JPH06130998A - Compressed voice decoding device - Google Patents

Abstract

PURPOSE:To improve the quality of a reproduced voice signal even when a present frame is a bad frame with information omitted. CONSTITUTION:When the present frame is the bad frame and some interpolation processing is not performed, a voice is instantaneously absent and the quality of the reproduced voice signal is degraded. The information of frames in the past can be utilized for the interpolation but even in this case, the interpolation is not satisfactorily performed and the quality of the reproduced voice signal is degraded sometimes. Therefore, when the present frame is the bad frame, an interpolation processing part 23a interpolates the information of the present frame by using the information of frames in the past and future rather than the present frame, and the quality of the reproduced voice signal is improved by performing decoding processing while using the interpolation information.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a compressed speech decoding apparatus, for example, a code excitation linear predictive coding system (CEL).
It is applicable to a digital mobile communication system using P).

[0002]

2. Description of the Related Art Conventionally, TDMA (time division multip) is used.
In a digital mobile communication system using the (le access) method, a voice signal digitized by an analog / digital converter or the like has a length corresponding to a fixed time length (voice frame: hereinafter simply referred to as a frame). It is processed for each data. For such a data series for each frame,
A compression coding process (information source coding process) and an error correction coding process (communication path coding process) are applied. In a digital mobile communication system using the TDMA method, the data subjected to these processes is applied to the TDMA transmission frame format and transmitted. On the receiving side, the TDMA frame data of the corresponding channel is demodulated, and after the error correction decoding processing, the audio signal for each frame is sequentially reproduced by the compression audio decoding device.

In such a digital mobile communication system, a system based on a code excitation linear predictive coding system is adopted as a high compression coding system for voice signals. It is known that this method can obtain high-quality reproduced voice at a coding rate of 4 kbit / s to 8 kbit / s.

The following documents (1) and (2) can be cited as references for the digital mobile communication system using the TDMA system and the code-excited linear predictive coding system, respectively.

Reference (1): NSJayant and JHChen, "
Speech Coding with Time-VaryingBit Allocations to
Excitation and LPC Parameters ", Proc.ICASSP, pp65-6
8,1989. (2): EIA / TIA Spec. "DUAL MODE BASE STARION COMP
ATIBILITY STANDARD "IS-54, pp64,1991.

[0006]

However, in digital mobile communication, there is a fading phenomenon due to multiple reflection of radio waves, and this phenomenon remarkably deteriorates the radio wave reception sensitivity depending on the position of the mobile device. When the mobile unit moves in such a radio wave state, a data error depending on the moving speed of the mobile unit occurs when the received signal is demodulated. In general, error correction coding is applied to transmission data, so that a slight data error can be corrected on the receiving side. However, the effect is lost for the data burst error.

Therefore, it is unavoidable that a frame (hereinafter referred to as a bad frame) in which the audio data cannot be decoded occurs, and the reproduced audio quality is significantly deteriorated.

Therefore, it is conceivable to perform the interpolation process using the data already received for the bad frame (for example, the data of the previous frame), but even in this case, the reproduced voice quality is still deteriorated. Met.

The present invention has been made in consideration of the above points, and provides a compressed speech decoding apparatus capable of favorably interpolating a bad frame and obtaining a reproduced speech signal of high quality. It was something I tried to do.

[0010]

In order to solve such a problem, in the first aspect of the present invention, there is provided a compressed speech decoding apparatus for decoding compressed speech coded data formed by processing for each speech frame. , When the decoding processing target frame is a bad frame in which the received information is missing,
Interpolation processing means for interpolating the information of the decoding target frame using the information of the past frame and the future frame from the decoding target frame is provided.

According to the second aspect of the present invention, compressed speech coded data formed by processing for each speech frame, wherein vocal tract information is a code formed by multi-stage vector coding of line spectrum pair parameters. In a compressed speech decoding device for decoding compressed speech coded data including, for at least the first-stage line spectrum pair parameter subjected to multi-stage vector decoding, holds information of frames before and after the decoding processing target frame, When the decoding process target frame is a bad frame in which the reception information is missing, the line spectrum pair parameter of the decoding process target frame is changed from the line spectrum pair parameters held for the frames before and after the decoding process target frame. An interpolation processing means for interpolating is provided.

[0012]

When the decoding processing target frame is a bad frame in which the received information is missing, the sound is momentarily lost and the quality of the reproduced sound signal deteriorates unless any interpolation processing is performed. It is possible to use the information of the already received frame past the decoding target frame, but if the change of the original audio signal from the past frame to the decoding target frame is large, interpolation is good. The quality of the reproduced audio signal is deteriorated.

