JP3089967B2 - Audio coding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech coding apparatus, and more particularly to a speech coding apparatus for coding a speech signal with a high quality in a short frame unit of 5 ms to 10 ms or less.

[0002]

2. Description of the Related Art As a method of encoding a speech signal, for example, an M-LCEP Speech by K. Ozawa et al.
Coding at 4kb / s with Multi-Mode and Mult-Codebook
”(IEICE Trans. Commun., Vol. E77-B, No. 9, pp. 1114-
1121, 1994). Less than,
An outline of the encoding method described in this paper will be described.

In this technique, every frame (for example, 4
At 0 ms), a linear prediction (LPC) analysis is performed on the audio signal to extract a spectrum parameter representing a spectral characteristic of the audio signal, and calculate a feature amount from a signal in a frame unit or a signal in which a perceptual weight is applied to the signal in a frame unit. Then, using the feature amount, for example, mode discrimination such as discrimination between a vowel portion and a consonant portion is performed, and encoding is performed by switching an algorithm or a code book according to the mode discrimination result.

[0004] The encoder further divides the frame into subframes (for example, 8 ms), and for each subframe, parameters in the adaptive codebook (delay parameters and gain parameters corresponding to the pitch period) based on past excitation signals. Is extracted, and the speech signal of the sub-frame is pitch-predicted by the adaptive codebook, and the residual signal obtained by pitch prediction is subjected to a speech codebook (a vector quantization code) including a predetermined type of noise signal. Book), an excitation signal is quantized by selecting an optimal excitation code vector and calculating an optimal gain.

[0005] The excitation code vector is selected so as to minimize the error power between the signal synthesized from the selected noise signal and the above-mentioned residual signal, and an index indicating the type of the selected code vector and a gain are selected. In addition, the spectrum parameters and the parameters of the adaptive codebook are transmitted in combination by the multiplexer unit.

[0006]

In the above-mentioned conventional speech coding method, it is necessary to shorten the frame length in order to reduce the processing delay. For example, when the frame length is reduced to 5 ms or less, However, there is a problem that the temporal change of the feature amount increases, which causes unstable and erroneous mode switching, thereby deteriorating sound quality.

In pitch extraction, T
Is calculated by calculating the frame length.
If ms, N = 40. That is, in the pitch extraction using the following expression, since the section length for calculating E _T is short, a pitch that changes greatly with time is required, and the sound quality also deteriorates.

[0008]

(Equation 1)

SUMMARY OF THE INVENTION It is an object of the present invention to provide a speech coding apparatus in which sound quality degradation due to erroneous mode discrimination does not easily occur.

It is another object of the present invention to provide a speech coding apparatus in which sound quality is hardly degraded due to erroneous pitch extraction.

[0011]

According to the first aspect of the present invention ,
Is (a) dividing means for dividing a speech signal into frame units of the signal of a predetermined unit, (b) a feature quantity obtained from the signal of the current frame which the dividing means is divided, dividing means less
Are calculated from the signal of the frame divided in the past for one frame.
A mode discriminating means for discriminating the mode of the audio signal using the weighted sum with the feature amount obtained, and (c) dividing means using an algorithm specified according to the mode discriminated by the mode discriminating means. It is provided to the speech coding apparatus and a coding means for coding the speech signal in units of frames and
You.

[0012]

[0013]

That is, according to the first aspect of the present invention, the characteristic amount (for example, the pitch prediction gain) obtained from the signal of the current frame and the characteristic amount obtained from the signal of the frame divided at least one frame in the past by the dividing means. By configuring the speech encoding apparatus to determine the mode of the speech signal using the weighted sum of the above, an erroneous mode discrimination does not occur.

[0015]

[0016]

[0017]

[0018]

[0019]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below in detail with reference to embodiments.

[0020]

FIG. 1 shows a schematic configuration of a speech coding apparatus according to an embodiment of the present invention. Hereinafter, with reference to this figure, illustrating the operation of the speech coding apparatus of the embodiment <br/>.

The frame dividing circuit 11 converts the audio signal input from the input terminal 50 into a frame (for example, 5 m
s), and a sub-frame division circuit 1
2 divides the frame output from the frame dividing circuit 11 into shorter frames (for example, 2.5 ms).

