JP2892462B2 - Code-excited linear predictive encoder - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a code-excited linear predictive encoder for encoding original speech.

[Prior Art] Since voice information has a large amount of information, an original voice is encoded and compressed when transmitting or performing recognition processing. As one of such encoding methods,
A code-excited linear predictive coding scheme has been proposed. For example, it is described in the following literature.

Literature [MRSchroeder, BSAtal
Co-author, "Code-Excited Linear Predictio"
n (CELP): High Quality Speech at Very Low Bit Rate
(Code Excited Linear Prediction: High Quality Speech at Low Bit Rate) ", Proc. ICASSP, pp 937-940, March 1985. FIG. 2 shows a conventional conceptual diagram of such a code excited linear predictive encoder. .

In FIG. 2, the code excitation linear predictive encoder 10
The original speech vector S is input to the input terminal 11. The original audio vector S is a vector composed of a plurality of adjacent samples when the original audio is sampled (the same sample may be an element of a plurality of vectors, and one sample is any one of the samples). It may be an element of only one vector). The original speech vector S is provided to the short-time analyzer 12 and the subtractor 18.

The short-time analysis unit 12 calculates a short-term prediction coefficient αj for the original speech vector S. One example is a short LPC
The short-term prediction coefficient αj is calculated by the analysis. The short-term analysis unit 12 calculates the short-term prediction coefficient αj thus obtained.
Is transmitted to the short-time synthesis filter 13 and the quantizer 14. In addition, the short-time analysis unit 12 sends the vector S1 from which the formant component (short-time component) of the voice has been removed to the long-time analysis unit 15.

The long-time analysis unit 15 performs a long-time analysis (LPC analysis as an example of the analysis) on the vector S1, calculates a pitch prediction coefficient β and a lag (pitch cycle) L, To the generator 14.

The quantizer 14 quantizes the short-term prediction coefficient αj, the pitch prediction coefficient β, the lag L, and the index I of the optimal excitation code vector described later, and outputs the result from the output terminal 112 as a total code C.

The index I given to the quantizer 14 is obtained by the processing of the excitation code table 17, the long-time synthesis filter 16, the short-time synthesis filter 13, the subtractor 18, the perceptual weighting filter 19, the square calculation unit 110, and the average operation unit 111. Is determined as follows.

In the excitation code table 17, each index i (i =
Excitation code vectors ei are stored in association with 1 to n), and each excitation code vector ei is applied to the long-time synthesis filter 16 in time sequence or collectively.

The long-time synthesis filter 16 performs inverse processing (synthesis processing) of the long-time analysis unit 15 on the excitation code vector ei using the pitch prediction coefficient β and the lag L, and converts the processed output vector STi into the short-time synthesis filter 13. To send to. The short-time synthesis filter 13 generates a short-time prediction coefficient α
The inverse processing (synthesis processing) of the short time analysis unit 12 is performed using j, and the processing output vector SWi is sent to the subtractor 18.

Each of the n vectors SWi means a synthesized sound (predicted vector) synthesized by each excitation code vector ei, and corresponds to the original speech vector S, respectively.

The subtracter 18 calculates each predicted vector SWi and the original speech vector S
And the resulting error vector e
ri is sent to the perceptual weighting filter 19.

The perceptual weighting filter 19 is provided in consideration of human perceptual characteristics, and weights the error vector eri in accordance with the perceptual characteristics to obtain a vector ε.
i is sent to the square calculator 110.

The square calculation unit 110 calculates the square of each component of the perceptually weighted error vector εi, and the average operation unit 111 calculates the average gi of the sum of the squares of the obtained components of the vector εi. Further, the average operation unit 111 includes n average values g1 to g.
The smallest average value gI (the one with the smallest prediction error) is detected from n, and the index is output from the excitation code table 17 to the quantizer 14 as the index I of the optimal excitation code vector.

A decoder (not shown) extracts the excitation code vector eI from the index I, and inverts the excitation code vector eI by the long-time analysis unit 15 using the pitch prediction coefficient β and the lag L (synthesis processing). ) And the inverse processing (synthesis processing) of the short-time analysis unit 12 using the short-term prediction coefficient αj is sequentially performed to obtain a reproduced sound Ss corresponding to the original sound S.

