JPH02165200A - Encoding/decoding device - Google Patents

Abstract

PURPOSE:To contrive the improvement of a voice quality even in a low bit rate by switching and quantizing in accordance with the pitch correlation of an input sound signal by two code books being a optimum to each, in the case when the periodicity of an input voice is strong and in other case. CONSTITUTION:In an encoding part (a), a vector generating part 7 fetches and outputs a vector from a first code book 91 and a second code book 92, in the case when it is shown that a pitch correlation coefficient is large and in other case, respectively, by a vector number and a long-term prediction switching signal inputted from an optimum vector selecting part 8. Also, in a decoding part (b) of an encoded signal, a vector generating part 15 fetches and outputs a vector from a first code book 181 and a second code book 182, in the case when it is shown that the pitch correlation coefficient is large and in other case, respectively, by an optimum vector number and a long-term prediction switching signal. In such a way, the quality of a satisfactory sound signal is obtained even in a low bit rate.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to an encoding/decoding device used for digitally transmitting audio signals.

(Prior Art) FIG. 2 is a block diagram of a conventional audio signal encoding/decoding device, and FIG. 1(a) shows an encoding section, and FIG. 1(b) shows a decoding section.

In the encoding unit shown in FIG. 3(a), the audio signal applied from the input terminal 1 is subjected to linear prediction analysis by the short-term prediction analysis unit at regular intervals, and the short-term prediction coefficients extracted thereby are processed by the short-term prediction analysis filter 3. and is input to the short-term prediction synthesis filter 4. [The period prediction analysis filter 3 removes the short-term correlated portion from the input audio signal using the applied short-term prediction coefficient, and inputs the signal to the long-term prediction analysis section 5. The long-term prediction analysis unit 5 extracts a pitch period and a pitch correlation coefficient from the input signal from which the short-term correlation portion has been removed, and applies them to the long-term prediction synthesis filter 6.

In this long-term prediction synthesis filter 6, the vector generation unit 7 gives a pitch correlation to the vector extracted from the codebook 9 using the input pitch period and pitch correlation coefficient according to the vector number applied from the optimal vector selection unit 8, It is input to the short-term prediction synthesis filter 4, where a short-term correlation is given by the short-term prediction coefficient input from the short-term prediction analysis section 2.

The difference distortion measure calculation unit 10 calculates the difference distortion measure between the input audio signal divided into fixed time periods and the signal input from the short-term prediction synthesis filter 4.

The optimal vector selection section 8 sequentially outputs vector numbers to the vector generation section 7 and the difference distortion measure calculation section 10.
Outputs the optimal vector number that minimizes the output of . In addition, as the difference strain measure mentioned above, for example, a squared distance or the like is used.

The multiplexing unit 11 combines the short-term prediction coefficients output from the short-term prediction analysis unit 2, the pitch period and pitch correlation coefficient output from the long-term prediction analysis unit 5, and the optimum vector number output from the optimum vector selection unit 8. Multiplex. The multiplexed signal is output from the output terminal 12 as encoded data.

ggl(b) in FIG. 2 is a block diagram showing a decoding section that decodes the output of the encoding section.

The encoded data having the configuration shown in FIG. 3A, which is applied from the input terminal 13, is separated by the separation unit 14 into an optimal vector number, a pitch period and a pitch correlation coefficient, and a short-term prediction coefficient. Part 15.
Long-term prediction synthesis filter 16. and is applied to the short-term prediction synthesis filter 17. The vector generator 15 extracts a vector from the codebook 18 according to the applied optimal vector number, and the long-term prediction synthesis filter 16
The input pitch cycle and pitch correlation coefficient give pitch correlation to the output of the vector generator 15, and the short-term prediction synthesis filter 17 applies short-term correlation to the output of the long-term prediction synthesis filter 16 according to the input short-term prediction coefficient. The audio signal is reproduced by inputting the input signal to the output terminal 19, and is output as a decoded audio signal from the output terminal 19.