The spectrum parameter calculation circuit 13 cuts out the speech by applying a window (for example, 24 ms) longer than the subframe length to the speech signal of at least one subframe, and sets the spectrum parameter to a predetermined order. (For example, P = 10th order), and the spectrum parameter calculation circuit 13 of the embodiment is
It is configured to calculate spectral parameters using Burg analysis. For details of Burg analysis, see "Signal Analysis and System Identification" by Nakamizo (Corona, 1998).
Since it is described on pages 82 to 88, its description is omitted. It should be noted that, as the spectrum parameter calculation circuit, for example, a circuit that is calculated by another calculation method such as LPC analysis can be used.

Further, a spectrum parameter calculation circuit 1
3, the linear prediction coefficient α _i calculated by the Burg method
A process of converting (i = 1,..., 10) into LSP parameters suitable for quantization and interpolation is also performed. The conversion from the linear prediction coefficient to the LSP in the spectral parameter calculation circuit 13 of the embodiment is performed by the method of Sugamura et al.
SP) Speech Information Compression by Speech Analysis / Synthesis Method ”(Transactions of the Institute of Electronics, Information and Communication Engineers, J64-A, pp.599-606, 1981)
Year).

That is, the spectrum parameter calculation circuit 13 converts the linear prediction coefficients obtained by the Burg method in the second subframe into LSP parameters, obtains the LSP of the first subframe by linear interpolation, and obtains the LSP of the first subframe. Is transformed back to linear prediction coefficients, and the first,
Linear prediction coefficient α _il of two subframes (i = 1,..., 1
0, l = 1,..., 5) are output to the perceptual weighting circuit 17, and the LSPs of the first and second subframes are output to the spectrum parameter quantization circuit 14.

The spectrum parameter quantization circuit 14
This is a circuit for quantizing LSP parameters of a predetermined subframe. The spectrum parameter quantization circuit 14 of the embodiment quantizes the LSP parameters of the second subframe using vector quantization as a quantization method. It has become something. For the specific procedure of the vector quantization of the LSP parameter, see Japanese Patent Laid-Open No.
No. 0 (Japanese Patent Application No. 2-297600) and Japanese Unexamined Patent Application Publication No.
No. 363000 (Japanese Patent Application No. 3-261925),
JP-A-5-6199 (Japanese Patent Application No. 3-155049)
No.) and “LSP Coding Using” by T. Nomura
VQ-SVQ With Interpolation in 4.075 kbpsM-LCELP Sp
eech Coder "(Proc. Mobile Multimedia
Communications, pp. B.2.5, 1993).

Then, the spectrum parameter quantization circuit 14 restores the LSP parameters of the first and second subframes based on the LSP parameters quantized in the second subframe. In the spectral parameter quantization circuit 14 of the embodiment, the quantization L of the second sub-frame of the current frame is
The LSPs of the first and second sub-frames are restored by linearly interpolating the SP parameters and the quantized LSP parameters of the second sub-frame of the previous frame.

Here, LSP before quantization and LSP after quantization
After selecting one type of code vector that minimizes the error power from the SP, the LSP of the first to fourth sub-frames can be restored by linear interpolation. Further, in order to further improve the performance, after selecting a plurality of code vectors for minimizing the error power, for each candidate, the cumulative distortion is evaluated, and the combination of the candidate for minimizing the cumulative distortion and the interpolation LSP is determined. You can choose to do so.

Also, instead of linear interpolation, LSP interpolation patterns are prepared for a predetermined number of bits (for example, 2 bits), and the accumulation of the first and second subframes is performed for each of these patterns. A combination of a code vector and an interpolation pattern that minimizes distortion may be selected. By doing so, the transmission information increases by the number of bits of the interpolation pattern, but the temporal change in the LSP frame can be represented more precisely.

Here, the interpolation pattern may be created by learning using LSP data for training, or a predetermined pattern may be stored. In the latter case, for example, T. Taniguc
hi) et al. “Improved CELP sppech coding at 4kb
/ s and below ”(Proc. ICSLP, pp.41-44,
1992) can be used. Also,
Furthermore, in order to improve the performance, after selecting the interpolation pattern, LS
An error signal between the true value of P and the interpolation value of the LSP may be obtained, and the error signal may be further represented by an error codebook.