[Problems to be Solved by the Invention] As described above, in the conventional encoding method, a square calculation and an averaging process are performed in order to evaluate an optimal excitation code vector. However, for the squared signal, that is, the signal having twice the bandwidth of the original signal, the signal is actually extracted through the upper window, the sum is calculated, the average value is obtained, and the average value is used. When evaluating the optimal excitation code vector, there is a disadvantage that accurate evaluation cannot be performed due to the influence of the frequency characteristics of the window. That is, the quality of the reproduced sound is deteriorated.

The present invention has been made in view of the above points,
Provide a code excitation linear prediction encoder that can correctly evaluate a prediction vector based on each excitation code vector, detect a truly optimal excitation code vector, and thereby improve the sound quality of an encoded speech signal. What you want to do.

[Means for Solving the Problem] In order to solve the problem, according to the present invention, an analysis parameter obtained by analyzing an original speech vector and a plurality of excitation code vectors are synthesized using the analysis parameter. It outputs an index from an optimal excitation code vector that minimizes an error from the original speech vector among the plurality of obtained prediction vectors, and is a square vector of an error vector between each prediction vector and the original speech vector. In the code excitation linear predictive encoder that determines the optimal excitation code vector based on the weighting circuit, a weighting circuit that weights each square vector in consideration of the band is provided, and the optimal square vector is determined based on the weighted square vector. The excitation code vector is determined.

[Operation] According to the present invention, among the plurality of prediction vectors obtained by synthesizing a plurality of excitation code vectors using an analysis parameter obtained by analyzing an original speech vector, an optimum error from the original speech vector is minimized. The present invention relates to a method of determining an excitation code vector based on a square vector of an error vector between each prediction vector and an original speech vector. Then, instead of using the square vector of the error vector between each prediction vector and the original speech vector as it is as the square vector for determining the optimal excitation code vector, the weighting circuit converts each square vector into a band. After weighting in consideration of the above, an optimal excitation code vector is determined.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 shows the configuration of one embodiment of the present invention.

In the code example linear prediction encoder 20 of this embodiment as well, the short-time analysis unit 22 performs a short-time analysis (for example, a short-time analysis by LPC analysis) on the original speech vector S provided via the input terminal 21. The short-term prediction coefficient αj obtained as described above is supplied to the quantizer 24 and the short-time synthesis filter 23, and the long-term analysis unit 25 outputs the formant component (short-time component) of the speech supplied from the short-time analysis unit 22. The pitch prediction coefficient β and the lag (pitch period) L obtained by performing the long-term analysis (LPC analysis as an example of the analysis) on the removed vector S1 are given to the long-time synthesis filter 26 and the quantizer 24, and quantized. Vessel 24,
The configuration in which the short-term prediction coefficient αj, pitch prediction coefficient β, lag L, and index I of the optimal excitation code vector determined as described later are quantized and output from the output terminal 212 as a total code C is a conventional encoder. Same as 10.

However, the encoder 20 of this embodiment is different from the conventional encoder 10 in the configuration in which each excitation code vector ei is evaluated to obtain the index I of the optimal excitation code vector eI.

The excitation code table 27 has n indexes i
Excitation code vector e associated with (i = 1 to n)
i are stored in advance, and each excitation code vector ei is
The time-sequentially or collectively is supplied to the long-time synthesis filter 26.

The long-time synthesis filter 26 performs an inverse process (synthesis process) of the long-time analysis unit 25 on the excitation code vector ei by using the pitch prediction coefficient β and the lag L given from the long-time analysis unit 25. The output vector STi is sent to the short-time synthesis filter 23. The short-time synthesis filter 23 also has the vector S
Reverse processing (synthesis processing) of the short-time analysis unit 22 on Ti using the short-time prediction coefficient αj given from the short-time analysis unit 22
To obtain a processing output vector (excitation code vector ei
Is transmitted to the subtracter 28).

The subtractor 28 calculates each prediction vector SWi and the original speech vector S
And the resulting error vector e
ri is sent to the perceptual weighting filter 29. The perceptual weighting filter 29 is provided in consideration of human perceptual characteristics, weights the error vector eri in accordance with the perceptual characteristics, and obtains the obtained vector εi by a square calculator.
Send to 210. The square calculator 210 calculates the square of each component of the perceptually weighted error vector εi.