As described above, the conventional device also calculates the coefficients of the short-term prediction synthesis filter and long-term prediction synthesis filter from the input audio signal, drives those synthesis filters, and minimizes the difference distortion measure between the output and the input audio signal. By selecting a vector, the input audio signal can be encoded, and the audio signal can be decoded by decoding the coefficients of the long-term prediction synthesis filter and short-term prediction synthesis filter and the vectors that drive these filters from the encoded data. Ta.

(Problem to be Solved by the Invention) However, the above-mentioned conventional encoding/decoding device encodes both voiced sound signals with periodicity and unvoiced sound signals without periodicity by referring to the same codebook. Because of this, it was not possible to obtain good quality audio signals at low bit rates.

Furthermore, for unvoiced sound signals without periodicity, there is a problem in that the pitch period and pitch correlation coefficient are sent out as useless information.

The present invention eliminates the above-mentioned problems and provides an audio signal code and code that can obtain good audio signal quality even at low bit rates.
The purpose is to provide a decoding device.

(Means for Solving the Problems) The object of the present invention is to divide an audio signal into cases where the pitch correlation coefficient is large and cases where the pitch correlation coefficient is small, and to encode the audio signal with reference to a codebook adapted to each case. This is achieved by increasing the number of vector quantization bits without performing long-term prediction when the pitch correlation coefficient is small.

(Function) With the above configuration, the present invention performs encoding with reference to a codebook that is adapted to each of a highly periodic audio signal and a weakly periodic audio signal, so that the quality of the encoded audio signal is improved, and For audio signals with weak periodicity, the audio quality can be further improved by increasing the vector quantization bit rate instead of using the pitch period and pitch correlation coefficient, which are useless information.

(Embodiment) FIG. 1 shows a block diagram of an audio signal encoding/decoding device according to an embodiment of the present invention, in which (a) shows the encoding section, (b)
) is a decoding unit, 91, 9 . are the first. This is the second codebook, and other code explanations refer to the corresponding code explanations in FIG. 2.

First, in the diagram (a) showing the encoding unit, the audio signal input from the input terminal 1 is subjected to linear prediction analysis by the short-term prediction analysis unit 2 at fixed time intervals, where short-term prediction coefficients are extracted and short-term prediction analysis is performed. The signal applied to the filter 3 and from which the short-term correlation has been removed is applied to the long-term predictive analysis section 5 . This long-term prediction analysis section 5 extracts the pitch period and pitch correlation coefficient from the signal from the short-term prediction analysis filter 3, and also outputs a long-term child 8111 switching signal indicating whether the correlation coefficient is larger or smaller than a predetermined value. of,
It is output to the vector generator 7 and the long-term prediction synthesis filter 6.

If the vector number inputted from the optimal vector selection section 8 and the long-term prediction switching signal indicate that the pitch correlation coefficient is large, the vector generation section 7 selects the first codebook 9. If not, the vector is extracted from the second codebook 9□ and output. The first codebook 9 is optimized for signals with a large pitch correlation coefficient, and the second codebook 98 is optimized for signals with a small pitch correlation coefficient.

Also, the second codebook 9. is larger than the first food book by the amount of encoded information of the pitch period and the pitch correlation coefficient.

If the input long-term prediction switching signal indicates that the pitch correlation coefficient is large, the long-term prediction synthesis filter 6
The pitch period and pitch correlation coefficient from the long-term prediction analysis section 5 are used to give periodicity to the output of the vector generation section 7 and output it, otherwise.

The output of the vector generator 7 is output as is. The short-term prediction synthesis filter 4 uses the short-term prediction coefficients from the short-term prediction analysis section 2 to give short-term correlation to the output of the long-term prediction synthesis filter 6 and outputs the result.

The difference distortion measure calculation unit 10 calculates the difference distortion measure between the input audio signal divided into fixed time periods and the output of the short-term prediction synthesis filter 4. Note that the difference skewness measure is 1, for example, a squared distance or the like.

The optimal vector selection section 8 sequentially inputs the vector numbers to the vector generation section 7, and outputs the vector number for which the output of the difference distortion measure calculation section 10 is the minimum as the optimal vector number.