The spectrum parameter quantization circuit 14
LS of the first and second subframes restored in the above manner
P and the quantized LSP of the second sub-frame are calculated for each sub-frame by a linear prediction coefficient α _il ′ (i = 1,..., 10, l = 1,
.. 5), and outputs the result to the impulse response calculation circuit 16, and outputs an index representing the code vector of the quantized LSP of the second subframe to the multiplexer 28.

Further, the perceptual weighting circuit 17 calculates the perceptual weight from the spectrum parameter calculating circuit 13 for each subframe.
The linear prediction coefficient α _il before quantization (i = 1,..., 10, l =
1,..., 5) are input, perceptual weighting is performed on the audio signal of the subframe, and a perceptual weighting signal is output. The proposed mode-discriminating circuit 20 _A receives a perceptual weighting signal in units of frames from the perceptual weighting circuit 17, on the basis of the feature amount and the feature amount of the past one frame of the current frame, the mode determination.

FIG. 2 shows the configuration of the proposed mode discriminating circuit. As shown, the proposed mode discriminating circuit 20 _A
Is composed of a feature amount calculation circuit 31, a frame delay (D) 32, a weighted sum calculation circuit 33, and a mode discrimination circuit 34, and an audibility weighting signal is input from the input terminal 52 in frame units.

The feature value calculating circuit 31 converts the input information into
Calculate and output the pitch prediction gain G as a feature value
I do. In the weight calculation circuit 32, the feature amount calculation circuit 31
And the previous one stored in the frame delay unit 32
Weighted sum G with (past) frame features_AV(1)
It is obtained by the formula and output. Note that in equation (1), ν _i
Is a weight coefficient.

[0035]

(Equation 2)

The mode discriminating circuit 34 calculates the value G of the weighted sum.
_The mode is determined by comparing the _AV with a plurality of predetermined thresholds, and a mode determination result is output. For example, when dividing into four types of modes, three types of thresholds are set in the mode determination circuit 34. Mode discrimination result mode determination circuit 34 of this proposal-mode determination circuit 20 _A is outputted, as is shown in FIG. 1, is output to the adaptive code book circuit 22 and the audio quantization circuit 24 and the multiplexer 28.

The response signal calculation circuit 18 is stored for each subframe based on the linear prediction coefficient α _il from the spectrum parameter calculation circuit 13 and the linear prediction coefficient α _il ′ from the spectrum parameter quantization circuit 14. A response signal with the input signal d (n) set to “0” for one subframe is calculated using the value of the filter memory that is present, and is output to the subtractor 21. The response signal x _z (n) output from the response signal calculation circuit 18 is expressed by equation (2). In the expression (2), γ is a weighting coefficient for controlling the audibility weighting amount.

[0038]

(Equation 3)

The subtracter 21 receiving the output of the response signal calculation circuit 18 subtracts the response signal by one subframe from the perceptual weighting signal according to the equation (4), and outputs the result of the subtraction to the adaptive codebook circuit 22. .

[0040]

(Equation 4)

The impulse response calculation circuit 16 calculates z
The conversion calculates the impulse response h _w (n) of the weighting filter represented by the equation (4) by a predetermined number L, and outputs it to the adaptive codebook circuit 22 and the sound source quantization circuit 24.

[0042]

(Equation 5)

The adaptive codebook circuit 22 obtains a pitch parameter, performs pitch prediction according to equation (5), and outputs an adaptive codebook prediction difference signal Z (n). In Equation (5), b (n) is an adaptive codebook pitch prediction signal, and the adaptive codebook pitch prediction signal is
If β and T are the gain and delay of the adaptive codebook, V (n) is the adaptive code vector, and the symbol * is the convolution operation symbol, it is expressed by equation (6).

[0044]

(Equation 6)

The non-uniform pulse number type sparse excitation codebook 25 is a sparse codebook in which the number of non-zero components of each vector is different.
4, the non-uniform pulse number type sparse sound source codebook 2
The excitation code vector c _j (n) is selected for all or a part of the excitation code vector stored in No. 5 so as to minimize the expression (7).

