WO2000008633A1

WO2000008633A1 - Exciting signal generator, voice coder, and voice decoder

Info

Publication number: WO2000008633A1
Application number: PCT/JP1999/004137
Authority: WO
Inventors: Hiroyuki Ehara; Toshiyuki Morii
Original assignee: Matsushita Electric Industrial Co., Ltd.
Priority date: 1998-08-06
Filing date: 1999-08-02
Publication date: 2000-02-17
Also published as: JP2000056799A; AU4932499A

Abstract

An MA adaptive code vector is generated by using a finite number of noise code vectors used in the past, an adaptive code book gain, and a pitch period, and the amount of phase shift is calculated from the MA adaptive code vector, thereby shifting the phase of the noise code vector by the calculated amount of phase shift.

Description

Description Excitation signal generator, speech encoder and speech decoder

The present invention relates to a CELP (Code Excited Linear Prediction) type speech coding apparatus in a mobile communication system or the like that encodes and transmits a speech signal. Background art

In the field of digital mobile communications and voice storage, voice information is compressed for efficient use of radio waves and storage media, and is used for voice coding equipment for efficient coding. In particular, a method based on the CELP (Code Excited Linear Prediction) method has been widely put to practical use in middle and low bit rates.

Regarding CELP technology, Mr. Schroeder and BSAtal gave a "Code-Excited. Linear Prediction (CELP): High-quality Speech at Very Low Bit Rates", Proc ICASSP-85, 25.1.1, pp.937-940. , 1985 ".

In the CEL P-type speech coding method, speech is divided into a certain fixed frame length (about 5 ms to 50 ms), linear prediction of speech is performed for each frame, and a prediction residual (excitation signal) is obtained by linear prediction for each frame. Is encoded using an adaptive code vector composed of known waveforms and a noise code vector.

The adaptive code vector is based on the adaptive codebook that stores the driving excitation vector generated in the past, and the noise code vector is the noise that stores a predetermined number of vectors with a specified shape prepared in advance. It is selected from the codebook and used. The random code vector stored in the random codebook includes a random noise sequence vector. For example, a vector generated by arranging some pulses or different pulses at different positions is used. In particular, when the bit length is reduced by increasing the frame length, the quality is improved by using the noise code vector in synchronization with the pitch peak position of the adaptive code vector (phase "Adaptive CELP method" is disclosed in Japanese Unexamined Patent Application Publication No. 7-92999 "Speech excitation signal encoding method and apparatus" and "Study on pitch position synchronization CELP sound source encoding method" by Tazaki et al. 28, pp. 285-286 ", etc. These methods of synchronizing the noise code vector to the pitch peak position of the adaptive code vector are used in the vowel part of the speech signal. By utilizing the feature that prediction residuals tend to remain near the pitch peak, the prediction residual signal of the audio signal is efficiently represented.

Fig. 1 shows an example of the configuration of the excitation signal generator provided in the phase-adaptive CELP encoder. In the excitation signal generator shown in the figure, an excitation signal is generated by adding the adaptive code vector multiplied by the adaptive codebook gain and the noise code vector after phase adaptation processing multiplied by the noise codebook gain. . The phase adaptation process is performed using the phase calculated using the adaptive code vector.

However, in the conventional excitation signal generator described above, since the phase is calculated using the adaptive code vector output from the adaptive codebook, there is a problem that propagation of a transmission line error is likely to occur. . Disclosure of the invention

An object of the present invention is to perform a phase calculation using an MA-type adaptive code vector output from an MA-type adaptive codebook instead of an adaptive code vector output from an adaptive codebook, to thereby reduce a transmission path error. An object of the present invention is to provide an excitation signal generating device, a voice coding device, and a voice decoding device capable of suppressing propagation.

In order to achieve the above object, the present invention generates an MA type adaptive code vector using a finite number of noise code vectors used in the past, an adaptive codebook gain, and a pitch period. Then, the phase is calculated using this. BRIEF DESCRIPTION OF THE FIGURES

FIG. 1 is a block diagram showing a configuration of a conventional phase-adaptive excitation signal generator; FIG. 2 is a block diagram showing a configuration of an excitation signal generator according to the first embodiment of the present invention;

FIG. 3 is a block diagram showing a configuration of an MA-type adaptive codebook provided in the excitation signal generation device according to the first embodiment;

FIG. 4 is a flowchart showing the flow of the excitation signal generation process in the first embodiment;

FIG. 5 is a flowchart showing the flow of the MA-type adaptive codebook generation process according to the first embodiment;

FIG. 6 is a block diagram showing a configuration of a speech coding apparatus and a speech decoding apparatus according to Embodiment 2 of the present invention;

FIG. 7 is a block diagram illustrating a configuration of an audio signal transmitting device and a receiving device according to the third embodiment of the present invention. BEST MODE FOR CARRYING OUT THE INVENTION

Hereinafter, embodiments of the present invention will be specifically described with reference to FIGS. 1 to 6.

