JPH04114516A - Sound encoding device - Google Patents

Abstract

PURPOSE:To improve a quantizing system and to suppress the deterioration of the quality of sound by excecuting quantization through the use of an adaptive quantizer which adaptively changes bit distribution and deciding the bit distribution whose error becomes the minimum in an error evaluation unit. CONSTITUTION:The number of quantizing bits as against the code vector number of an adaptive code book 2 is set to nac, the number of the quantizing bits as against the multiplier of a multiplier 3 to nag, the number of the quantizing bits as against the code vector number of a noise code book 7 to be nwc and the number of the quantizing bits as against the multiplier of a multiplier 8 to be nwg. The adaptive quantizers 6 and 11 prepare plural combinations under a condition that the value of the total is constant and quantize the combinations of the number of the quantizing bits. The error evaluation unit 12 selects the combination of the number of the quantizing bits, in which the error becomes the smallest, and inputs information on the combination of the number of the quantizing bits to a multiplex unit 13. Thus, the deterioration of the quality of sound is suppressed even in the sound-encoding of a low bit rate.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a voice encoding device used in digital mobile communications and the like.

BACKGROUND ART FIGS. 3 and 4 show the configuration of a conventional speech encoding device. FIG. 3 shows an encoder, and FIG. 4 shows a decoder. In FIG. 3, numeral 31 is a linear prediction analyzer, which calculates linear prediction coefficients from the input audio signal. 32 and 37 are codebooks corresponding to sound sources, each of which stores a plurality of signal sequences (hereinafter referred to as code vectors) each having a length corresponding to a specific length section (hereinafter referred to as frame) to be analyzed. 32 is an adaptive codebook that is updated every frame and stores a plurality of code vectors obtained from past signal sequences, and 37 is a plurality of fixed code vectors obtained in advance from a white noise sequence. This is a noise codebook that stores . 33.38
is a multiplier that corrects the amplitude of each codevector in the codebook 32.37, and its output is sent to the linear predictive analyzer 3.
1 is input to a synthesis filter 34 and 39 using the linear prediction coefficients determined by 1. 35.40 calculates the error between the output of the synthesis filter 34.39 and the input signal sequence to be quantized (hereinafter referred to as input vector), and calculates the code vector number that minimizes the error and the multiplier of the multiplier 33.38. 36.41 is a fixed quantizer that quantizes the selected code vector number and the multiplier of multiplier 33.38 with a predetermined number of bits to obtain the code. . The quantized linear prediction coefficients, multiplier multipliers, and codes of the code vectors are assembled as transmission information in the multiplexer 42 and transmitted to the decoder.

In FIG. 4, 43 is a decomposer that separates the transmitted information into codes indicating linear prediction coefficients, multiplier multipliers, and code vector numbers; This is an inverse quantizer that calculates the number of . The code vector is selected from the adaptive codebook 45 and the noise codebook 48, the amplitude of which is corrected by multipliers 46 and 49, and input to the synthesis filter 50. The output of the synthesis filter 50 becomes decoded speech.

Next, the operation of the above conventional example will be explained.

In FIG. 3, the code vectors stored in the adaptive codebook 32 are amplitude-corrected by the multiplier 33, filtered by the synthesis filter 34, and the signal sequence ci
(i is the code vector number) is generated. The difference between this signal C1(i″11 and the input vector S[11 is 61
The error minimizer 35 determines the number of the code vector with the smallest error sum of squares and the multiplier to be multiplied by the multiplier 33. make changes. Further, the code vectors stored in the noise codebook 37 are amplitude-corrected by a multiplier 38 and filtered by a synthesis filter 39 to generate a signal sequence xi[n] (i is the code vector number).

The difference ei between this signal xi[n] and the difference signal dilp)
[i″+1 is determined, and the error minimizer 40 calculates the error 2
Code vector number and multiplier 3 with the smallest multiplicative sum
A multiplier to be multiplied by 8 is determined, and a fixed quantizer 41 performs quantization using a predetermined number of bits.

The quantized outputs from the fixed quantizers 36 and 41 and the outputs of the linear predictive analyzer 31 are multiplexed by a multiplexer 42, and the fourth
The signal is transmitted to the decoder shown in the figure.

In FIG. 4, the transmitted encoded speech is transmitted to the decomposer 4.
3, it is decomposed into three parts: a linear prediction coefficient, a multiplier multiplier, and a code indicating the code vector number, and the linear prediction coefficient is added to the synthesis filter 50.

The multiplier of the square multiplier and the code indicating the code vector number are sent to a dequantizer 44.47, where they are dequantized.The multiplier is sent to the multiplier 46.49, and the code vector number is sent to the adaptive codebook 45 and It is added to the noise codebook 48. Multipliers 46 and 49 correct the amplitudes of the code vectors, and a synthesis filter synthesizes them together with linear prediction coefficients to obtain decoded speech.

Problems to be Solved by the Invention However, in the conventional audio encoding device described above, parameters to be transmitted are quantized by a fixed quantizer according to a predetermined bit allocation, regardless of the characteristics of the audio. However, with low bit rate encoding, it is not possible to perform highly accurate encoding, and there is a problem in that the audio quality deteriorates.

The present invention solves these conventional problems,
It is an object of the present invention to provide an excellent speech encoding device that can suppress deterioration of speech quality.

Means for Solving the Problems In order to achieve the above object, the present invention compares the errors between the vector quantizer of the speech encoding device and the adaptive quantizer that adaptively changes the bit allocation, and This system is constructed with an error evaluator that determines the bit allocation that reduces the bit allocation, making it possible to suppress deterioration of audio quality even in low pit rate encoding.