Therefore, in the first aspect of the present invention, when the decoding processing target frame is a bad frame in which the received information is missing, the interpolation processing means has past frames and future frames from the decoding processing target frame. The information of the decoding processing target frame is interpolated using the information of 1. and the decoding processing is performed by using this interpolation information to improve the quality of the reproduced audio signal. In this way, since the information of the frame in the future from the decoding processing target frame is used,
The compressed audio decoding device does not set the frame of the input compressed audio encoded data as the decoding processing target frame, but sets the frame of the compressed audio encoded data input before that as the decoding processing target frame. To do.

In the second aspect of the present invention, compressed speech coded data (for example, vocal tract information and excitation source information are separately coded) (for example,
Code excitation linear predictive coding method), and the decoding structure of vocal tract information in the target
The same technical idea as that of the present invention is applied. As the vocal tract information, it is known that the line spectrum pair parameter has good interpolating characteristics, and it has already been made that this line spectrum pair parameter is multistage vector quantized and inserted into the compressed speech coded data. The second aspect of the present invention relates to a compressed voice decoding device for receiving such compressed voice coded data.

In the second aspect of the present invention, the interpolation processing means is
When at least the first-stage line spectrum pair parameter subjected to multi-stage vector decoding holds the information of the frames before and after the decoding process target frame, and the decoding process target frame is a bad frame in which the reception information is missing Then, the line spectrum pair parameter of the decoding process target frame is interpolated from the line spectrum pair parameter held for the frames before and after the decoding process target frame, and is used for the decoding process.

[0016]

EXAMPLES The present invention will be described in detail below based on examples.

(A) Schematic Configuration of Digital Mobile Communication System First, a schematic configuration and operation of a digital mobile communication system to which an embodiment is applied will be described. Here, FIG. 1 is a block diagram showing the configuration of this system.

In FIG. 1A showing the transmitting side, a series of discrete audio signals input from the input terminal 11 is accumulated in the buffer 12 for the frame length, and thereafter processed in units of this frame length.

The audio data corresponding to one frame length is compression-encoded by the audio compression encoding unit 13. Here, the voice compression encoding unit 13 obtains compressed voice encoded data by performing a process such as a linear prediction analysis to determine a parameter indicating a feature of the voice based on, for example, a code excitation linear predictive encoding system. .

The voice data encoded by the voice compression encoding unit 13 is added with redundant data for error correction and error detection in the error correction encoding unit 14, and the data order is changed according to a predetermined rule. It is given to the TDMA frame format processing unit 15 after being subjected to the interleaving process, converted into a corresponding channel format by the TDMA frame format processing unit 15, and given to the modulation unit 16. Modulator 1
In 6, in order to wirelessly transmit the coded data that has also undergone the channel coding process, a modulation process for converting a baseband signal into a wireless band signal and a power amplification process are performed.
It is transmitted from the transmission terminal (antenna) 17.

In FIG. 1B showing the receiving side, the received signal captured by the receiving terminal (antenna) 18 is demodulated by the demodulating section 19 to convert it from the radio frequency band to the base band signal band. From this demodulated signal, the TDMA frame format processing unit 20 selects the corresponding channel format signal and sends the demodulated data to the error correction decoding unit 21.

In the error correction decoding unit 21, the data is deinterleaved to restore it to the original time order, and then the error correction is performed to decode it into compressed speech coded data. In this error correction processing, when a fatal error (error that cannot be corrected) is detected in the data of the frame currently being decoded (current frame), the error correction decoding unit 21 performs compressed voice decoding. The error detection information 25 is transmitted to the conversion unit 23.

The compressed speech coded data obtained by the error correction decoding unit 21 is delayed by being accumulated in the buffer 22 for one frame and is subjected to processing by the compression decoding unit 23. In the compressed voice decoding unit 23, the decoding process is switched according to the error detection information 25.

When the error detection information 25 does not instruct the error detection, the compressed audio decoding unit 23 determines the compressed audio encoded data of the current frame, that is, the error correction decoding unit 21.
From the viewpoint, the compressed voice coded data delayed by one frame time is decoded to generate a reproduced voice signal and output from the output terminal 24.

Of course, the compressed voice decoding unit 23 corresponds to the voice compression encoding unit 13 on the transmission side, and the voice compression encoding unit 13 has an encoding configuration based on the code excitation linear predictive encoding system. If so, the compressed voice decoding unit 2
3 also has a decoding structure based on the code-excited linear predictive coding system.

