JP3060229B2 - Signal processing device and signal processing method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a signal processing device and a signal processing method for compressing and outputting an input signal.

[Summary of the Invention]

The present invention divides a signal into a plurality of frequency bands, treats each frequency band signal as a vector for each predetermined unit, determines the classification of each vector based on the properties of these vectors, and determines between the vectors and between each vector. A codebook corresponding to the classification is selected from a number of codebooks independent of the number of frequency bands that can be commonly selected between frequency bands, and each vector is vector-quantized using the selected codebook. Accordingly, the present invention provides a signal processing device and a signal processing method that can reduce the size of the device configuration or perform more adaptive vector quantization.

[Conventional technology]

As one method of high-efficiency signal encoding, while dividing the input signal into multiple channels on the time axis or frequency axis,
There is an encoding technique based on bit allocation (bit allocation) that adaptively allocates the number of bits for each channel. For example, a coding technique based on the above-mentioned bit allocation of a voice signal includes a band division code (sub-band coding: SBC) for dividing a voice signal on a time axis into a plurality of frequency bands and coding the signal, So-called adaptive transform coding (ATC), which transforms a signal of interest into a signal on the frequency axis (orthogonal transform), divides the signal into a plurality of frequency bands, and adaptively codes each band, or the above-mentioned SBC and so-called adaptive prediction. So-called adaptive bit allocation, which combines with coding (APC), performs band division on signals on the time axis, converts each band signal to baseband (low band), and then performs multi-order linear prediction analysis to perform predictive coding. (APC-AB).

Here, the above-mentioned band division coding is performed by a signal transmission device as shown in FIG. In FIG. 3, a digital audio signal supplied to an input terminal 110 is first converted into a frequency division filter (for example, QMF: quadrature mi
Bandpass and low-frequency conversion are performed by mirror filter groups 131 _{1 to} 131 _n such as rror filters. That is, in the frequency division filter groups 131 _{1 to} 131 _n , after being band-divided by a band-pass filter (band-pass filter: BPF), each signal is passed through a low-pass filter and passed through the center frequency of the pass band. , And these signals are down-sampled at quantizers 134 _{1 to} 134 _n at an appropriate sampling frequency. Each signal subjected to data compression by being requantized in this manner is output from the terminal 138 via the multiplexer 136, transmitted to the terminal 148 of the decoder 140 via the transmission path, and transmitted from the terminal 148. Dequantizers 144 _{1 to} 144 _n via a demultiplexer 149
After that, the signals are converted into respective band signals on the time axis by the frequency converters 142 _{1 to} 142 _n , and are output from the terminal 150 as decoded voice signals through the adder 146.

Here, in the data compression processing of the signal by the encoder 130, quantization bits are adaptively allocated to each band so as to reduce the influence of noise generated by data compression on the decoded speech signal. Improving quality. The decoder 140 also decodes the signal based on the bit allocation information transmitted from the encoder 130.

Conventionally, in order to obtain the above-mentioned bit allocation information, a method of transmitting energy value information of each band as auxiliary information (side information) separately from a signal of each band has been adopted. That is, in this method, the encoder
The energy detection means 133 _{1 to} 133 _n calculate the energy value of the signal of each band from each signal divided into bands by each of the frequency division filter groups 131 _{1 to} 131 _n of 130, and assign / step calculation based on the calculated value. Means 135 is used to calculate the optimal number of bits to be allocated when quantizing the signal of each band and the quantization step, and using this result, the signal of each band is regenerated by quantization 134 _{1 to} 134 _n. Quantized. Further, the output signal of the allocating / step calculating means 135, that is, auxiliary information, is sent to the allocating / step calculating means 145 of the decoder 140, and the information from the allocating / step calculating means 145 is dequantized 144 _{1 to} 144
_n , where the signal is decoded by performing a process reverse to that of the quantizers 134 _{1 to} 134 _n described above.

In such band division coding, noise shaping or the like can be considered in accordance with human auditory characteristics, and a band in which voice energy is unbalanced or large, or a band that greatly contributes to subjective quality such as intelligibility. Can assign more information. The signal of each band is quantized and dequantized with the allocated number of quantization bits, whereby the degree of auditory disturbance of quantization noise can be reduced, and the number of bits can be reduced as a whole. Further, by performing band division coding, quantization noise is generated only in the divided band, and does not affect other bands. The method of transmitting the energy value information as auxiliary information as described above has an advantage that the energy value of the signal of each band is simultaneously used as the quantization step width (normalization factor) of the signal of each band.