(Embodiment 1)

FIG. 2 shows a configuration of the excitation signal generator according to the first embodiment of the present invention. The excitation signal generator shown in the figure includes an adaptive codebook 101, a first random codebook 102, and a second random codebook 103.

The adaptive codebook 101 buffers the excitation signal generated in the past, and generates an adaptive code vector using the pitch period (pitch lag) P. The adaptive code vector generated by adaptive codebook 101 is applied to adaptive codebook gain G by multiplier 104. After being multiplied by 1, it is output to the adder 105.

The first noise code book 102 stores a predetermined number of noise code vectors having different shapes and stores the first noise code vector specified by the index S 1 of the noise code vector. Output. The first noise code vector is phase-shifted in the phase adaptor 106 by a shift amount described later. The first noise code vector after the phase shift is multiplied by the noise codebook gain G 2 in the multiplier 10 Ί and output to the adder 105.

The second random codebook 103 stores a predetermined number (finite number) of random code vectors having different shapes, and stores a second random code specified by an index S 2 of the random code vector. The vector is output to the multiplier 108. Multiplier 108 multiplies the second noise code vector by noise codebook gain G 2 and outputs the result to adder 105.

The second noise code vector multiplied by the noise codebook gain G 2 is simultaneously provided to the MA adaptive codebook 109. MA-type adaptive codebook 109 generates MA-type adaptive code vector using second noise code vector after noise codebook gain multiplication, adaptive codebook gain G1 and pitch period P. And outputs it to the phase calculator 110. The phase calculator 110 calculates the phase shift amount using the MA type adaptive code vector output from the MA type adaptive code book 109 and the pitch period P, and uses the shift amount as the phase adaptor Output to 6.

The excitation signal output from the adder 105 is also input to the adaptive codebook 101 and used to update the adaptive codebook.

The operation of the excitation signal generator configured as described above will be described.

First, the pitch period (pitch lag) P is stored in the adaptive codebook 101, the phase calculator 110, and the MA adaptive codebook 109, and the first noise codebook index S1 is stored in the first noise codebook. In 102, the second random codebook index S2 is in the second noise codebook 103, the adaptive codebook gain G1 is in the multiplier 104 and the MA adaptive codebook 109, Noise codebook gain G2 is input to multipliers 107 and 108, respectively. Is done.

The adaptive codebook 101 buffers the excitation signal generated in the past as time-series data.Starting at the point specified by the pitch period P, the adaptive codebook is cut out of the adaptive codebook to obtain a multiplier. Output to 104.

At this time, if the data length behind the position of the point specified by the pitch period P is shorter than the adaptive code vector length to be output, the adaptive code vector is generated by periodicizing the pitch period (output The adaptive code vector length is equal to the excitation signal vector length output from the excitation signal generator). The multiplier 104 multiplies the adaptive code vector output from the adaptive code book 101 by the adaptive code book gain G 1 to generate a vector of the adaptive code component of the excitation signal vector.

The first random codebook 102 extracts the first random code vector specified by the first random codebook index S1 and outputs it to the phase adaptor 106. In the phase adaptor 106, the phase shift of the first noise code vector is performed, so that the vector length stored in the first noise code book 102 is output from the excitation signal generator. It has a length longer than the excitation signal vector length by the maximum shift length that can be performed by the phase adaptor 106.

The phase adaptor 106 shifts the first noise code vector by the shift value calculated by the phase calculator 110, cuts out only the portion used for generating the excitation signal vector, and multiplies the multiplier by adding Output to 7. The multiplier 107 multiplies the vector output from the phase adaptor 106 by the noise codebook gain G2, and outputs the result to the adder 105.

The second noise codebook 103 takes out the second noise code vector specified by the index S2 and outputs it to the multiplier 108. The noise code vector stored in the second noise code book 103 is the same as the excitation signal vector length generated from the present excitation signal generator. The multiplier 108 multiplies the second noise code vector output from the second noise codebook 103 by the noise codebook gain G2.