On the other hand, when the error detection information 25 indicates error detection (when the current frame is a bad frame), the compressed audio decoding unit 23 outputs the compressed audio encoded data of the current frame. Since it cannot be used, the reproduced audio signal is generated by interpolation using the built-in interpolation processing unit 23a. Unlike the conventional method, the interpolation processing unit 23a does not only include the information of the compressed speech coded data of the frame immediately before the current frame (hereinafter, referred to as the previous frame) but also the frame immediately after the current frame (hereinafter, referred to as the next frame). An appropriate interpolation process is performed by using the compressed voice coded data (referred to as “call”), and decoding is performed to generate a reproduced voice signal of the current frame.

FIG. 2 shows a conceptual time relationship of output signals in each part of the above system.

The input audio signal sequence shown in FIG. 2 (a) is divided into frames A, B and C as a time interval sequence, and compression coding processing is performed in time corresponding to each frame. As shown in (b), compressed voice coded data Ca, Cb, and Cc are formed in the next frame section. Then, as shown in FIG. 2 (c), this is inserted into a predetermined position determined by the corresponding channel of the TDMA frame, and exchanged between the transmitting side and the receiving side. The compressed voice coded data shown in FIG. 2 (d1) separated from the TDMA frame is subjected to compression decoding processing after being delayed by one frame, that is, the compressed voice coded data of the next frame can also be used. The compression decoding processing is executed in this state, and a sequence of reproduced audio signals as shown in FIG. 2 (e1) is obtained.

Here, when the frame B becomes a bad frame and the compressed voice coded data Cb cannot be used due to a communication path error or the like, as shown in FIG. 2D2, the preceding and following frames A and C are used. The encoded data Cbd corresponding to the frame B using the encoded data Ca and Cc of
Of the reproduced signal sequence B corresponding to the frame B
produces d.

As described above, the receiving side has a structure for delaying the compressed voice coded data, and when the current frame to be decoded is a bad frame, the compressed voice coded data of the frames before and after the current frame are decoded. Since the reproduction audio signal is obtained by using the interpolation, the information of the bad frame can be reproduced well, and the quality of the reproduction audio signal can be improved more than before.

Incidentally, it is possible to use only the information of the previous frame for the interpolation of the bad frame, but when the change of the information between the previous frame and the current frame (bad frame) is large, the interpolation is not successful and the quality is deteriorated. Since deterioration cannot be avoided, it is preferable to interpolate the information before and after the bad frame as described above.

A specific method of interpolating using the information of the frames before and after the bad frame will be described in the order of the method of interpolating vocal tract information and the method of interpolating excitation source information.

(B) Method of interpolating vocal tract information FIG. 3 shows a compressed speech decoding apparatus (applying a method of interpolating vocal tract information using information of frames preceding and following a bad frame (decoding process target frame: current frame)). 2 shows a specific configuration example of the compressed voice decoding unit). In addition, in FIG.
Although the processing configuration of the excitation source information is also shown, it is a drawing intended to explain in detail a method of interpolating vocal tract information.

The compressed speech decoding apparatus shown in FIG. 3 has a structure in which the buffer 22 and the compressed speech decoding unit 23 in the conceptual structure shown in FIG. 1 are integrated.

The compressed speech decoding apparatus is intended for compressed speech coded data encoded by the following encoding apparatus (not shown). That is, it is premised on an encoding device which is based on a code-excited linear predictive encoding system and which encodes line spectrum pair parameters (LSP parameters) for vocal tract information by three-stage vector quantization and inserts them into compressed speech encoded data. And
Note that encoding the line spectrum pair parameter and inserting it into the compressed voice encoded data is described in, for example, Japanese Patent Application No. 3-327443, and the compression encoding device on which this premise is known. belongs to.

In FIG. 3, the demultiplexing circuit 50 is provided with compressed speech coded data and error detection information, and the demultiplexing circuit 50 demultiplexes the compressed speech coded data and separates them from the first stage to the third stage. Vector quantization code VQ1n ~
VQ3n is applied to the corresponding codebooks for multistage vector decoding (inverse quantization) (hereinafter referred to as VQ codebooks) 51 to 53.

In FIG. 3, the suffix "p" for each code means the value in the previous frame for the current frame in which the reproduced audio signal is output, and the suffix "n" for each code is as follows. It means a value in a frame, and a value without suffix "p" or "n" means a value in the current frame.

The respective VQ codebooks 51, 52 and 53 are respectively given codes VQ1n, VQ2n and VQ3n.
Line spectrum vs. parameter component LSP1 determined by
n, LSP2n and LSP3n are output. The summation (vector combination) of these line spectrum pair parameter components LSP1n, LSP2n and LSP3n is obtained by the adders 54 and 55, and given to the buffer 56 as the line spectrum pair parameter LSPn of the next frame. The buffer 56 functions almost as a one-frame delay circuit, and outputs the line spectrum pair parameter LSP for the current frame to the switch 57.