[Problems to be solved by the invention]

In the above-described band division coding (SBC), the time waveform of each band divided by the frequency band of the input signal is quantized independently and adaptively by the number of allocated bits for each channel. The quantization is performed by vector quantization or the like in which a series of sample values (connection of a plurality of words) is divided into an appropriate size, collected into one block, and quantized as one vector.

In this vector quantization, an identification code (index) corresponding to a code vector (representative vector) most similar to the vector of the input signal (for example, the closest distance) is obtained as an output after quantization. The code vector and the identification data are stored in a memory generally called a code book.

For this reason, when the above-described vector quantization is applied to band division coding, conventionally, the number of codebooks for the vector quantization, that is, the codebook, according to the number of band divisions of the band division coding, The size of the memory (capacity of the memory or the number of memories) for the above has been determined.

By the way, in the above-described band division coding, for example,
When the number of band divisions is large, the number of codebooks according to the number of band divisions is required as described above,
The memory for the code book must be increased, and the device becomes larger. Also, for example,
When the number of band divisions is small, the number of codebooks is also small, so that it is difficult to perform adaptive quantization according to the characteristics of an input signal.

Therefore, the present invention has been made to solve such a problem, and the memory for the codebook can be reduced even if the number of band divisions is large. It is an object of the present invention to provide a signal processing device and a signal processing method capable of performing quantization according to the properties of a signal.

[Means for solving the problem]

The signal processing device of the present invention has been proposed in order to achieve the above-described object, and includes a dividing unit that divides a signal into a plurality of frequency bands, and a signal in each of the plurality of frequency bands for each predetermined unit. Type determining means for treating each vector as a vector and determining the classification of each vector based on the properties of these vectors, and a code of a number independent of the number of frequency bands which can be commonly selected between the respective vectors and between the respective frequency bands. A book, a code book according to the determined classification, and a quantization unit that performs vector quantization of each of the vectors using the selected code book.

Further, a signal processing method of the present invention has been proposed to achieve the above-mentioned object, and divides a signal into a plurality of frequency bands, and converts the signals of the plurality of frequency bands into a vector for each predetermined unit. The classification of each vector is determined based on the properties of these vectors, and the above-described determination is made from a number of codebooks independent of the number of frequency bands that can be commonly selected between the vectors and between the frequency bands. A codebook corresponding to the selected classification is selected, and each vector is vector-quantized using the selected codebook.

[Action]

According to the present invention, each vector of a plurality of (N) bands is classified into a plurality of types (n) based on the properties of these vectors.
And the respective vectors are quantized by the codebooks corresponding to the n types, so that the number of codebooks can be set arbitrarily regardless of the number of bands to be divided.

〔Example〕

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

FIG. 1 shows a schematic configuration of an apparatus according to an embodiment of the present invention. The apparatus of the embodiment shown in FIG. 1 is preferably applied to an encoding apparatus for performing the above-described adaptive bit allocation, for example, an apparatus for performing band division encoding shown in FIG.

In the apparatus shown in FIG. 1, first, an input signal via an input terminal 1 is converted into a band-pass filter (BPF) 10 which is N band dividing means for dividing the input signal into a plurality of (N) frequency bands. It is divided into N channels by ₁ to 10 _N.
Each signal of the N-channel, after being sub-sampled by the quantizer 11 ₁ to 11 _N, the respective data is sent to the gain calculator 12 ₁ to 12 _N and a divider 13 ₁ to 13 _N. here,
In the above gain calculator 12 ₁ to 12 _N, and m sample (word) 1 block of each data, determine the gain (RMS value) of the respective blocks. The dividers 13 _{1 to} 13 _N perform an operation of interrupting each word (sample) of each block with the RMS value. That is, in these gain calculator and divider, normalization (gain normalization) of each block is performed.

Next, each block which these normalized is performed is sent to the type determination circuit 14 ₁ to 14 _N and the vector quantizer 15 ₁ to 15 _N.

The type determination circuit 14 ₁ to 14 _N, each band (channel)
, The data of a plurality of words (m words, m samples) are treated as vectors (input vectors), the properties of these input vectors are determined, and each input vector is determined based on the determined properties of the input vectors. into a plurality of types (n types), and selects the codebook 20 ₁ to 20 _n corresponding to each type. That is, from this kind discriminating circuit 14 ₁ to 14 _N, the codebooks 20 ₁ to 20 _n
To being adapted to index representing the selected code book is sent, by index indicating the codebook is sent to the code book 20 ₁ to 20 _n, from the code book 20 ₁ to 20 _n The code vector in the selected code book will be output. In addition,
Index indicating the respective codebooks, output terminals 2 ₁
1 to 2 _N to a decoding device having a configuration reverse to that of the device shown in FIG. 1, in which an index indicating each mode book is used for decoding a signal. ing.

Furthermore, the vector quantizer 15 ₁ to 15 _N is arranged to perform a vector quantization using the code vector of the codebook which has been selected as described above. That is, the in the vector quantizer 15 ₁ to 15 _N, compared with the type code vector (representative vector) of the determination circuit 14 ₁ in the codebook selected by to 14 _N and the input vector is carried out. For example distance calculation that distance by performing (similarity) are compared, shortest distance (similar) code vector and the corresponding identification code (index indicating the code vectors) of these vector quantizers 15 ₁ to 15 _N
Is obtained as a quantized output of Incidentally, the quantization output is made from the output terminal 3 ₁ to 3 _N as supplied to the decoding apparatus, the signal is decoded using the quantization output in the decoding apparatus.

Here, a codebook for performing vector quantization by the above-described band division coding or the like is usually designed for each divided channel. Instead of a code book for each channel, a code book 20 for each type classified into a plurality of types (n types) based on the statistical properties of each channel output (subband signal) in units of one block (m samples).
_{1 ~20 n} are designed.

When the outputs of the N channels are classified into n types, for example, a statistical property obtained from a first-order correlation coefficient ψ (τ) shown in the first equation can be used as a classification standard.

However, τ = 1 (one sample shift). That is, the output of the N channel is converted to the first order correlation function ψ
By using the magnitude of (τ), the magnitude of the above RNS value, or a combination thereof, it can be classified into n types.

By the way, in the conventional band division coding or the like, a signal is divided into a large number of bands, and as described above, the number of codebooks needs to be a number corresponding to the large number of band divisions. Therefore, a large-capacity memory is required for the code book, and the device becomes large in size.

Therefore, in the present embodiment, the number of classifications n (the number of codebooks n) of the plurality of types is N> n for such a large number of band divisions N. That is, the first
In the apparatus of figure performs vector quantization using less of n codebooks 20 ₁ to 20 _n than the number of subbands N. Here, as described above, reducing the number n of codebooks means that the number of code vectors used in vector quantization is reduced. However, there is a possibility that the quantization distortion increases. In the present embodiment, as described above, since the codebook corresponding to the property of the input vector is selected, and the vector quantization is performed using the codebook, the quantization distortion is not increased. It does not hinder adaptive quantization. Further, for example, even the band dividing number N is large, it is possible to reduce the number of codebooks 20 ₁ to 20 _n, it is possible to reduce the memory capacity for codebook is possible to prevent the size of the device it can. Further, it is possible to contribute to downsizing of the device.

On the other hand, in the conventional band division coding, for example, when the number of band divisions is small, the number of codebooks corresponding to the number of band divisions is also small. Quantization becomes difficult to perform.

In the apparatus of the present embodiment, when the number N of band divisions is small (small), the number of classifications n (the number n of codebooks) is set to N <n, and n codebooks 20 ₁ larger than the number N of divisions are set. Vector quantization is performed using 2020 _n . That is, without being bound by the band dividing number, the band than dividing number N many types of codebooks 20 ₁ to 20 _n using a dynamic codebook, select the codebook in line with statistical properties of the signal the By using the selected codebook at the time of vector quantization, more adaptive quantization becomes possible, and the performance of the apparatus can be improved. For example, in band division coding for dividing into four bands,
The signals (normalized (gain-normalized)) of each band are classified (classified) into about eight types, and the codebook of vector quantization is adaptively switched according to the class, so that more adaptive signal processing can be performed. .

Here, the codebook 20 ₁ to 20 _n can be designed by using a configuration as shown in Figure 2.

In FIG. 2, training data (training set) having the same statistical properties as the input signal is stored in a memory 50.
The band is divided into N channels through the same band pass filters (BPF) 51 _{1 to} 51 _N as shown in FIG. Divided data in these N channels, which is also similar to the quantizer 52 ₁ to 52 _N and the device of FIG. 1, a normalization circuit 53 ₁ to 53 _N formed by the gain calculator and a divider, etc. , Are subjected to subsampling and normalization (gain normalization), and then sent to the classification circuits 54 _{1 to} 54 _N. This classification circuit 54 _{1 to} 5
4 _N is a type determination circuit 14 ₁ to 14 _N and the same criteria in the first diagram classifying the training data. That is, the statistical properties of the data in units of one block (m words) of each channel data from the memory 50 are determined, and the data is classified into a plurality of types (n types) according to the statistical properties.

In this way, the training data of each channel is classified into n type are collected for each data having the same nature, to be a new n type training set 1 training set n memory 60 ₁ to 60 _n Stored, and n codebooks are designed using the n types of training sets. Note that each codebook can be designed, for example, with a so-called multi-stage or the like so as to support dynamic bit allocation.

〔The invention's effect〕

In the signal processing device and the signal processing method of the present invention,
For example, even when the number of band divisions is large, the number of codebooks can be reduced without deteriorating the quantization efficiency, and the memory for the codebook can be reduced. Is possible. Further, even when the number of band divisions is small, adaptive vector quantization according to the properties of the input signal is performed by increasing the number of codebooks and selecting a codebook according to the properties of the vector. be able to.

[Brief description of the drawings]

FIG. 1 is a block circuit diagram showing a schematic configuration of a signal processing device according to one embodiment of the present invention, FIG. 2 is a block circuit diagram showing a configuration for designing a codebook used in the device of the embodiment, and FIG.
FIG. 1 is a block circuit diagram showing a schematic configuration of an apparatus for performing band division coding. 10 ₁ to 10 _N , 51 _{1 to} 51 _N …… Band pass filter 11 _{1 to} 11 _N , 52 _{1 to} 52 _N … Quantizer 12 _{1 to} 12 _N … Gain calculator 13 _{1 to} 13 _N … Division vessel 53 ₁ ~53 _n ...... normalization circuit 14 ₁ ~14 _n ...... kind discriminating circuit 15 ₁ ~15 _n ...... vector quantizer 54 ₁ ~54 _n ...... classification circuit 20 ₁ ~20 _n ...... codebook 50,60 _{1 to} 60 _n …… Memory

Continued on the front page (72) Inventor Makoto Akune 6-7-35 Kita-Shinagawa, Shinagawa-ku, Tokyo Sony Corporation (72) Inventor Yoshihito Fujiwara 6-35, Kita-Shinagawa, Shinagawa-ku, Tokyo Sony Stock In-house (56) References JP-A-63-198478 (JP, A) JP-A-62-139089 (JP, A) JP-A-64-29121 (JP, A) (58) Fields investigated (Int. . ^7, DB name) H03M 7/30

Claims (2)

(57) [Claims] A dividing unit that divides a signal into a plurality of frequency bands; and a signal in each of the plurality of frequency bands is treated as a vector for each predetermined unit, and a classification of each vector is determined based on a property of the vector. Type determination means, codebooks independent of the number of frequency bands selectable in common between the vectors and frequency bands, and selection means for selecting a codebook according to the determined classification And a quantizing means for vector-quantizing each of the vectors using the selected codebook. 2. A signal is divided into a plurality of frequency bands, the signals in each of the plurality of frequency bands are treated as vectors for each predetermined unit, and the classification of each vector is determined based on the properties of these vectors. From the number of codebooks independent of the number of frequency bands that can be commonly selected between each vector and each frequency band, a codebook corresponding to the determined classification is selected, and each of the vectors is selected as described above. A signal processing method characterized by performing vector quantization using a codebook.