When the long-term prediction switching signal from the long-term prediction analysis unit 5 indicates that the pitch correlation coefficient is large, the multiplexing unit 11 uses the long-term prediction switching signal, the short-term prediction coefficient, the pitch period, the pitch correlation coefficient, and the optimal vector. If not, the long-term prediction switching signal, short-term prediction coefficient, and optimal vector number are multiplexed and output from the output terminal 12 as a coded audio signal.

FIG. 1(b) is a block diagram of the decoding section for the encoded signal.

In this configuration, first, an encoded signal is input from the input terminal 13, and the long-term prediction switching signal is separated in the 1-separation unit 14. If it indicates that the pitch correlation coefficient is large, the remaining signal is input to the short-term prediction coefficient. , the pitch period and the pitch correlation coefficient, and the optimal vector number,
If not.

Separate only short-term prediction coefficients and optimal vector numbers.

If the optimum vector number and the long-term prediction switching signal indicate that the pitch correlation coefficient is large, the vector generation unit 15 generates the first codebook 18. from.

If not, a vector is extracted from the second codebook 18□ and output. These codebooks are the same as those of the encoding unit shown in FIG. When the long-term prediction switching signal indicates that the pitch correlation coefficient is large, the long-term prediction synthesis filter 16 gives a pitch correlation to the output of the vector generation unit 15 using the pitch period and the pitch correlation coefficient, and outputs it. If not, the output of the vector generator 15 is output as is. The short-term prediction synthesis filter 17 uses the short-term prediction coefficients to apply short-term correlation to the output of the long-term prediction synthesis filter 16, and outputs the output from the output terminal 19 as an audio signal.

As described above, the present invention provides an encoding/decoding device that can obtain good audio quality by switching the codebook for vector quantization depending on whether the pitch correlation of the input audio signal is large or small. Furthermore, when the pitch correlation coefficient is small, the speech quality is further improved by increasing the bit rate of vector quantization instead of encoding the pitch period and pitch correlation coefficient.

(Effects of the Invention) As is clear from the detailed explanation above, the present invention is capable of converting input audio signals by using two codebooks that are optimal for cases where input audio has strong periodicity and cases where it does not have strong periodicity. Since it switches and quantizes according to the pitch correlation, it can be encoded with good quality, and if the pitch correlation coefficient is small,
This has the effect of improving voice quality by increasing the bit rate of vector quantization instead of encoding the pitch period or pitch correlation coefficient.

[Brief explanation of the drawing]

FIGS. 1(a) and (b) are block diagrams showing an encoding/decoding unit according to an embodiment of the present invention, and FIGS. 2(a) and (b) are block diagrams showing a conventional encoding/decoding unit. It is. 2... Short-term prediction analysis section, 3... Short-term prediction analysis filter, 4.17... Short-term prediction synthesis filter, 5...
- Long-term prediction analysis unit, 6, 16... Long-term prediction synthesis filter, 7, 15... Vector generation unit, 8...
Optimal vector selection section, 9゜91, 9□, 18.18
．． , 1g, . . . code book, 10 . . . difference distortion measure calculation unit, 11 . Patent applicant Matsushita Electric Industrial Co., Ltd.

Claims (1)

[Claims] Find the coefficients of the short-term prediction synthesis filter and long-term prediction synthesis filter from the input speech signal, drive these short-term prediction synthesis filters and long-term prediction synthesis filter, and select the vector that minimizes the difference distortion measure between the output and input speech. The input speech is encoded by doing this, and from this encoded signal, the coefficients of the long-term prediction synthesis filter and the short-term prediction synthesis filter and the vectors that drive these filters are decoded.
In an encoding/decoding device that reproduces an audio signal, the audio signal is divided into large and small parts with respect to a fixed value of the pitch correlation coefficient, and is encoded using a codebook adapted to each of them. An encoding/decoding device characterized in that the bits and number of vector quantization are increased for encoding without long-term prediction, and the same codebook as above is used for decoding.