The demultiplexing circuit 50, in accordance with the error detection information provided from an error correction decoding unit (communication path decoding unit) (not shown), if the current frame is not a bad frame (normal processing), the buffer 56 side. The selection control signal SW to be connected to the switch 57 is given to the switch 57. When the current frame is a bad frame, the demultiplexing circuit 50 connects the selection control signal SW to the interpolation circuit 64 side described later.
To the switch 57.

During normal processing when the switch 57 is connected to the buffer 56, the line spectrum pair parameter LSP for the current frame from the buffer 56 is supplied to the LPC coefficient converter 58. The LPC coefficient converter 58 converts the line spectrum pair parameter LSP into an LPC coefficient (linear prediction coefficient) α and gives it to the synthesis filter 59 as a synthesis filter coefficient.

The demultiplexing circuit 50 extracts the excitation source code obtained by demultiplexing the compressed speech coded data from the excitation source decoding unit 60.
Give to. The excitation source decoding unit 60 includes, for example, an adaptive excitation codebook, a statistical excitation codebook, and a configuration in which excitation signals (excitation code vectors) from these excitation codebooks are weighted and added by gains. If not, the excitation signal e obtained from the excitation source code of the current frame is given to the synthesis filter 59.

The synthesizing filter 59 synthesizes the given excitation signal e with the given LPC coefficient α to obtain a reproduced voice signal for the current frame.

The above is the basic decoding structure, but in addition to this, a structure is provided for interpolating when the current frame is a bad frame. The switch 57 described above also forms part of the interpolation configuration.

The first-stage line spectrum pair parameter component LSP1n for the next frame, which is output from the first-stage VQ codebook 51, is supplied to the cascade connection section composed of the three buffers 61 to 63. Therefore, buffer 6
1 stores the first stage line spectrum pair parameter component LSP1n for the next frame, buffer 62 stores the first stage line spectrum pair parameter component LSP1 for the current frame, and buffer 63 stores the first stage line spectrum pair parameter component LSP1n for the current frame. The line spectrum pair parameter component LSP1p of the stage will be stored.

The first stage line spectrum pair parameter component LSP1n for the next frame by the buffer 61,
The first-stage line spectrum pair parameter component LSP1p for the previous frame by the buffer 63 is the interpolation circuit 6
Given to 4. The interpolation circuit 64 uses the line spectrum pair parameter components LSP1n and LSP1p of the first stage of the preceding and following frames for these current frames to determine the line spectrum pair parameter L for the current frame (bad frame).
SPd (d means an interpolated value; the same applies hereinafter) is given to the switch 57. The interpolation circuit 64 performs interpolation by, for example, a simple average or a weighted average.

If the current frame is a bad frame,
The switch 57 is connected to the interpolation circuit 64 side based on the selection control signal SW, and the interpolated line spectrum pair parameter LSPd is given to the LPC coefficient converter 58 and converted into an LPC coefficient (synthesis filter coefficient) αd. It is then given to the synthesis filter 59. When the current frame is a bad frame, the excitation source decoding unit 60 also supplies the excitation signal ed formed by the interpolation process to the synthesis filter 59. An example of a specific method of forming the excitation signal by interpolation will be described later. Therefore, the synthesis filter 5
In the case where the current frame is a bad frame, 9 is a synthesizing process using the LPC coefficient αd formed by the interpolation process with respect to the excitation signal ed formed by the interpolation process,
Obtain the reproduced audio signal for the current frame.

As described above, when the current frame is a bad frame, when the vocal tract information is obtained by interpolating from the information of the frames before and after it, the interpolation can be smoothly performed to improve the quality of the reproduced audio signal. You can

Here, the line spectrum pair parameter is said to have the best interpolation characteristic among various vocal tract parameter expressions, and can be satisfactorily interpolated. Also,
In the case of multi-stage vector quantization, the first-stage information represents most of the original vector information, and as described above, the reproduced voice quality can be improved even if the interpolation information is obtained only by the first-stage information. In this case, the interpolation structure can be simplified. Of course, the information of the second and third stages may also be used for interpolation.

The method of interpolating the vocal tract information from the information of the frames before and after the current frame, which is a bad frame, is not limited to the above-mentioned method, and a method other than the line spectrum pair parameter may be applied. The quantization may also be one to which another quantization method such as matrix quantization is applied.