Also in the configuration for evaluating each excitation code vector ei, that is, the configuration for detecting the optimum excitation code vector and extracting its index, the configuration from the above-described excitation code table 27 to the square calculation unit 210 is the same as the conventional configuration. . That is, a prediction vector of the original speech vector is formed from each excitation code vector, an error vector of the prediction vector from the original speech vector is obtained, the error vector is weighted in consideration of perceptual characteristics, and then squared. The configuration up to this point is the same as the conventional one.

However, in the case of this embodiment, the square vector of the error vector εi output from the square calculating section 210 is
Instead of being directly given to 211, it is given to the averaging section 211 after being weighted by the weighting circuit 213.

The weighting circuit 213 performs a weighting operation shown in the equation.

γi (m) = (εi (m)) ² W (m) (where W (m) is a weighting function and m is the number of components of a vector) As the weighting function W (m), the attenuation amount at a high frequency is It is better to obtain a large number and have a narrow bandwidth. For example, a hamming window that is frequently used at present may be used.

The weighting circuit 213 is provided because the squared signal has twice the bandwidth of the original signal. Therefore, when the signal is simply extracted by a square window, the influence of the frequency characteristics of the window is obtained. This is because there is a risk that accurate evaluation may not be possible in response to this.

The average operation unit 211 calculates the average gi of the sum of the squares of the components of the obtained vector γi. Furthermore, the average operation unit 211
Detects the minimum average value gI (the one with the smallest prediction error) from the n average values g1 to gn, and uses the index as the index I of the optimal excitation code vector from the excitation code table 27 to the quantizer 24. Output.

The above encoding process is repeated every time the original speech vector S changes.

 Note that the configuration and processing of the decoder are the same as those in the conventional art.

Therefore, according to the above-described embodiment, when each excitation code vector is evaluated and the index of the optimal excitation code vector is detected, the squared vector whose band has been doubled is operated after the weighting process. Therefore, it is possible to correctly evaluate the error between the prediction vector and the original speech vector, and to execute optimal encoding. That is, the quality of the reproduced sound can be improved as compared with the related art.

Note that, in the above-described embodiment, an example in which the present invention is applied to an encoder having both a short-time analysis unit and a long-time analysis unit has been described. However, the present invention can be applied to at least an encoder having a short-time analysis unit. .

Further, the present invention can be applied to an encoder in which the excitation code table is of an adaptive type.

Further, the operation on the weighted square vector is
Not only the average operation but also a simple sum operation may be used.

If the band of the signal after the squaring process can be made equivalent to that of the embodiment, the weighting circuit 213 may be provided in a stage preceding the square calculating unit 210.

[Effects of the Invention] As described above, according to the present invention, after weighting the square vector of the error vector in which the perceptual characteristics are considered in consideration of the band, the representative value of the error vector (the square of each component) Since the average value and the sum of squares of each component are given to the calculation unit, the error between the prediction vector and the original speech vector can be correctly evaluated, and the optimum encoding can be executed. Quality can be improved than before.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a code excitation linear prediction encoder according to the present invention, and FIG. 2 is a block diagram showing a conventional encoder. 22 ... short-time analysis unit, 23 ... short-time synthesis filter, 24 ...
... Quantizer, 25 ... Long time analysis unit, 26 ... Long time synthesis filter, 27 ... Excitation code table, 28 ... Subtractor, 29
…… Perceptual weighting filter, 210… Square calculator, 211…
... Average operation unit, 213 ... Weighting circuit.

Claims (1)

(57) [Claims] An analysis parameter obtained by analyzing an original speech vector, and a plurality of prediction vectors obtained by synthesizing a plurality of excitation code vectors using the analysis parameter, the error from the original speech vector is minimized. And the index of the optimal excitation code vector
Error vector 2 between each predicted vector and the original speech vector
A code excitation linear predictive encoder that determines an optimal excitation code vector based on a square vector, comprising: a weighting circuit that weights each of the square vectors in consideration of a band, based on the weighted square vector. A code excitation linear predictive encoder characterized in that an optimal excitation code vector is determined by using the same.