The MA-type adaptive codebook 109 is the second after the noise codebook gain multiplication input in the past. Is generated using the noise code vector and the pitch period P inputted in the past, and further, the MA type adaptive code vector is cut out from the generated MA type adaptive codebook using the current pitch period P to calculate the phase. Output to 1 110

The phase calculator 110 searches for a phase position using the MA-type adaptive code vector and the current pitch period P. There are several methods for searching for the phase position.The method of maximizing the correlation value between the pulse train arranged in the pitch period P and the MA adaptive code vector, and the pitch period when using for CELP coding There is a method to maximize the correlation value between the vector obtained by applying a synthesis filter to the pulse train arranged in P and the vector obtained by applying the synthesis filter to the MA adaptive code vector.

Finally, the adder 105 adds the vectors output from the multiplier 104, the multiplier 107, and the multiplier 108 to generate an excitation signal vector. The generated excitation signal vector is also output to adaptive codebook 101 and used to update adaptive codebook 101.

FIG. 3 shows a detailed configuration of the MA-type adaptive codebook 109.

The MA-type adaptive codebook 109 forms the second noise code vector after multiplication by the noise codebook gain output from the multiplier 108 into a first delay unit 201 and a second delay unit 200. 2. The third delay unit 203 delays one unit time (one unit time is the time corresponding to the length of the excitation signal vector generated in one generation process).

The second noise code vector multiplied by the noise codebook gain three unit times before and output from the third delay device 203 is buffered in the first MA-type adaptive codebook 204. The first MA-type adaptive codebook 204 extracts a first MA-type adaptive code vector starting from a point indicated by a pitch period two unit times before, which will be described later.

The result obtained by multiplying the first MA-type adaptive code vector extracted from the first MA-type adaptive codebook 204 by the adaptive codebook gain two unit times later described later in a multiplier 205 Is output to the adder 206. In this adder 206, the output of the multiplier 205 and the output of the second delay unit 202 output the noise codebook gain two unit times ago. The second noise code vector after the multiplication is added. This added value is output to the second MA type adaptive codebook 207.

The second MA-type adaptive codebook 207 is composed of the first MA-type adaptive codebook and the vector output from the adder 206, and is described below one unit time before. The MA type adaptive code vector is cut out starting from the point indicated by the pitch period.

The vector cut out from the second MA-type adaptive codebook 207 is multiplied by an adaptive codebook gain one unit time before described later by a multiplier 208, and the result is output to an adder 209. . In the adder 209, the output of the multiplier 209 and the second noise code vector multiplied by the noise codebook gain one unit time ago output from the first delay unit 201 are added. I do. The added value is output to the third MA type adaptive codebook 210.

The third MA-type adaptive codebook 210 is formed by connecting the second MA-type adaptive codebook 207 and the vector output from the adder 209, and is specified by the pitch period P. Then, the MA-type adaptive code vector is cut out from the starting point and output to the phase calculator 110.

The applied codebook gain G 1 is sequentially delayed by one unit time by the fourth delay unit 2 1 1 and the fifth delay unit 2 1 2. The adaptive codebook gain two unit times before output from the fifth delay unit 2 1 2 is given to the multiplier 205 described above, and the adaptive codebook gain before one unit time output from the fourth delay unit 211 is output. The adaptive codebook gain is provided to the multiplier 208 described above.

The pitch period P is sequentially delayed by one unit time by the sixth delay unit 2 13 and the seventh delay unit 2 14. The pitch period two unit times before output from the seventh delay unit 2 14 is given to the first MA type adaptive codebook 204 described above, and output from the sixth delay unit 2 13 1 The pitch period before the unit time is given to the second MA type adaptive codebook 207 described above.

The operation of the MA-type adaptive codebook configured as described above will be described. The second noise code vector after the noise codebook gain multiplication input from the multiplier 108 (hereinafter simply referred to as the noise code vector in this paragraph) is input to the first delay unit, The delay unit 201 outputs the noise code vector S [—1] input one unit time ago (past). The second delay unit 202 receives the noise code vector S [—1] input one unit time ago, and further inputs the noise code vector input one unit time ago (ie, two unit time past). Outputs S [— 2]. The third delay unit 203 receives the noise code vector S [−2] input in the past two unit times, and further inputs the noise code vector input one unit time in the past (ie, three unit times in the past). Output the vector S [— 3]. This is equivalent to buffering all the noise code vectors input in the past three unit times and extracting the noise code vectors input in each unit time.

Similarly, the fourth delay unit 2 1 1 receives the adaptive codebook gain G 1, outputs the adaptive codebook gain G [—1] input one unit time ago, and outputs the fifth delay unit 2 1 2 is to input the adaptive codebook gain G [—1] input in the previous one unit time, and to further calculate the adaptive codebook gain G [—2] input one unit time in the past (ie, two unit time past). Output. This is equivalent to buffering all adaptive codebook gains input in the past two unit times and extracting adaptive codebook gains input in each unit time.