(C) Method of Interpolating Excitation Source Information FIG. 4 is a compressed speech decoding apparatus to which the method of interpolating excitation source information using information of frames preceding and following a bad frame (decoding process target frame: current frame) is applied ( 2 shows a specific configuration example of the compressed voice decoding unit). Note that the one shown in FIG. 4 simply uses the information of the previous frame for vocal tract information.

The compressed speech decoding apparatus shown in FIG. 4 has the following coding apparatus (code excitation linear predictive coding apparatus:
(Not shown). That is, having both an adaptive excitation codebook and a statistical excitation codebook,
The synthesis filter coefficient (vocal tract parameter) α, the index L for the optimal adaptive excitation code vector, the gain for the optimal adaptive excitation code vector (hereinafter referred to as the adaptive excitation code gain) β, the optimal statistical excitation code obtained for each frame The present invention is compatible with a code-excited linear predictive coding apparatus that quantizes and multiplexes an index H for a vector and a gain for an optimal statistical excitation code vector (hereinafter referred to as statistical excitation code gain) γ to form compressed speech coded data. It is a thing.

Therefore, the compressed speech decoding apparatus (code excitation linear predictive decoding apparatus) includes a demultiplexing circuit 100, an adaptive excitation codebook 101, and a statistical excitation codebook 10.
2, adaptive gain multiplier 103, statistical gain multiplier 1
04, the adder 105, and the synthesis filter 106 have a basic configuration.

The demultiplexing circuit 100 shown in FIG.
It is shown as a block having a wider function than the demultiplexing circuit 50 shown in FIG. That is, the demultiplexing circuit 100
Is a block having a decoding function for the synthesis filter coefficient and the like, and a function (buffer etc.) capable of outputting not only the current frame but also various information of the previous frame and the next frame.

The basic structure is the adaptive excitation codebook (10
Since it is premised on the coding method including 1), it is difficult to apply the information of the next frame to the interpolation, but it is used to improve the following three points.

(1) It is used for interpolation so that the periodicity of the excitation code vector can be reproduced well even if a bad frame occurs at the rising portion of the voice.

(2) When the current frame is a bad frame, the quality of the reproduced voice is improved by the addition ratio indicating the ratio of combining the adaptive excitation code vector and the statistical excitation code vector. Used to interpolate.

(3) When the current frame is a bad frame, the power is used to interpolate the quality of reproduced voice to be good.

In FIG. 4, the switch 107, the pulse multiplier 108, the pulse generation circuit 109, the lag calculation circuit 110, and the rise detection circuit 111 among the units provided outside the basic structure are for the first improvement. The addition ratio calculation circuit 120 and the switches 121 and 122 are for the second improvement, and the power interpolator 130 and the power multiplier 131 are for the third improvement. The function of each element will be clarified through the description of the operation.

First, the normal operation when the current frame is not a bad frame will be described.

In normal times, the demultiplexing circuit 100 separates the compressed speech coded data of the current frame to obtain the adaptive excitation codebook index L, the statistical excitation codebook index H, the adaptive excitation code gain β, and the statistical excitation code. The gain γ and the synthesis filter coefficient α are output to each unit.

As a result, the adaptive excitation codebook 101
Outputs an adaptive excitation code vector ea determined by the index L, and this adaptive excitation code vector e
After a is multiplied by the adaptive excitation code gain β by the multiplier 103, the adder 105 is added via the switch 107 described later.
And the statistical excitation code vector es determined by the index H is output from the statistical excitation code book 102, and the statistical excitation code vector es is multiplied by the statistical excitation code gain γ by the multiplier 104 and then added to the adder 105. Given.

During the normal operation, both the switches 121 and 122 are designed to select the gains β and γ obtained by the demultiplexing circuit 100 separating the compressed voice coded data of the current frame. Further, the switch 107 is adapted to select the adaptive multiplier 103 side.

The adder 105 synthesizes the excitation code vectors of these respective components to obtain the optimum excitation code vector (synthetic excitation code vector) e for the current frame,
It is given to the synthesis filter 106 via the multiplier 131 to which 1 is given as the multiplication coefficient PW.
Synthesizes the synthesized excitation code vector e with a synthesis filter coefficient α to obtain a reproduced voice signal S.

The composite excitation code vector e for the current frame via the multiplier 131 is fed back to the adaptive excitation codebook 101, and the adaptive excitation codebook 1
01 adaptively changes the stored excitation code vector accordingly. The excitation code vector e is also given to the lag calculation circuit 110.

Unless the current frame is a bad frame,
Such normal decoding process is repeated.

Next, the operation when the current frame is a bad frame will be described.