[0046]

(Equation 7)

In the sound source quantization circuit 24, two or more types of code vectors are selected, so that one of the best code vectors is specified at the time of gain quantization described below. However, at the time of this selection,
The code vector may be specified as one type. When applying Equation (7) to only some of the excitation code vectors, a plurality of excitation code vectors are preliminarily selected, and Equation (7) is applied to the preselected excitation code vectors. It can also be applied.

The gain quantization circuit 26 reads the gain code vector from the gain codebook 27, and generates the excitation code vector from the excitation code vector selected by the excitation quantization circuit 24 so as to minimize the expression (8). The combination of the vector and the gain code vector is selected, and an index representing the selected excitation code vector and gain code vector is output to the multiplexer 28. In addition,
In the equation (8), β _K ′ and γ _K ′ are k-th code vectors in the two-dimensional gain codebook stored in the gain codebook 27.

[0049]

(Equation 8)

The weighting signal calculation circuit 19 reads out the code vector corresponding to each index based on the output parameters of the spectrum parameter calculation circuit 13 and each index, and firstly, based on the equation (9), the driving sound source signal v ( Find n).

[0051]

(Equation 9)

Next, the spectrum parameter calculation circuit 13
Output parameter and spectrum parameter quantization circuit 1
The weighting signal s _w (n) is calculated for each subframe by the equation (10) using the output parameter of No. 4 and the calculated weighting signal is output to the response signal calculation circuit 18.

[0053]

(Equation 10)

As described above, in the speech coding apparatus according to the first embodiment, the mode information averaged over a time length longer than the frame length is output, so that the sound quality due to an incorrect mode discrimination is output. Deterioration can be suppressed.

[0055]

[0056]

[0057]

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0098]

[0099]

[0100]

[0101]

[0102]

[0103]

[0104]

As described in detail above, claim 1 is as follows.
According to the inventions described in ( 2) and ( 3) , even if the frame length is reduced to 5 to 10 ms in order to reduce the delay, the sound quality does not deteriorate due to the temporal fluctuation of the mode discrimination, so that good sound quality can be maintained. become.

[0105]

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a speech encoding device according to an embodiment of the present invention .

[Figure 2] a block diagram showing the structure of a proposed mode-discriminating circuit used in the speech coding apparatus of the embodiment of the present invention Oh
You.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 11 Frame division circuit 12 Subframe division circuit 13 Spectrum parameter calculation circuit 14 Spectrum parameter quantization circuit 15 LSP codebook 16 Impulse response calculation circuit 17 Perception weighting circuit 18 Response signal calculation circuit 19 Weighting signal calculation circuit 20 Proposed mode discrimination circuit 21 Subtractor 22 Adaptive codebook circuit 23 Pattern storage circuit 24 Sound source quantization circuit 25 Non-uniform pulse number type sparse sound source codebook 26 Gain quantization circuit 27 Gain codebook 28 Multiplexer 31 Feature amount calculation circuit 32 Frame delay unit 33 Weighted sum calculation Circuit 34 mode discrimination circuit

Continuation of front page (56) References JP-A-2-139600 (JP, A) JP-A-6-4099 (JP, A) JP-A-6-222797 (JP, A) JP-A-7-225599 (JP) , A) (58) Fields surveyed (Int. Cl. ⁷ , DB name) G10L 11/00-21/06 H03M 7/30 H04B 14/04 JICST file (JOIS)

Claims (2)

(57) [Claims] 1. A dividing means for dividing an audio signal into a signal of a predetermined unit of frame, and a characteristic obtained from a signal of a current frame divided by the dividing means.
The amount of data collected and the dividing means
Weighting with features obtained from divided frame signals
A mode discriminator for discriminating a mode of the audio signal by using the sum; and an algorithm specified in accordance with the mode determined by the mode discriminator. An audio encoding apparatus comprising: encoding means for performing encoding. 2. A pitch prediction gain as the feature quantity.
The speech encoding device according to claim 1, wherein the speech encoding device is used.