Similarly, the sixth delay unit 2 13 receives the pitch period P, outputs the pitch period P [—1] input in the past for one unit time, and the seventh delay unit 2 14 The pitch period P [-1] input in the past unit time is input, and the pitch period P [-2] input in the past one unit time (that is, two unit times past) is output. This is equivalent to buffering all pitch periods input in the past two unit times and extracting the pitch period input in each unit time.

In this way, the random code vector (S [—1], S [1 2], S [— 3]) and 2 to: The adaptive codebook gain (G [— 1], G [— 2]) before unit time and 2 to:! Using the pitch period (P [-2], P [1-1], P), an MA-type adaptive codebook is generated as follows.

First, the noise code vector in the past three unit times becomes the first MA-type adaptive codebook 204. The first MA type adaptive codebook 204 is a buffer having the maximum pitch period length that the pitch period can take, and the noise code vector past 3 unit time is copied at the end of the buffer. All parts before the copied part are 0.

The first MA-type adaptive codebook 2 0 4 uses the pitch period P [-2] of 2 unit time past to calculate the end point of the MA-type codebook (the end point of the random code vector of 3 unit time past) from P [ — Extracts and outputs the first MA-type adaptive code vector starting from the point that has been traced back by the pitch period length indicated by [2].

At this time, if the pitch period length is shorter than the vector length to be output (the length of one unit time), the periodic processing is performed with the pitch period length indicated by P [—2] to obtain a vector of a predetermined length. Output from Also, the first MA type adaptive codebook 204 itself is output to the second MA type adaptive codebook 207.

The vector output from the first MA-type adaptive codebook 204 is multiplied by the adaptive codebook gain G [—2] two unit times past in the multiplier 205 and output to the adder 206 Is done. The adder 206 adds the vector output from the multiplier 205 to the noise code vector S [−2] in the past two unit times and outputs the vector to the second MA-type adaptive codebook 207 I do.

The second MA-type adaptive codebook 2 07 is a buffer having the same length as that of the first MA-type adaptive codebook, and the vector output from the adder 210 is located at the end of this buffer. The first part is copied, and the first MA adaptive codebook 204 is copied in the previous part.

At this time, since copying to the second MA-type adaptive codebook is performed in order from the end of the first MA-type adaptive codebook, the first one unit time of the first MA-type adaptive codebook The minutes are not copied to the second MA adaptive codebook. The second MA-type adaptive codebook 207 goes back from the end of the second MA-type adaptive codebook 207 by a pitch period length represented by a pitch period P [—1] one unit time in the past. With the point as the starting point, the second MA-type adaptive code vector is cut out and output in the same manner as when the first MA-type adaptive code vector is cut out. Further, the second MA-type adaptive codebook 2107 itself is output to the third MA-type adaptive codebook 210.

The vector output from the second MA-type adaptive codebook 207 is multiplied by the adaptive codebook gain G [-1] of one unit time past in the multiplier 208 and output to the adder 209. Is done. The adder 209 adds the noise code vector S [—1] in the past one unit time and the vector output from the multiplier 209, and outputs the result to the third MA type adaptive codebook 210. .

The third MA-type adaptive codebook 210 is a buffer having the same length as the second MA-type adaptive codebook, and the vector output from the adder 209 is located at the end of this buffer. The second MA-type adaptive codebook 207 is copied to the part before it is copied.

At this time, since copying to the third MA-type adaptive codebook is performed in order from the end of the second MA-type adaptive codebook, the first unit time of the second MA-type adaptive codebook corresponds to the third unit time. It is not copied to the MA-type adaptive codebook. The third MA-type adaptive codebook 210 is a third MA-type adaptive codebook, starting from a point that has been advanced from the end of the third MA-type adaptive codebook 210 by the pitch period indicated by the current pitch period P. The type adaptive code vector is cut out and output in the same way as when the first MA type adaptive code vector was cut out. The third MA type adaptive code vector is input to the phase calculator 110.

As described above, in the present embodiment, the MA adaptive codebook generates four vectors in order to calculate the noise code vector and the adaptive codebook gain and the phase period based on the pitch period within the past three unit times. It is not affected by transmission line errors in the past more than an hour. In the above description, the noise code vector, adaptive codebook gain, and pitch period within the past three unit times are used.However, with the same configuration, information over the past four unit times or the past two unit times is used. The configuration used is also possible.

Next, a flow of processing of the excitation signal generating method according to the above embodiment will be described with reference to FIG.