At this time, the demultiplexing circuit 100 gives the following information to each unit. The adaptive excitation codebook 101 contains the interpolation index Ld of the current frame.
As the index Lp of the previous frame. The statistical excitation codebook 102 is given the index Hp of the previous frame as the interpolation index Hd of the current frame. The synthesis filter 106 uses the synthesis filter coefficient α of the previous frame as the interpolation synthesis filter coefficient αd of the current frame.
give p. The rising edge detection circuit 111 is supplied with the adaptive excitation code gain βp and the statistical excitation code gain γp in the previous frame, and the adaptive excitation code gain βn and the statistical excitation code gain γn in the next frame. The addition ratio calculation circuit 120 indicates to the adaptive excitation code gain βn of the next frame.
And the statistical excitation code gain γ n are given. Switch 12
A selection control signal for selecting the interpolation excitation code gains βd and γd formed by the addition ratio calculation circuit 120 is given to 0 and 121. Selection control for the switch 107 to select the pulse multiplier 108 side when the rising edge detection circuit 111 outputs a significant detection signal and to select the adaptive multiplier 103 side when an insignificant detection signal is output. Give a signal. The power interpolator 130 is supplied with the adaptive excitation code gain βn and the statistical excitation code gain γn of the next frame.

The rising edge detection circuit 111 uses the adaptive excitation code gain βp and the statistical excitation code gain γp of the previous frame, and the adaptive excitation gain βn and the statistical excitation gain γn of the next frame to detect the voice of the current frame (bad frame). It is detected whether or not it is the rising part of.
The rising edge detection circuit 111 has, for example, equations (1) and (2)
When both of the conditional expressions are satisfied, the current frame (bad frame) is determined to be the rising portion of the voice by judging from the information of the frames before and after it.

√ (βn × βn + γn × γn)> √ (βp × βp + γp × γp) (1) | βn / γn |> | βp / γp | (2) The condition shown in the equation (1) is This is a condition for determining whether or not the power of the next frame is larger than that of the previous frame. The condition shown in equation (2) is a condition for determining whether or not the contribution rate of the adaptive excitation code vector in the combined excitation code vector is larger in the next frame than in the previous frame.

The power naturally increases at the rising portion of the voice. In addition, the adaptive excitation code vector contributes significantly to the pitch periodicity compared to the statistical excitation code vector, and the pitch periodicity increases at the rising part of the speech, so the increase in the contribution rate of the adaptive excitation code vector also increases the speech rising. It is a judgment element of the part.

As described above, the switch 107 selects the pulse multiplier 108 side only when the rising edge detection circuit 111 detects a rising edge, and when the current frame is not a bad frame and when the current frame is a bad frame. However, when the current frame is not the rising portion, the adaptive multiplier 103 side is selected.

The lag calculation circuit 110 stores the past composite excitation code vector for a predetermined period enough to calculate the lag, and when the rising edge detection circuit 111 detects a rising edge, the lag (pitch period) and the current The lag information P, which is the phase of the first pulse position in the frame, is calculated and given to the pulse generation circuit 109.

The pulse generation circuit 110 has periodic pulses according to the given lag information P with the lag included therein as intervals, and the first pulse phase is exactly as given information. A pulse excitation code vector em is generated and given to the pulse multiplier 108.

The pulse multiplier 108 outputs the pulse excitation code vector em output from the pulse generation circuit 110.
Is multiplied by the interpolation adaptive excitation code gain βd given from the addition ratio calculation circuit 120 and input to the switch 107.

In the addition ratio calculation circuit 120, the adaptive excitation code vector eagp output from the adaptive multiplier 103 for the previous frame after the gain multiplication and the previous frame after the gain multiplication output from the statistical multiplier 104 are output. Is given, and the adaptive excitation code gain βn and the statistical excitation code gain γn of the next frame are given as described above.

The addition ratio calculation circuit 120 first obtains the addition ratio pp for the previous frame and the addition ratio pn for the next frame according to the equations (3) and (4) using these pieces of information.

Pp = √ (eagp × eagp) / √ (esgp × esgp) (3) pn = | βn / γn | (4) where multiplication of the same excitation code vectors is the square of vector element It represents the sum (inner product).

For the previous frame, the adaptive excitation code vector eagp after gain multiplication applied to the adder 105.
And the statistical excitation code vector esgp after gain multiplication are known, the addition ratio pp may be obtained using these information. On the other hand, for the next frame, the addition ratio must be obtained before the adaptive excitation codebook 101 is updated for processing the next frame.
It is not possible to obtain the addition ratio pn using the information given in 05. However, no matter what the adaptive excitation code vector and the statistical excitation code vector are, the adaptive excitation code gain βn and the statistical excitation code gain γn
The ratio of stipulates the addition ratio pn,
Therefore, the addition ratio pn of the next frame is estimated by the equation (4).

