (JP, A) JP-A-3-121624 (JP, A) JP-A-3-240319 (JP, A) JP-A-5-947 145427 (JP, A) JP-A-6-291674 (JP, A) JP-A-7-248799 (JP, A) JP-A-7-261800 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) H03M 7/30

Claims (10)

(57) [Claims] An audio signal modulated by a PCM (Pulse Code Modulation) method is converted to an MDCT (Modified Discrete Cos
In an audio signal encoding apparatus for performing orthogonal transformation by an ine transform) method and encoding MDCT coefficients obtained by the transformation, a data separation unit for separating the MDCT coefficients into amplitude component data and phase component data; Vectoring means for converting each data separated by the means into vector data; and data processing substantially the same as the data processing by the data separating means and the vectorizing means for MDCT coefficients obtained from a learning sample of an acoustic signal. And a code book in which different index data are stored in correspondence with the respective vector data relating to the amplitude component and the phase component obtained by performing the above-described processing, and the respective vector data relating to the amplitude component and the phase component determined by the vectorizing means. Is searched using the codebook, and the same or similar Vector quantization means for reading out index data relating to the amplitude component and the phase component associated with the vector data of the textbook, and outputting the index data read out by the vector quantization means to a transmission path. An audio signal encoding device. 2. The data separating means separates amplitude component data a (m) and phase component data θ (m) using MDCT coefficients x (n) and x (n + 1) adjacent in time series. 2. The audio signal encoding apparatus according to claim 1, wherein the audio signal is obtained by the following equations (1) and (2). (Equation 1) (Equation 2) (However, n and m are spectrum numbers, and n = 0, 2, 4, 6,...
.., N−1, m = 0, 1, 2, 3,..., (N−1) / 2, and N is the number of spectra in one frame. ) 3. The method of converting amplitude component data and phase component data into vector data by the vectorizing means, wherein the number of conversion dimensions in a low frequency band is smaller than the number of conversion dimensions in a high frequency band. The audio signal encoding apparatus according to claim 1 or 2. 4. When the vectorization means converts phase component data into vector data, the difference between the first data value and the subsequent data value of the phase component data group to be converted is defined as vector data. The audio signal encoding apparatus according to claim 1, 2 or 3. 5. The code book according to claim 1, wherein the number of vector data related to the phase component of the codebook is larger than the number of vector data related to the amplitude component.
Audio signal encoding device. 6. The number of conversion dimensions of phase component data into vector data by the vectorization means is made smaller than the number of conversion dimensions of amplitude component data into vector data, and the number of vector data related to the phase component of the codebook is reduced by the amplitude component. 6. The audio signal encoding apparatus according to claim 1, wherein the number of vector data is larger than the number of vector data. 7. A vectorization means for converting the amplitude component data into logarithm and converting it into vector data, and using a maximum power value in the vector data as a normalization coefficient to normalize vector data relating to other amplitude components. And a codebook is obtained by performing substantially the same data processing as the data processing by the data separating means, the vectorizing means, and the normalizing means on the MDCT coefficients obtained from the learning sample of the audio signal. The index data different from each other is stored in correspondence with the amplitude component data thus obtained, and the normalization coefficient of the normalization means is output to the transmission path together with the index data read by the vector quantization means. Claim 2, Claim 3, Claim 4,
The audio signal encoding device according to claim 5 or 6. 8. A decoding device to which index data output to a transmission path by an audio signal encoding device according to claim 1, 2, 3, 4, 5, or 6 is input. In the codebook, the same data as the storage data of the codebook provided in the encoding device is stored, and the codebook is searched using the input index data related to the amplitude component and the phase component. Vector inverse quantization means for reading vector data relating to the amplitude component and phase component matching the respective index data; and amplitude component data relating to the MDCT coefficient by combining the vector data read by the vector inverse quantization means. Vector synthesizing means for generating phase component data; and synthesizing the amplitude component data and the phase component data obtained by the vector synthesizing means to obtain MDCT coefficients. An audio signal decoding apparatus, comprising: a data synthesizing means to be obtained; and an inverse MDCT means for reproducing an audio signal modulated by the PCM method based on the MDCT coefficient obtained by the data synthesizing means. 9. A decoding apparatus to which index data and a normalization coefficient output to a transmission path by the audio signal encoding apparatus according to claim 7 are input, and wherein codebook storage data provided in the encoding apparatus is provided. Using the codebook storing the same data as the above and the input index data relating to the amplitude component and the phase component, the codebook is searched, and the vector data relating to the amplitude component and the phase component matching the respective index data are retrieved. Vector inverse quantization means for reading out the vector component, inverse normalization means for inversely normalizing the vector data relating to the amplitude component read by the vector inverse quantization means using the normalization coefficient, and the inverse normalization means The normalized vector data and the vector data related to the phase component read by the vector inverse quantization means are combined to generate the amplitude component data related to the MDCT coefficient. Vector synthesizing means for generating data and phase component data, data synthesizing means for synthesizing the amplitude component data and phase component data obtained by the vector synthesizing means to obtain MDCT coefficients, and MDCT coefficients obtained by the data synthesizing means. A decoding apparatus for an audio signal, comprising: an inverse MDCT means for reproducing an audio signal modulated by the PCM method on the basis of the above. 10. The data synthesizing means outputs amplitude component data a
10. The method according to claim 8, wherein x (n) and x (n + 1), which are MDCT coefficients, are obtained by the following Expressions 3 and 4, respectively, using (m) and the phase component data b (m). Sound signal decoding device. (Equation 3) (Equation 4) (However, n and m are spectrum numbers, and n = 0, 2, 4, 6,...
.., N−1, m = 0, 1, 2, 3,..., (N−1) / 2, and N is the number of spectra in one frame. )