In step 301, an adaptive code vector acv [0 to N-1] is generated from an adaptive codebook acb [-Pmax ~ -1]. Here, Pmax is the maximum value of the pitch period (pitched lag) that can be taken, N is the number of signal samples in one unit time, and [] indicates an array variable. The adaptive codebook acb [] is a buffer that stores only Pmax samples of the excitation signal vector generated in the past, and outputs acb [-P to N-P_1] as the adaptive code vector acv [0 to Nl]. . P is the pitch period. If N-ρ-1≥0, it is out of the range of the adaptive codebook acb [], so acv [] is generated by repeatedly using the part of acb [-P ~ -1].

Next, in step 302, the MA-type adaptive codebook ma-acv [0 to N-l] is generated from the MA-type adaptive codebook ma-acb [-Pmax: -1]. The method of generating ma-acb [] will be described later using FIG. MA-type adaptive codebook ma—acb [] is a buffer that stores only Pmax samples of vectors generated from sound source generation information in the past finite time, and MA—acb [-P to N-P-1] Type adaptive code vector ma—Output as acv [0 to Nl]. When N-P-1 ≥ 0, the MA-type adaptive code vector ma- acv [] is generated in the same way as when the adaptive code vector acv [] is generated from the adaptive codebook acb [].

Next, in step 303, the phase ph is calculated. The phase is calculated by searching for the position of the first impulse of the impulse train vector that maximizes the cross-correlation between the MA-type adaptive code vector ma—acv [] and the impulse train vectors arranged with the pitch period P Is used. When this excitation signal generation method is used for CELP coding, search for the position of the first impulse in the impulse train vector that minimizes distortion in the area after the excitation signal is subjected to the synthesis filter. "Phase adaptive PS I—CEL P speech coding", IEICE Technical Report, SP 94—96 (1 995-02) P. 37 -P. 44 ".

Next, in step 304, a first random code vector scvl [0 to N-1] is generated from the first random codebook SCBl [Slsize] [-MAXph to N-1]. The noise codebook SCB1 [] [] stores a vector with a length of N + MAXph as Slsize. Here, MAXph is the maximum value that the phase Ph can take (ph≥0). The vector SCB1 [S1] [] specified by the first noise code index S1 is extracted, and the SCB1 [S1] [-ph to Nl-ph] is cut out using the phase ph to obtain the first noise. Let the sign vector be scvl [0 to Nl].

Next, in step 305, a second random code vector scv2 [0 to N-l] is generated from the second random codebook SCB2 [S2size] [0 to N-l]. The random codebook SCB2 [] [] stores S2size types of vectors having a length of N. The vector SCB2 [S2] [0 to N-1] specified by the second noise code index S2 is set as a second noise code vector scv2 [0 to N-1].

Then, at step 306, adaptive code vector acv and base multiplied by the adaptive codebook gain G 1 in the [0~Nl] vector, the first noise code base vector _SCV 1 [0 to N-1] and the second noise code An excitation signal vector exc [0-Nl] is generated by adding a vector obtained by multiplying the sum vector with the vector scv2 [0-Nl] by the noise codebook gain G2.

Finally, in step 307, the adaptive codebook is updated. The adaptive codebook is updated by performing acb [n] = acb [n + N〗 for n = -Pmax -1 and shifting the buffer, then adding the newly generated excitation signal vector exc [0 to Nl]. This is done by copying to acb [-N ~ -1].

Next, a method of generating an MA-type adaptive codebook will be described with reference to FIG. FIG. 5 is a flowchart showing a specific process of the method of generating an MA-type adaptive codebook in the present embodiment. First, in step 401, the contents of the MA adaptive codebook ma_acb [-Pmax to -l] are cleared to zero. Next, in step 402, the second noise code vector after the noise codebook gain multiplication (hereinafter simply referred to as the noise code vector in this paragraph) is stored in the buffer buf—scb [0] [0 to Nl]. . Next, in step 403, the pitch period P is stored in the buffer buf_p [0]. Next, in step 404, the adaptive codebook gain G1 is stored in buf—g [0].

Here, each buffer will be described.

1) 1 ^ -301> [0-3] [0-11-1] stores the noise code vector generated in the past, and is stored in buf-scb [0] [0-N-1]. Is the currently generated noise code vector, and buf-scb [l] [0-Nl] is the noise code vector generated one unit time in the past, buf-scb [2] [0-N -1] stores the noise code vectors generated two unit times ago. Also, buf_p [0 to 2] stores the pitch cycle P used in the past, buf_p [0] indicates the current pitch cycle, and buf_p [1] indicates the pitch used in the past one unit time. The period, buf_p [2] stores the pitch period used in the past 2 units of time. Also, the adaptive codebook gain G 1 used in the past is stored in buf-g [0-2], the current adaptive codebook gain is stored in buf-g [0], and the current adaptive codebook gain is stored in buf-g [l]. Stores the adaptive codebook gain used in the past one unit time, and buf-g [2] stores the adaptive codebook gain used in the past two unit times.

Next, in step 405, an MA-type adaptive codebook ma-acb [] is generated. This is done by copying buf-scb [3] [0-N-1] to ma-1 acb [-N--1]. Next, in step 406, the MA type adaptive codebook is updated. First, ma-act> [-buf-p [2] -Nl-buf_p [2]] is copied to temporary vector ¹ Tmp [0-Nl]. This is the same as generating an adaptive code vector using the MA adaptive codebook generated in step 405 as an adaptive codebook and using buf_p [2] as a pitch period. Next, the MA adaptive codebook is shifted by performing ma acb [n] = ma—acb [n + N] for n = −Pmax N−1. Finally, The vector obtained by multiplying the temporary vector Tmp [0 to Nl] by buf—g [2] and buf—scb [2] [0 to N-1〗 is calculated by calo, and ma—acb [-N ~ -1 ].

Next, in step 407, the MA type adaptive codebook is updated again. First, ma acb [-buf_p [l] -N-1-buf-p [l]] is copied to the temporary vector Tmp [0-N-l]. This is the same as generating the adaptive code vector using the 適応 -type adaptive codebook updated in step 406 as the adaptive codebook and buf_p [1] as the pitch period. Next, ma-acb [n] = ma-act> [n + N] is performed on n = -Pmax N-1 to shift the MA adaptive codebook. Finally, the vector obtained by multiplying the temporary vector Tmp [0 ~ N-1] by buf-g [l] and buf_scb [1〗 [0 ~ N-1] are calculated as ma_acb [-N ~- 1].

At this point, the final MA adaptive codebook is created. Then, using the current pitch period buf_p [o] from the final MA-type adaptive codebook, the MA-type adaptive code vector ma—acv [0 to Nl] is calculated as ma—acv [0 to N—1] = ma—acb [-buf_p [0〗 -N-1-buf—p [0]]

Finally, in step 409, the three buffers are updated and all processing ends. By expanding the three buffers, storing the parameters used in the past—evening, and repeating the same processing as in steps 406 and 407, the MA-type adaptive codebook is expanded to increase the phase adaptability. It is also possible to increase.

In the present embodiment, two noise codebooks are provided, and the phase adaptation process is performed only on one of the noise codebooks. Therefore, compared with the case where the phase adaptation process is performed on both of the two noise codebooks. The resistance to transmission line errors has become stronger.

In the present embodiment, an example of a configuration in which a noise codebook that always performs (MA type) phase adaptation is used is shown. However, a noise codebook that uses only a general noise codebook that does not perform phase adaptation at all is shown. The present invention is also applicable to a configuration example in which a mode for generating a vector and a mode for generating a random code vector from a random code book performing phase adaptation as described in the present embodiment are switched.

(Embodiment 2) FIG. 6 shows an embodiment using the excitation signal generator shown in the first embodiment. FIG. 6A shows a speech encoding device, and FIG. 6B shows a speech decoding device.

In the speech coding apparatus shown in FIG. 6A, an input signal composed of a digitized speech signal or the like is inputted to the LPC analyzer 501 and the adder 506. The LPC analyzer 501 performs a linear prediction analysis to calculate a linear prediction coefficient (LPC) and outputs it to the LPC quantizer 502.

The LPC quantizer 502 quantizes the input LPC, outputs the quantized LPC to the synthesis filter 505, and outputs a code L representing the quantized LPC to the decoder.

The synthesis filter 505 constructs an LPC synthesis filter using the input quantized LPC. An excitation signal generated by the excitation signal generator 503 is input to the synthesized filter to perform filter processing, and a synthesized speech signal is output to the adder 506.

The adder 506 calculates an error between the input data and the synthesized speech signal, and outputs the error to the distortion calculator 507. The distortion calculator 507 calculates the distortion of the synthesized voice signal with respect to the input voice signal in consideration of the auditory weights, etc., based on the error signal output from the adder 506, and sends the resultant to the parameter determiner 504. Output.

The parameter determining unit 504 determines parameters (P, S 1, S 2, G 1, G 2) for generating an excitation signal output so as to minimize the distortion output from the distortion calculating unit 507. Adjust. Finally, a combination of parameters that minimizes distortion is output to the decoder side.

On the other hand, in the decoder of FIG. 6B, the encoded LPC information L transmitted from the encoder side is provided to the LPC decoder 508. LPC decoder 508 decodes and decodes the quantized LPC from PC information L and outputs it to synthesis filter 510. The synthesis filter 510 constructs an LPC synthesis filter using the decoding LPC input from the LPC decoder 508, applies a synthesis filter to the excitation signal input from the excitation signal generator 509, and decodes the synthesized speech. Output a signal. The excitation signal generator 509 generates the excitation signal using the parameters (P, SI, S2, Gl, G2) for generating the excitation signal transmitted from the encoder side. Output to the composite file.

Note that the synthesis filter 510 on the decoder side and the synthesis filter 505 on the encoder side are exactly the same if there is no error in the transmitted information. In addition, the excitation signal generator 503 on the decoder side and the excitation signal generator 509 on the encoder side perform exactly the same operation to generate the same excitation signal if there is no error in the transmitted information. . Further, when post-processing such as a post filter for improving the auditory quality is added to the decoded synthesized speech signal output from the synthesis filter 510, the quality of the decoded speech signal is further improved.

(Embodiment 3)

FIG. 7 is a block diagram showing an audio signal transmitter and a receiver provided with the audio encoding or decoding device according to the second embodiment. Figure 7A shows the transmitter and Figure 7B shows the receiver.

In the audio signal transmitter of FIG. 7A, the audio is converted into an electrical analog signal by the audio input device 601 and output to the AZD converter 602. The analog audio signal is converted into a digital audio signal by the AZD converter 602 and output to the audio encoder 603. Speech encoder 603 performs speech encoding processing, and outputs the encoded information to RF modulator 604. The RF modulator performs an operation for transmitting the information of the encoded voice signal as a radio wave such as modulation, amplification, and code spreading, and outputs the information to the transmission antenna 605. Finally, a radio wave (RF signal) 606 is transmitted from the transmitting antenna 605.

On the other hand, in the receiver in FIG. 7B, a radio wave (RF signal) 606 is received by the receiving antenna 607, and the received signal is sent to the RF demodulator 608. The RF demodulator 608 performs processing such as code despreading / demodulation for converting radio signals into encoded information, and outputs the encoded information to the audio decoder 609. The audio decoder 609 performs a decoding process on the encoded information and converts the digital decoded audio signal into a DZA converter 61. Output to 0. The DZA converter 610 converts the digitized decoded audio signal output from the audio decoder 609 into an analog decoded audio signal and outputs the analog decoded audio signal to the audio output device 611. Finally, the audio output device 6 11 1 converts the electrical analog decoded audio signal into decoded audio and outputs it.

The transmitting device and the receiving device can be used as a mobile device or a base station device of a mobile communication device such as a mobile phone. Note that the medium for transmitting information is not limited to radio waves as described in the present embodiment, but may use optical signals or the like, and may use a wired transmission path.

Note that the audio encoding device or the encoding and decoding device described in the second embodiment and the transmitting device and the transmitting and receiving device described in the third embodiment include a magnetic disk, a magneto-optical disk, and a ROM cartridge. It is also possible to realize by recording as software on a recording medium such as, for example, and by using the recording medium, a speech encoding device and a Z decoding device can be used by a personal computer or the like using such a recording medium. And a transmission device / reception device can be realized. That is, by installing the recorded program in a computer, the same function as that of the above-described excitation signal generator can be provided.

In the present invention, a phase adaptation process of a noise code vector is performed based on an MA type adaptive code vector generated using a finite number of noise code vectors used in the past, an adaptive codebook gain, and a pitch period. Therefore, the propagation of a transmission path error can be shortened as compared with a method in which a noise code vector is adaptively shifted based on information extracted from an adaptive code vector. Also, since the phase adaptation processing is performed only on one of the noise codebooks, the resistance to transmission path errors can be increased as compared with the case where the phase adaptation processing is performed on both of the two noise codebooks.

Further, according to the present invention, in the excitation signal generating device in the phase adaptive CELP encoding device, the noise code vector is not adaptively shifted by the information extracted from the adaptive code vector. Can minimize the effects of As described above, according to the present invention, since the phase adaptation processing is performed only from the data used in the past finite time, the propagation of errors due to the phase adaptation processing can be suppressed within a limited time. It is possible to provide an excitation signal generating device, a voice coding device, and a voice decoding device.

The present specification is based on Japanese Patent Application No. 10-232393, filed on Aug. 6, 1998. All this content is included here. Industrial applicability

INDUSTRIAL APPLICABILITY The present invention can be used in a communication terminal device such as a base station device and a mobile station in a digital radio communication system.

Claims

The scope of the claims

1. Excitation in which a noise codebook that generates a noise code vector used to generate an MA-type adaptive code vector and a noise codebook that generates a noise code vector to which a phase shift is given are configured by different codebooks. Signal generator.

2. The excitation signal generator according to claim 1, wherein the MA type adaptive code vector is generated using a finite number of noise code vectors, adaptive codebook gains, and pitch periods used in the past.

3. Means for calculating the phase shift amount of the noise code vector using the MA-type adaptive code vector, and means for generating the excitation signal vector using the noise code vector after the phase shift and the adaptive code vector The excitation signal generation device according to claim 1, comprising:

4. An adaptive codebook storing excitation signal vectors generated in the past, a first noise codebook storing a plurality of vectors having a predetermined shape, and a plurality of different codebooks different from the first noise codebook. And a finite number of noise code vectors and adaptive codebooks generated by the second noise codebook in the past, and the MA adaptive code vector using the gain and the pitch period. Means for generating a phase shift amount using the MA type adaptive code vector and the current pitch period, and a first noise generated from the vector of the first noise codebook. Phase adaptation means for giving a phase shift to the sound code vector by the phase shift amount, and means for generating an excitation signal vector using the first noise code vector after the phase shift and the adaptive code vector. Excitation signal Generating device.

5. The excitation signal generator according to claim 1, means for extracting a parameter representing a spectrum characteristic from an audio signal, means for quantizing and encoding the parameter, and the excitation signal generator. Means for synthesizing an audio signal from the generated excitation signal and the parameters, means for calculating distortion between the synthesized audio signal and the input audio signal, and means for calculating the distortion so that the calculated distortion is minimized. Means for determining the parameters of the excitation signal generation device.

6. The excitation signal generating apparatus according to claim 1, means for decoding a parameter representing a spectrum characteristic of the audio signal quantized in the audio encoding apparatus, and the means for decoding in the audio encoding apparatus. Means for generating a decoded speech from the excitation signal generated by the excitation signal generation apparatus based on the parameter and the decoded parameter.

8. An audio input device that converts an audio signal into an electrical signal, an AZD converter that converts a signal output from the audio input signal device into a digital signal, and an output signal from the A / D converter 5. An audio encoder according to claim 4, which encodes a digital signal to be encoded, an RF modulator for performing a modulation process on encoded information output from the audio encoder, and an output from the RF modulator. And a transmitting antenna for converting the converted signal into radio waves and transmitting the converted signal.

9. A receiving antenna for receiving a received radio wave, an RF demodulator for demodulating a signal received by the receiving antenna, and a voice decoding according to claim 6, for decoding information obtained by the RF demodulator. , A DZA converter for DZA converting a digital audio signal decoded by the audio decoding device, and an audio output device for converting an electrical signal output by the DZA converter into an audio signal. An audio signal receiving device comprising:

10. A mobile station device comprising at least one of the voice signal transmitting device according to claim 8 and the voice signal receiving device according to claim 9, and performing wireless communication with a base station device. 11. A base station device comprising at least one of the voice signal transmitting device according to claim 8 and the voice signal receiving device according to claim 9, and performing wireless communication with a mobile station device. 1 2.

A procedure for generating an MA-type adaptive code vector using a finite number of noise code vectors used in the past, an adaptive codebook gain, and a pitch period;

A procedure for calculating the phase shift amount of the noise code vector using the MA type adaptive code vector,

Excitation signal using noise code vector after phase shift and adaptive code vector. A step of generating a signal vector,

A machine-readable recording medium on which a program for executing the program is recorded.

13. A process of generating an MA-type adaptive code vector using a finite number of noise code vectors used in the past, an adaptive codebook gain, and a pitch period, and a process of generating noise using the MA-type adaptive code vector. An excitation signal generation method, comprising: calculating a phase shift amount of a code vector; and generating an excitation signal vector using the noise code vector after the phase shift and the adaptive code vector.

14. An excitation signal generated by the excitation signal generation method according to claim 13, wherein a step of extracting parameters representing spectral characteristics from the audio signal, and a step of quantizing and encoding the parameters are performed. Synthesizing the audio signal from the parameter and the parameter; calculating the distortion between the synthesized audio signal and the input audio signal; and controlling the excitation signal generating device so that the calculated distortion is minimized. A step of determining the parameters; and a speech encoding method comprising:

15. The excitation according to claim 13, wherein a step of decoding parameters representing the spectral characteristics of the voice signal quantized by the voice coding apparatus is performed, and the parameters determined by the voice coding apparatus are based on the parameters. Generating a decoded speech from the excitation signal generated by the signal generation method and the decoded parameters.