Next, the addition ratio calculation circuit 120 estimates the addition ratio p of the current frame according to the equation (5) using the addition ratio pp of the previous frame and the addition ratio pn of the next frame.

P = θ × pp + (1−θ) × pn (5) Here, θ is 0 <θ <1, and is determined experimentally, for example.

Finally, the addition ratio calculation circuit 120
While using the addition ratio p for the current frame,
For example, according to the equations (6) and (7), the adaptive excitation code gain βd and the statistical excitation code gain γd for the current frame (bad frame) are determined, and the corresponding adaptive multiplication is performed via the corresponding switches 121 and 122. Bowl 1
03, to the statistical multiplier 104. The adaptive excitation code gain βd is also given to the pulse multiplier 108.

Βd = p × γn (6) γd = γn (7) Note that each gain βd and γd is determined based on, for example, the addition ratio p estimated around the adaptive excitation code gain βn of the next frame. Alternatively, it may be determined based on the addition ratio p estimated based on the average value of the statistical excitation code gains γp and γn of the previous frame and the next frame.

The power interpolator 130 generates a multiplication coefficient PW for power interpolation, and the power interpolation multiplier 131 uses the composite excitation code vector e or edd output from the adder 105 for power interpolation. The multiplication coefficient PW is multiplied and given to the synthesis filter 106. Power interpolator 13
When the current frame is not a bad frame, 0 outputs the multiplication coefficient PW (= 1) that allows the multiplier 131 to pass the combined excitation code vector e output from the adder 105 as described above, and outputs the current frame. Is a bad frame, the multiplication coefficient PW is formed and output as follows.

The power interpolator 130 has a structure for storing the combined excitation code vector ep of the previous frame,
From the combined excitation code vector ep of the previous frame, and the adaptive excitation code gain βn and the statistical excitation code gain γn of the next frame provided from the demultiplexing circuit 100, (8)
The power information pw shown in the formula is obtained.

Pw = λ × ε × √ (βn × βn + γn × γn) + (1-λ) × √ (ep × ep) (8) where 0 <ε and 0 <λ <1 These values are determined by experiment, for example. The power information of the previous frame is obtained by utilizing the synthetic excitation code vector ep already obtained. On the other hand, since the combined excitation code vector of the next frame cannot be used, the power information is estimated using the adaptive excitation code gain βn and the statistical excitation code gain γn, which are information reflected in the magnitude of power. In this way, since the power information is estimated for the next frame by a different operation from that for the previous frame, the correction coefficient ε for the case where it is obtained from the combined excitation code vector is multiplied as in the previous frame. Also, the weighting factor λ is applied in consideration of how the power information of the previous frame and the power information of the next frame contribute to the power information pw for the current frame.

When various interpolation values for the current frame (bad frame) are applied, the composite excitation code vector edd output from the adder 105 is given to the power interpolator 130.

The power interpolator 130 outputs the combined excitation code vector edd and the power information p for the current frame.
Based on w, the multiplication coefficient PW is determined according to the equation (9).

PW = pw / √ (edd × edd) (9) Here, since the power information pw is absolute size information, the power information √ included in the composite excitation code vector edd
It is converted into the rate of change of the power information (multiplication coefficient PW) by normalizing using (edd × edd). The power information of the previous frame can be used as the denominator for normalization.

Since each unit added to the basic configuration operates as described above, when the current frame is a bad frame, the reproduced voice Sd is obtained by the following series of processes.

When the rising edge detection circuit 111 determines that the current frame, which is a bad frame, is not the rising edge portion of the voice, the adaptive excitation codebook 101 determined by the index Ld of the previous frame.
The adaptive excitation code vector ead output from is multiplied by the adaptive excitation code gain βd output from the addition ratio calculation circuit 120 by the multiplier 103, and then applied to the adder 105 via the switch 107. Statistical excitation code vector esd output from statistical excitation codebook 102, which is determined by index Hd
Is multiplied by the statistical excitation code gain γd output from the addition ratio calculation circuit 120 by the multiplier 104 and then given to the adder 105. The adder 105 synthesizes these excitation code vectors to obtain an interpolated synthetic excitation code vector edd, and supplies it to the power interpolator 130 and the power interpolation multiplier 131. The power interpolator 130 also uses this information to generate the multiplication coefficient PW for power interpolation, and the multiplier 131 also performs the power interpolation by multiplying the interpolated combined excitation code vector edd by the multiplication coefficient PW. The interpolation synthesis code vector ed obtained by
Give to 6. The synthesizing filter 106 synthesizes the synthesized excitation code vector ed using the synthesizing filter coefficient αd of the previous frame to obtain a reproduced voice signal Sd.