Effects The present invention has the following effects due to the above structure. In other words, quantization is performed using an adaptive quantizer that adaptively changes bit allocation, and an error evaluator determines the bit allocation with the smallest error to improve quantization accuracy and prevent deterioration of audio quality. can be suppressed.

Embodiment FIGS. 1 and 2 show the structure of an embodiment of the present invention. FIG. 1 is an encoder, FIG. 2 is a decoder, and in FIG. 1, 1 is a linear prediction analyzer. 2 is an adaptive codebook, and 7 is a noise codebook, which are connected to multipliers 3 and 8, respectively. The outputs of multipliers 3 and 8 are input to synthesis filters 4 and 9, respectively. 5 and 10 are the code vector numbers that minimize the error between the output of the synthesis filters 4 and 9 and the input vector to be quantized, and the multiplier 3.
, 8 is an error minimizer. 6.11 is an adaptive quantizer that quantizes the selected code vector number and the multipliers of the multipliers 3 and 8 using the number of pits adaptively allocated for each frame to obtain a code. 12 is based on the error between the outputs of the synthesis filters 4 and 9 and the input vector,
This is an error evaluator that allocates the number of quantization pits to the adaptive quantizer 6.11. Information indicating the number of allocated pits is input to the multiplexer 13 together with the output of the linear prediction analyzer 1 and the outputs of the adaptive quantizers 6 and 11.

In FIG. 2, 21 is a decomposer that separates transmitted information; 14. Reference numeral 17 denotes an inverse quantizer that calculates the multiplier of the multiplier and the code vector number from the transmitted code and the number of allocated bits. The code vector is adaptive codebook 1
5, the noise code block 18 is selected, and the multiplier 16
．． The amplitude is corrected by 19 and input to the synthesis filter 20. The output of the synthesis filter 20 becomes decoded speech.

Next, the operation of the above embodiment will be explained. In FIG. 1, the number of quantization bits for the code vector number of adaptive codebook 2 is nac, the number of quantization bits for the multiplier of multiplier 3 is nag, and the number of quantization bits for the code vector number of noise codebook 7 is nwc, and the number of quantization bits for the number of stations of the multiplier 8 is nwg. In the adaptive quantizers 6 and 11, nac, nag, nwc, nwg under the condition that the values of naC + nag-1 - nWC-1 - nWg are constant.
A plurality of combinations are prepared, and quantization is performed for each combination of quantization bit numbers. Error evaluator 12
compares the error with the input vector for each combination of quantization bit numbers, selects the combination of quantization bit numbers that results in the smallest error, and multiplexes information indicating that combination of quantization bit numbers. input to the device 13. In the adaptive quantizers 6 and 10, the error evaluator 12
The code quantized by the number of quantization bits that results in the smallest error selected in is input to the multiplexer 13.

When determining the combination, for example, the number of pits required to quantize the code vector numbers Nac and Nwc is [tog * (1" Jac) x 1,000 bits [togz (
Nwc)] + t bits (X indicates the largest integer value not exceeding X), so the required number of pits is na
The remaining number of pits in c and nwc may be allocated to nag and nwg.

The operations of the other parts are the same as those shown in FIG. 3, so the explanation will be omitted.

In this way, the encoded audio is transmitted from the multiplexer 13.

The transmitted encoded audio signal is received by the decoder of FIG. The decomposer 21 converts the transmitted information into linear prediction coefficients,
It is decomposed into four pieces of information: the multiplier of the multiplier, a code indicating the code vector number, and information indicating the number of allocated bits. The information indicating the number of allocated bits is sent to the dequantizer 14.17 together with the multiplier of the multiplier and the code indicating the code vector number, and is subjected to dequantization processing. Since the subsequent operations are the same as those shown in FIG. 4, their explanation will be omitted.

As described above, according to the above embodiment, the adaptive quantizers 6 and 11 perform quantization according to the number of quantization bits that results in the least error, which is determined by the error evaluator 12.
This has the effect of suppressing deterioration of voice quality.

Effects of the Invention As is clear from the above embodiments, the present invention adaptively allocates the number of quantization bits in vector quantization, and performs quantization with the number of quantization bits that minimizes the quantization error. This has the advantage that deterioration in audio quality can be suppressed even in low bit rate audio encoding.

[Brief explanation of the drawing]

FIG. 1 is a schematic block diagram of an encoder of a speech encoding device according to an embodiment of the present invention, FIG. 2 is a schematic block diagram of a decoder of the same device, and FIG. 3 is a schematic block diagram of an encoder of a conventional speech encoding device. FIG. 4 is a schematic block diagram of a conventional decoder of the same device. 1... Linear prediction analyzer, 2, 15. Adaptive code footer, 3. 8.16.19... Multiplier, 4,9.20.
・Synthesis filter, 5, 10...Error minimizer, 6, 1
1 Adaptive quantizer, 7, 18... Noise codebook, 12. Error evaluator, 13... Multiplexer, 14.17
...Dequantizer, 21...Decomposer. Name of agent: Patent attorney Akira Kokaji and two others

Claims (1)

[Claims] a linear prediction analyzer that calculates linear prediction coefficients from input audio;
The adaptive codebook, the noise codebook, and an adaptive quantizer that adaptively changes the number of quantization bits for the adaptive codebook and the noise codebook are compared with each other to determine the bit allocation that minimizes the error. A speech encoding device having an error evaluator.