On the other hand, the rising edge detection circuit 111
When it is determined that the current frame, which is a bad frame, is the rising portion of the speech, the lag calculation circuit 110 estimates the lag information P for the current frame from the past combined excitation code vector, and thereby the pulse The pulse excitation code vector em is output from the generation circuit 109. This pulse excitation code vector e
m is multiplied by the adaptive excitation code gain βd output from the addition ratio calculation circuit 120 by the multiplier 108, and then applied to the adder 105 via the switch 107. Also, the statistical excitation code vector esd output from the statistical excitation codebook 102, which is determined by the index Hd of the previous frame, is multiplied by the statistical excitation code gain γd output from the addition ratio calculation circuit 120 by the multiplier 104, and then post-added. Given to the container 105. The adder 105 synthesizes the excitation code vectors of these components to obtain an interpolated synthetic excitation code vector edd. The operation after the adder 105 is the same as the operation when the bad frame (current frame) is not the rising portion of the voice, and thus the description thereof is omitted.

As described above, since the excitation source information is interpolated using not only the information of the previous frame but also the information of the next frame, the quality of the reproduced audio signal can be improved. According to the interpolation configuration shown in FIG. 4, specifically, the reproducibility at the rising portion of the voice is improved, the pitch periodicity can be reproduced well, and how the adaptive excitation code vector and the statistical excitation code vector are calculated. It is possible to smoothly reproduce a change in addition ratio, which indicates whether to synthesize at a certain ratio, and also a smooth change in power, thereby improving the quality of a reproduced audio signal.

Incidentally, FIG. 4 merely shows one example of the interpolation method of the excitation source information using the information of the next frame.

(D) Other Embodiments In the above embodiment, when the current frame is a bad frame, the information of the preceding and following frames is used. However, the information of the past past the previous frame and the information of the next frame are used. Future information may also be used for interpolation.

Although the present invention has been made in consideration of the case of being applied to digital mobile communication, the applicable system is not limited to this, and a recording / reproducing system may be used. Therefore, the compression speech coding system is not limited to the one based on the code excitation linear predictive coding system.

Furthermore, the length of the processing frame may be different between the vocal tract information and the excitation source information. For example, the present invention can be applied even when the length of the vocal tract information processing frame divided into several pieces is the excitation source information processing frame.

[0097]

As described above, according to the present invention, in the compressed voice decoding apparatus for decoding the compressed voice coded data whose processing unit is each voice frame, the decoding target frame is the received information. When a bad frame is missing, the interpolation processing means for interpolating the information of the decoding target frame by using the information of the past frame and the future frame from the decoding target frame is provided. The interpolation can be favorably performed, and the quality of the reproduced audio signal can be improved as compared with the conventional case.

[Brief description of drawings]

FIG. 1 is a block diagram showing a digital mobile communication system to which an embodiment is applied.

FIG. 2 is an explanatory diagram showing a temporal relationship of frames in each processing unit of the system shown in FIG.

FIG. 3 is a block diagram showing a specific interpolation structure of vocal tract information according to an embodiment.

FIG. 4 is a block diagram showing a specific interpolation configuration of excitation source information according to the embodiment.

[Explanation of symbols]

21 ... Error correction decoding unit, 22, 56, 61-63 ... Buffer, 23 ... Compressed speech decoding unit, 23a ... Interpolation processing unit, 50 ... Demultiplexing circuit, 51-53 ... Vector decoding (dequantization) VQ codebook for, 57 ... switch,
64 ... Interpolation circuit.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yoshihiro Ariyama 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd.

Claims (2)

[Claims] 1. A compressed audio decoding apparatus for decoding compressed audio encoded data formed by processing for each audio frame, wherein a decoding processing target frame is a bad frame in which reception information is missing. In addition, the compressed speech decoding apparatus is characterized by further comprising interpolation processing means for interpolating the information of the decoding processing target frame by using the information of the past frame and the future frame from the decoding processing target frame. 2. Compressed speech coded data formed by processing for each speech frame, wherein vocal tract information includes a code formed by multistage vector coding of line spectrum pair parameters. In a compressed speech decoding apparatus that decodes a frame, the multi-stage vector-decoded line spectrum pair parameter of at least the first stage holds information of frames before and after the frame to be decoded, and the frame to be decoded is Interpolation processing means for interpolating the line spectrum pair parameter of the decoding process target frame from the line spectrum pair parameter held for the frames before and after the decoding process target frame when the received information is missing A compressed speech decoding apparatus comprising: