JP4236675B2 - Speech code conversion method and apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce deterioration in sound quality by making it possible to perform voice code conversion even between voice coding systems differing in subframe length and to perform code conversion corresponding to the rate of an EVRC by making a delay time as short as possible. <P>SOLUTION: Voice code conversion sections for a full rate, a half rate, and a 1/8 rate are provided corresponding to rates of a first voice code, and when a rate of the first voice code is the 1/8 rate, the voice code conversion part for the 1/8 rate (1) quantizes a reversely quantized value of an LSP code included in a voice code by a first voice encoding system by a second encoding system to obtain the LSP code of a second voice code, (2) generates a target signal and an algebraic composite signal and obtains an algebraic code by the second voice encoding system such that the difference between those target signal and algebraic composite signal becomes minimum, and (3) uses a reversely quantized value, an algebraic code, etc., of a pitch lag code obtained by voice code conversion for the full rate or half rate to obtain and output a gain code of the second voice encoding system. <P>COPYRIGHT: (C)2007,JPO&amp;INPIT

Description

  The present invention relates to a voice code conversion method and apparatus for converting a voice code obtained by encoding using a first voice coding method into a voice code of a second voice coding method, and in particular, the Internet, a mobile phone system, and the like. The present invention relates to a speech code conversion method and apparatus for converting a speech code obtained by encoding speech using the first speech encoding method used in the above to a speech code of a different second speech encoding method.

  In recent years, the number of mobile phone subscribers has increased explosively, and the number of users is expected to increase in the future. Voice communication using the Internet (Voice over IP; VoIP) is becoming widespread in fields such as corporate networks (intranet) and long distance telephone services. In a voice communication system such as a cellular phone or VoIP, a voice coding technique for compressing voice is used to effectively use a communication line. Different voice coding technologies are used in mobile phones depending on the country or system. In cdma2000, which is expected as a next-generation mobile phone system, EVRC (Enhanced Variable Rate CODEC) is used as a voice coding method. Method) is adopted. On the other hand, in VoIP, ITU-T recommendation G.729A is widely used as a voice encoding method. Below, the outline of G.729A and EVRC will be described first.

(1) Description of G.729A Configuration and operation of encoder FIG. 15 is a configuration diagram of an encoder of the ITU-T recommendation G.729A system. In FIG. 15, an input signal (audio signal) X having a predetermined number of samples (= N) per frame is input to the LPC analyzer 1 in units of frames. If the sampling rate is 8 kHz and one frame period is 10 msec, one frame is 80 samples. The LPC analysis unit 1 expresses the human vocal tract as
H (z) = 1 / [1 + Σαi · z ⁻ⁱ ] (i = 1 to P) (1)
The coefficient αi (i = 1,..., P) of this filter is obtained. Here, P is the filter order. Generally, in the case of telephone band voice, a value of 10 to 12 is used as P. The LPC (Linear Prediction) analysis unit 1 performs LPC analysis using a total of 240 samples of 80 samples of the input signal, 40 samples of the pre-reading, and 120 samples of the past signal to obtain LPC coefficients.

  The parameter converter 2 converts the LPC coefficient into an LSP (Line Spectrum Pair) parameter. Here, the LSP parameter is a parameter in the frequency domain that can be mutually converted with the LPC coefficient. Since the quantization characteristic is superior to the LPC coefficient, the quantization is performed in the LSP area. The LSP quantization unit 3 quantizes the converted LSP parameter to obtain an LSP code and an LSP inverse quantization value. The LSP interpolation unit 4 obtains an LSP interpolation value from the LSP inverse quantization value obtained in the current frame and the LSP inverse quantization value obtained in the previous frame. That is, one frame is divided into two first and second subframes of 5 msec, and the LPC analysis unit 1 determines the LPC coefficient of the second subframe, but does not determine the LPC coefficient of the first subframe. Therefore, the LSP interpolation unit 4 predicts the LSP inverse quantization value of the first subframe by interpolation using the LSP inverse quantization value obtained in the current frame and the LSP inverse quantization value obtained in the previous frame.

  The parameter inverse conversion unit 5 converts the LSP inverse quantization value and the LSP interpolation value into LPC coefficients, respectively, and sets them in the LPC synthesis filter 6. In this case, as the filter coefficient of the LPC synthesis filter 6, the LPC coefficient converted from the LSP interpolation value is used in the first subframe of the frame, and the LPC coefficient converted from the LSP dequantized value is used in the second subframe. . In the following description, 1 with a subscript, for example, 1 in lspi, li (n),.

The LSP parameter lspi (i = 1,..., P) is quantized by the LSP quantization unit 3 by scalar quantization or vector quantization, and then the quantization index (LSP code) is transmitted to the decoder side. The FIG. 16 is an explanatory diagram of the quantization method, and the quantization table 3a stores a large number of sets of quantization LSP parameters corresponding to the index numbers 1 to n. The distance calculation unit 3b has the following formula: d = Σi {lspq (i) -lspi} ² (i = 1 to P)
To calculate the distance. When q is changed from 1 to n, the minimum distance index detection unit 3c obtains q that minimizes the distance d, and transmits the index q as an LSP code to the decoder side.

  Next, sound source and gain search processing is performed. Sound source and gain are processed in subframe units. First, the sound source signal is divided into a pitch period component and a noise component, the pitch code component is quantized using the adaptive codebook 7 storing a past sound source signal sequence, and the noise component is quantized with an algebraic codebook. Or a noise codebook. In the following, a speech coding scheme that uses the adaptive codebook 7 and the algebraic codebook 8 as the excitation codebook will be described.

  The adaptive codebook 7 outputs a sound source signal (referred to as a periodic signal) for N samples sequentially delayed by one sample corresponding to the indexes 1 to L. FIG. 17 is a block diagram of the adaptive codebook 7 when one subframe has 40 samples (N = 40). The adaptive codebook 7 includes a buffer BF that stores the pitch period component of the latest (L + 39) samples. Identifies a periodic signal composed of 1 to 40 samples, identifies a periodic signal composed of 2 to 41 samples by index 2, ... identifies a periodic signal composed of L to L + 39 samples by index L The In the initial state, the contents of the adaptive codebook 7 contain all signals with an amplitude of 0, discard the oldest signal in time every subframe by the length of the subframe, and obtain the sound source signal obtained in the current subframe. It operates so as to be stored in the adaptive codebook 7.

  In the adaptive codebook search, the periodic component of the excitation signal is identified using the adaptive codebook 7 storing the past excitation signal. That is, while the starting point read from the adaptive codebook 7 is changed by one sample, the past excitation signal in the adaptive codebook 7 is extracted by subframe length (= 40 samples), and is input to the LPC synthesis filter 6 to be input to the pitch synthesis signal β. A / PL periodic signal (adaptive code vector), A is an impulse response of the LPC synthesis filter 6, and β is an adaptive codebook gain.

The arithmetic unit 9 calculates the input power X and the error power EL of β / A / PL as follows:
EL ＝ ｜ X−β ・ A ・ PL ｜ ² (2)
Ask for.
When the weighted composite output of the adaptive codebook output is A · PL, the autocorrelation of A · PL is Rpp, and the cross correlation between A · PL and the input signal X is Rxp, the error power in equation (2) is minimized. The adaptive code vector PL for the pitch lag Lopt is given by
^{PL = argmax (Rxp 2 / Rpp} ) (3)
Is represented by That is, the reading start point where the value obtained by normalizing the cross-correlation Rxp between the pitch composite signal A · PL and the input signal X by the autocorrelation Rpp of the pitch composite signal is the optimum start point. As described above, the error power evaluation unit 10 obtains the pitch lag Lopt that satisfies the expression (3). At this time, the optimum pitch gain βopt is expressed by the following equation: βopt = Rxp / Rpp (4)
Given in.

Next, the noise component contained in the sound source signal is quantized using the algebraic codebook 8. The algebraic codebook 8 is composed of a plurality of pulses having an amplitude of 1 or −1. As an example, FIG. 18 shows pulse positions when the subframe length is 40 samples. The algebraic codebook 8 divides N (= 40) sample points constituting one subframe into a plurality of pulse system groups 1 to 4, and for each combination obtained by taking one sample point from each pulse system group, A pulse signal having a +1 or -1 pulse at the sample point is sequentially output as a noise component. In this example, basically four pulses are arranged per subframe. FIG. 19 is an explanatory diagram of sample points assigned to each of the pulse system groups 1 to 4.
(1) Eight sample points 0, 5, 10, 15, 20, 25, 30, 35 are assigned to pulse system group 1,
(2) Eight sample points 1, 6, 11, 16, 21, 26, 31, 36 are assigned to pulse system group 2.
(3) Eight sample points 2, 7, 12, 17, 22, 27, 32, 37 are assigned to pulse system group 3,
(4) Pulse system group 4 has 16 sample points 3,4,8,9,13,14,18,19,23,
24,28,29,33,34,38,39 are assigned.

3 bits are required to represent the sample points of pulse system groups 1 to 3, 1 bit is required to represent the positive and negative of the pulse, and a total of 4 bits are required. Also, to represent the sample points of pulse system group 4 4 bits, 1 bit and 5 bits in total are required to express the positive and negative of the pulse. Accordingly, ¹⁷ bits are required to specify the pulse signal output from the noise codebook 8 having the pulse arrangement of FIG. 18, and the type of the pulse signal is 2 ¹⁷ (= 2 ⁴ × 2 ⁴ × 2 ⁴ × 2). ⁵ ) Exists.
As shown in FIG. 18, the pulse position of each pulse system is limited, and in the algebraic codebook search, the pulse with the smallest error power from the input speech in the reproduction area is selected from the combinations of pulse positions of each pulse system. Determine the combination. That is, the optimum pitch gain βopt obtained by the adaptive codebook search is set, and the adaptive codebook output PL is multiplied by the gain βopt and input to the adder 11. At the same time, a pulse signal is sequentially input to the adder 11 from the algebraic codebook 8 and the difference between the reproduced signal obtained by inputting the adder output to the LPC synthesis filter 6 and the input signal X is minimized. Identify the signal. Specifically, first, a target vector X ′ for algebraic codebook search is generated from the optimal adaptive codebook output PL obtained from the input signal X by adaptive codebook search and the optimal pitch gain _βopt by the following equation.

X ′ = X−β _opt APL (5)
In this example, the position and amplitude (positive / negative) of the pulse are expressed in 17 bits as described above, so there are 2 17 combinations. Here, when the kth algebraic code output vector is Ck, in the algebraic codebook search, D = | X′−GC · A · Ck | ² (6)
The code vector Ck that minimizes the evaluation function error power D is obtained. GC is the algebraic codebook gain. In the search of the algebraic codebook, the error power evaluation unit 10 normalizes the mutual value obtained by normalizing the square of the cross-correlation value Rcx between the algebraic composite signal A · Ck and the input signal X ′ with the autocorrelation value Rcc of the algebraic composite signal. A search is made for a combination of a pulse position and a polarity having the largest correlation value (Rcx * Rcx / Rcc). When the pitch lag is shorter than the subframe length, a pitch periodicizing unit is provided to improve the sound quality, and the pitch periodicizing process is performed by the pitch periodicizing unit to make the algebraic codebook output periodic. be able to. The output result of the algebraic codebook search is the position and sign (positive or negative) of each pulse, which are collectively called an algebraic code.

Next, gain quantization will be described. In the G.729A system, the algebraic codebook gain is not directly quantized, but the adaptive codebook gain Ga (= βopt) and the correction coefficient γ of the algebraic codebook gain GC are vector quantized. Here, between the algebraic codebook gain GC and the correction coefficient γ
GC = g ′ × γ
There is a relationship. g ′ is the gain of the current frame predicted from the logarithmic gain of the past 4 subframes. In the gain quantization table (gain codebook) (not shown) of the gain quantizer 12, 128 combinations (= 27) of the correction code γ for the adaptive codebook gain Ga and the algebraic codebook gain are prepared. The gain codebook search method is as follows: (1) For the adaptive codebook output vector and the algebraic codebook output vector, one set of table values is extracted from the gain quantization table and set in the gain variable sections 13 and 14. (2) The gain variable units 13 and 14 multiply the respective vectors by the gains Ga and Gc and input them to the LPC synthesis filter 6. (3) The error power evaluation unit 10 has the smallest error power with the input signal X. This is done by selecting a combination.

  From the above, the line coding unit 15 (1) an LSP code that is an LSP quantization index, (2) a pitch lag code Lopt, (3) an algebraic code that is an algebraic codebook index, and (4) a gain quantization index. The line code is generated by multiplexing the gain code, and is transmitted to the decoder. As described above, the G.729A encoding scheme can efficiently compress speech by modeling the speech generation process and quantizing and transmitting the feature parameters of the model.

Decoder Configuration and Operation FIG. 20 is a block diagram of a G.729A decoder. The line data sent from the encoder side is input to the line decoding unit 21, and an LSP code, pitch lag code, algebraic code, and gain code are output. The decoder decodes the audio data based on these codes. The operation of the decoder is partly duplicated because the function of the decoder is included in the encoder, but will be briefly described below.
When the LSP code is input, the LSP inverse quantization unit 22 performs inverse quantization and outputs an LSP inverse quantization value. The LSP interpolation unit 23 interpolates the LSP inverse quantization value of the first subframe of the current frame from the LSP inverse quantization value of the second subframe of the current frame and the LSP inverse quantization value of the second subframe of the previous frame. . Next, the parameter inverse conversion unit 24 converts the LSP interpolation value and the LSP inverse quantization value into LPC synthesis filter coefficients, respectively. The G.729A LPC synthesis filter 25 uses the LPC coefficient converted from the LSP interpolation value in the first first subframe, and uses the LPC coefficient converted from the LSP inverse quantization value in the next second subframe. .

The adaptive codebook 26 outputs a pitch signal having a subframe length (= 40 samples) from the reading start position indicated by the pitch lag code, and the noise codebook 27 indicates the pulse position and pulse polarity from the reading position corresponding to the algebraic code. Output. The gain dequantization unit 28 calculates an adaptive codebook gain dequantization value and an algebraic codebook gain dequantization value from the input gain code, and sets them in the gain variable units 29 and 30. The adder 31 adds the signal obtained by multiplying the adaptive codebook output by the adaptive codebook gain inverse quantization value and the signal obtained by multiplying the algebraic codebook output by the algebraic codebook gain inverse quantization value, A signal is created, and this sound source signal is input to the LPC synthesis filter 25. Thereby, reproduced sound can be obtained from the LPC synthesis filter 25.
In the initial state, the contents of the adaptive codebook 26 on the decoder side all contain a signal with an amplitude of 0. For each subframe, the oldest signal in time is discarded by the subframe length, while in the current subframe. It operates so as to store the obtained excitation signal in the adaptive codebook 26. In other words, the adaptive codebook 26 of the encoder and decoder is always maintained in the latest state.

(2) Explanation of EVRC
EVRC is characterized in that the number of transmission bits per frame is changed according to the nature of the input signal. That is, the bit rate is increased in a stationary part such as a vowel, and the number of transmission bits is reduced in a silent part or a transient part, thereby reducing the time average bit rate. Table 1 shows the EVRC bit rates.

  In EVRC, the rate is determined for the input signal of the current frame. In the rate determination, the frequency region of the input audio signal is divided into a low region and a high region, and the power of each band is calculated. Compare the power of each band with two predetermined thresholds, and if both the low-frequency power and high-frequency power are higher than the threshold, select the full rate, and either low-frequency power or high-frequency power only If is higher than the threshold, the half rate is selected. In addition, when both the low frequency power and the high frequency power are lower than the threshold, the 1/8 rate is selected.

FIG. 21 shows the configuration of the EVRC encoder. In EVRC, an input signal divided into 20 msec (160 sample) frames is input to an encoder. An input signal of one frame is divided into three subframes as shown in Table 2. The full rate and half rate have substantially the same encoder configuration, and only the number of quantization bits of each quantizer is different, so the full rate will be described below.

  As shown in FIG. 22, the LPC (Linear Prediction) analysis unit 41 obtains an LPC coefficient by LPC analysis using a total of 240 samples of 160 samples of the input signal of the current frame and 80 samples of prefetch. The LSP quantization unit 42 converts the LPC coefficients into LSP parameters and then quantizes them to obtain an LSP code, and the LSP inverse quantization unit 43 obtains an LSP inverse quantization value from the LSP code. The LSP interpolation unit 44 uses the LSP inverse quantization value obtained in the current frame (the LSP inverse quantization value in the third subframe) and the LSP inverse quantization value in the third subframe obtained in the previous frame. LSP inverse quantization values in the first, second, and third subframes of the current frame are obtained by interpolation calculation.

Next, the pitch analysis unit 45 obtains the pitch lag and pitch gain of the current frame. EVRC performs pitch analysis twice per frame. The analysis window positions for pitch analysis are as shown in FIG. The procedure for pitch analysis is as follows.
(1) The input signal of the current frame and the look-ahead signal are input to an LPC inverse filter composed of the LPC coefficients to obtain an LPC residual signal. If the LPC synthesis filter is H (z), the LPC inverse filter is 1 / H (z).
(2) Obtain the autocorrelation function of the LPC residual signal, and find the pitch lag and gain when the autocorrelation function is maximized.
(3) The above processing is performed at two analysis window positions. The pitch lag and pitch gain obtained in the first analysis are Lag1 and Gain1, respectively, and the pitch lag and pitch gain obtained in the second analysis are Lag2 and Gain2.
(4) When the difference between Gain1 and Gain2 is larger than a predetermined threshold, Gain1 and Lag1 are set as the pitch gain and pitch lag of the current frame. If it is equal to or smaller than the threshold, Gain2 and Lag2 are set as the pitch gain and pitch lag of the current frame, respectively.

  The pitch lag and pitch gain of the current frame are obtained by the above procedure. The pitch gain quantization unit 46 quantizes the pitch gain using a quantization table and outputs a pitch gain code, and the pitch gain dequantization unit 47 dequantizes the pitch gain code and inputs it to the gain variable unit 48 To do. In G.729A, the pitch lag and pitch gain are obtained in units of subframes, whereas in EVRC, the pitch lag and pitch gain are obtained in units of frames.

  Further, EVRC is different in that the input voice correcting unit 49 corrects an input signal in accordance with the pitch lag code. In other words, instead of obtaining the pitch lag and pitch gain that minimize the error from the input signal as in G.729A, in EVRC, the input speech correction unit 46 is determined by the pitch lag and pitch gain obtained by pitch analysis. The input signal is corrected so as to be closest to the adaptive codebook output. Specifically, the input speech correction unit 46 converts the input signal into a residual signal using an LPC inverse filter, and the pitch peak position in the residual signal area becomes the same position as the pitch peak position of the output of the adaptive codebook 47. This is achieved by time shifting.

  Next, the noisy sound source signal and gain are determined in units of subframes. First, the adaptive codebook synthesized signal obtained by passing the adaptive codebook output through the gain variable unit 48 and the LPC synthesis filter 51 is subtracted from the modified input signal of the input speech correcting unit 46 by the arithmetic unit 52 to obtain the target signal X for the algebraic codebook search. ′ Is generated. The EVRC algebraic codebook 53 is composed of a plurality of pulses as in G.729A, and 35 bits are allocated per subframe at the full rate. Table 3 shows the full-rate pulse positions.

代数符号帳の探索方法はG.729Aと同様であるが、各パルス系統から選ぶパルスの本数が異なる。5つのパルス系統のうち3系統に2パルスを割り当て、2系統に1パルスを割り当てる。ただし、1パルスを割り当てる系統の組み合わせはT3-T4,T4-T0, T0-T1, T1-T2の４通りに限定されている。従って、パルス系統とパルス本数の組合わせは表4のようになる。

The algebraic codebook search method is the same as G.729A, but the number of pulses selected from each pulse system is different. 2 pulses are assigned to 3 of 5 pulse systems, and 1 pulse is assigned to 2 systems. However, the combinations of systems to which one pulse is assigned are limited to four combinations of T3-T4, T4-T0, T0-T1, and T1-T2. Therefore, combinations of pulse system and number of pulses are as shown in Table 4.

以上のように1パルスを割り当てる系統と2パルスを割り当てる系統があるため、パルス本数によって各パルス系統に割り当てるビット数が異なっている。表５にフルレートの代数符号帳のビット配分を示す。

As described above, since there are a system for assigning one pulse and a system for assigning two pulses, the number of bits assigned to each pulse system differs depending on the number of pulses. Table 5 shows the bit allocation of the full rate algebraic codebook.

1本のパルス系統の組み合わせは表4より4通りあるため、2ビット必要である。パルス数が1本である２つのパルス系統における11個のパルス位置をそれぞれX,Y方向に配列すると、１１×11の格子点が形成でき、1つの格子点により２つのパルス系統のパルス位置を特定することができる。従って、パルス数が1本である２つのパルス系統のパルス位置を特定するために7ビット必要であり、パルス数が1本である２つのパルス系統のパルスの極性を表現するのに2ビット必要である。また、パルス数が２本である２つのパルス系統のパルス位置を特定するために7×３ビット必要であり、パルス数が２本である３つのパルス系統のパルスの極性を表現するのに1×３ビット必要である。尚、1系統のパルスの極性は同じである。以上より、EVRCにおいて代数符号はトータル35ビットで表現される。

Since there are four combinations of one pulse system from Table 4, 2 bits are required. If 11 pulse positions in two pulse systems with one pulse are arranged in the X and Y directions respectively, 11 × 11 lattice points can be formed, and the pulse positions of the two pulse systems can be formed by one lattice point. Can be identified. Therefore, 7 bits are required to specify the pulse position of two pulse systems with one pulse, and 2 bits are required to express the polarity of the pulses of two pulse systems with one pulse. It is. In addition, 7 × 3 bits are required to specify the pulse position of two pulse systems having two pulses, and 1 is used to express the polarity of the pulses of three pulse systems having two pulses. * 3 bits are required. Note that the polarity of one pulse is the same. From the above, the algebraic code is expressed by a total of 35 bits in EVRC.

In the algebraic codebook search, the algebraic codebook 53 sequentially inputs a pulse signal to the gain multiplier 54 and the LPC synthesis filter 55 to generate an algebraic synthesized signal, and the arithmetic unit 56 calculates the algebraic synthesized signal and the target signal X ′. Calculate the difference
D = | X′−GC · A · Ck | ²
The code vector Ck that minimizes the evaluation function error power D is obtained. GC is the algebraic codebook gain. In the search of the algebraic codebook, the error power evaluator 59 normalizes the mutual value obtained by normalizing the square of the cross-correlation value Rcx between the algebraic composite signal A · Ck and the target signal X ′ with the autocorrelation value Rcc of the algebraic composite signal. A search is made for a combination of a pulse position and a polarity having the largest correlation value (Rcx * Rcx / Rcc).

The algebraic codebook gain is not directly quantized, but the algebraic codebook gain correction coefficient γ is scalar quantized with 5 bits per subframe. The correction coefficient γ is a value (γ = Gc / g ′) obtained by normalizing the algebraic codebook gain Gc with the gain predicted from the past subframe as g ′. As described above, the multiplexing unit 60 uses (1) an LSP code as an LSP quantization index, (2) a pitch lag code, (3) an algebraic code as an algebraic codebook index, and (4) a pitch gain quantization index. A line data is created by multiplexing a certain pitch gain code and (5) an algebraic codebook gain code which is a quantization index of the algebraic codebook gain, and transmits it to the decoder.
The decoder is configured to decode the speech data using the LSP code, pitch lag code, algebraic code, pitch gain code, and algebraic codebook gain code sent from the encoder side. Since the EVRC decoder can be created in the same manner as the G.729 decoder corresponding to the encoder, the description thereof is omitted.

(3) Conventional voice code conversion method With the spread of the Internet and mobile phones, it is considered that the volume of voice calls between Internet users and mobile phone network users will increase in the future. However, since the voice encoding method used differs between the cellular phone network and the Internet, communication cannot be performed as it is. For this reason, conventionally, a speech code encoded in one network is converted into a speech code of a coding system used in the other network by a speech code conversion unit.
FIG. 23 shows a principle diagram of a conventional typical speech code conversion method. Hereinafter, this method is referred to as Prior Art 1. In the figure, only the case where the voice input by the user A to the terminal 71 is transmitted to the terminal 72 of the user B is considered. Here, it is assumed that the terminal 71 possessed by the user A has only the encoder 71a of the encoding scheme 1, and the terminal 72 possessed by the user B has only the decoder 72a of the encoding scheme 2.

  The voice uttered by the user A on the transmission side is input to the encoder 71 a of the encoding method 71 incorporated in the terminal 71. The encoder 71a encodes the input audio signal into an encoding method 1 audio code and sends it to the transmission line 71b. When a speech code is input via the transmission path 71b, the decoder 73a of the speech code conversion unit 73 once decodes the reproduced speech from the speech code of the encoding scheme 1. Subsequently, the encoder 73b of the voice code conversion unit 73 converts the reproduced voice signal into a voice code of the encoding method 2 and sends it to the transmission path 72b. The voice code of this encoding method 2 is input to the terminal 72 through the transmission path 72b. When the speech code is input, the decoder 72a decodes the reproduced speech from the speech code of the encoding method 2. Thereby, the user B on the receiving side can listen to the reproduced sound. The process of decoding the speech once encoded as described above and encoding the decoded speech again is called tandem connection.

  As described above, in the configuration of the conventional technique 1, since the speech code encoded by the speech encoding method 1 is once decoded into the encoded speech and then encoded again by the speech encoding method 2, the speech quality is improved. There were problems such as significant deterioration of the network and increased delay. That is, once encoded and information-compressed sound (reproduced sound) has a smaller amount of sound information than the original sound (original sound), and the sound quality of the reproduced sound is strictly worse than the original sound. In particular, in recent low bit rate speech coding schemes typified by G.729A and EVRC, a large amount of information contained in the input speech is discarded and coded in order to achieve a high compression rate. When tandem connection is repeated repeatedly, there is a problem that the quality of the playback audio deteriorates remarkably

As a method for solving such a problem of tandem connection, the speech code is decomposed into parameter codes such as LSP code and pitch lag code without returning to speech signals, and each parameter code is individually converted into another speech coding method. A method of converting to a code has been proposed (see Japanese Patent Application No. 2001-75427). FIG. 24 shows the principle diagram. Hereinafter, this is referred to as Prior Art 2.
The encoding method 1 encoder 71a incorporated in the terminal 71 encodes the audio signal emitted by the user A into the encoding method 1 audio code and sends it to the transmission line 71b. The voice code conversion unit 74 converts the voice code of the encoding method 1 input from the transmission line 71b into the voice code of the coding method 2 and sends it to the transmission line 72b. The decoder 72a of the terminal 72 uses the transmission line 72b. The user B can listen to the reproduced voice by decoding the reproduced voice from the voice code of the encoding method 2 input via the user.

  The encoding method 1 includes (1) a first LSP code obtained by quantizing an LSP parameter obtained from a linear prediction coefficient (LPC coefficient) obtained by linear prediction analysis for each frame, and (2) a periodic sound source. A first pitch lag code that specifies an output signal of an adaptive codebook for outputting a signal, and (3) an output signal of an algebraic codebook (or a noise codebook) for outputting a noisy excitation signal. 1 is obtained by quantizing an algebraic code (noise code), and (4) a pitch gain representing the amplitude of the output signal of the adaptive codebook and an algebraic codebook gain representing the amplitude of the output signal of the algebraic codebook. This is a method of encoding an audio signal with a gain code of 1. The encoding method 2 is obtained by quantization by a quantization method different from the first speech encoding method (1) a second LSP code, (2) a second pitch lag code, (3) a second In this method, a speech signal is encoded with an algebraic code (noise code) and (4) a second gain code.

  The speech code conversion unit 74 includes a code separation unit 74a, an LSP code conversion unit 74b, a pitch lag code conversion unit 74c, an algebraic code conversion unit 74d, a gain code conversion unit 74e, and a code multiplexing unit 74f. The code separation unit 74a uses a code of a plurality of components necessary for reproducing a voice signal from the voice code of the coding method 1 inputted from the encoder 71a of the terminal 1 via the transmission path 71b, that is, (1) An LSP code, (2) a pitch lag code, (3) an algebraic code, and (4) a gain code are separated and input to the code conversion units 74b to 74e, respectively. Each of the code conversion units 74b to 74e converts the input LSP code, pitch lag code, algebraic code, and gain code according to the speech coding method 1 into an LSP code, pitch lag code, algebraic code, and gain code according to the speech coding method 2, respectively. The code multiplexing unit 74f multiplexes the converted codes of the audio coding method 2 and sends them to the transmission path 72b.

  FIG. 25 is a block diagram of the voice code conversion unit 74 in which the configuration of each of the code conversion units 74b to 74e is clarified. The same parts as those in FIG. The code separation unit 74a separates the LSP code 1, the pitch lag code 1, the algebraic code 1, and the gain code 1 from the coding method 1 speech code input from the transmission line via the input terminal # 1, and the code conversion unit 74b. Input to 74e.

The LSP dequantizer 74b ₁ of the LSP code converter 74b dequantizes the LSP code 1 of the encoding scheme 1 and outputs an LSP dequantized value, and the LSP quantizer 74b ₂ outputs the LSP dequantized value. Is quantized using the LSP quantization table of encoding method 2 and LSP code 2 is output. The pitch lag dequantizer 74c ₁ of the pitch lag code converter 74c dequantizes the pitch lag code ₁ of the encoding scheme 1 and outputs a pitch lag dequantized value, and the pitch lag quantizer 74c ₂ outputs the pitch lag dequantized value. Is quantized using the pitch lag quantization table of encoding method 2 and pitch lag code 2 is output. Algebraic code dequantizer 74d ₁ of algebraic code converting unit 74d is the algebraic code 1 of encoding scheme 1 and dequantized outputs algebraic code dequantized value, the algebraic code quantizer 74d ₂ are surrogate number of codes The inverse quantization value is quantized using the algebraic code quantization table of encoding method 2 and algebraic code 2 is output. The gain dequantizer 74e ₁ of the gain code converter 74e dequantizes the gain code ₁ of the encoding scheme 1 and outputs a gain dequantized value, and the gain quantizer 74e ₂ outputs the gain dequantized value. Is quantized using a gain quantization table of encoding method 2 and gain code 2 is output.
The code multiplexer 74f multiplexes the LSP code 2, the pitch lag code 2, the algebraic code 2, and the gain code 2 output from the quantizers 74b _{2 to} 74e ₂ to create a speech code according to the encoding scheme 2 and output it. Send to terminal # 2 to transmission line.

  In the tandem connection method (conventional technology 1) in FIG. 23, the reproduced speech obtained by once decoding the speech code encoded by the encoding method 1 is input, and the encoding and decoding are performed again. For this reason, since speech parameters are extracted from reproduced speech that has a much smaller amount of information than the original sound by re-encoding (i.e., speech information compression), the resulting speech code is not necessarily optimal. There wasn't. On the other hand, according to the speech coding apparatus of the related art 2 in FIG. Compared with the tandem connection of the prior art 1, speech code conversion with much less deterioration is possible. Further, since there is no need to decode the speech once for speech code conversion, there is an advantage that the delay which has been a problem in the conventional tandem connection can be reduced.

G.729Aのフレーム長は10msecであり、サブフレーム長は5msecである。一方、EVRCのフレーム長は20msecであり、１フレームを3つのサブフレームに分割している。このため、EVRCのサブフレーム長は6.625msec(最終サブフレームのみ6.75msec)となり、G.729Aとはフレーム長だけでなく、サブフレーム長も異なっている。表７にG.729AとEVRCのビット割り当てを比較した結果を示す。 In the VoIP network, G.729A is used as a voice encoding method. On the other hand, EVRC is adopted in the cdma2000 network, which is expected as a next-generation mobile phone system. Table 6 shows the results of comparing the main specifications of G.729A and EVRC.

The frame length of G.729A is 10 msec, and the subframe length is 5 msec. On the other hand, the frame length of EVRC is 20 msec, and one frame is divided into three subframes. Therefore, the EVRC subframe length is 6.625 msec (only the final subframe is 6.75 msec), and not only the frame length but also the subframe length is different from that of G.729A. Table 7 shows the result of comparison of bit assignment between G.729A and EVRC.

VoIP網とcdma2000網との間で音声通信をする場合には、一方の音声符号を他方の音声符号に変換するための音声符号変換技術が必要である。このような場合に用いられる技術として、前述した従来技術１と従来技術２が知られている。
ところが、従来技術１では符号化方式１の音声符号から一旦音声を再生し、再生された音声を入力として音声符号化方式２で再度符号化するため、符号化方式の違いに影響されずに符号変換が可能である。ところが、この方法では再符号化する際にLPC分析とピッチ分析のために信号の先読み（すなわち、遅延）が生じるという問題や、音質が大幅に劣化するという問題がある。

When performing voice communication between the VoIP network and the cdma2000 network, a voice code conversion technique for converting one voice code into the other voice code is required. Conventional techniques 1 and 2 described above are known as techniques used in such a case.
However, in the prior art 1, since the voice is once reproduced from the voice code of the coding system 1, and the reproduced voice is encoded again by the voice coding system 2, the code is not affected by the difference in the coding system. Conversion is possible. However, this method has a problem that pre-reading (that is, delay) of a signal occurs due to LPC analysis and pitch analysis when re-encoding, and a problem that sound quality is greatly deteriorated.

  On the other hand, in the speech code conversion method of the prior art 2, since the subframe lengths of the encoding method 1 and the encoding method 2 are converted on the assumption that the subframe lengths of the encoding method 1 and the encoding method 2 are equal, the subframe lengths of the encoding method 1 and the encoding method 2 are different. There was a problem with the code conversion. That is, in the algebraic codebook, pulse position candidates are determined according to the subframe length, and therefore, the pulse positions are completely different between systems with different subframe lengths (G.729A and EVRC). There was a problem that it was difficult to associate the positions one-on-one.

Accordingly, an object of the present invention is to enable voice code conversion even between voice coding systems having different subframe lengths.
Another object of the present invention is to make it possible to reduce deterioration of sound quality and to reduce delay time.
Another object of the present invention is to enable code conversion corresponding to the EVRC rate.

According to a first aspect of the present invention, a first speech code obtained by encoding a speech signal with an LSP code, a pitch lag code, an algebraic code, a pitch gain code, and an algebraic codebook gain code based on the first speech coding scheme is provided. This is a voice code conversion method for converting to a second voice code based on a two-voice coding system, and a full-rate, half-rate, and 1 / 8-rate voice code conversion unit corresponding to the rate of the first voice code. When the rate of the 1st speech code is 1/8 rate, the speech code conversion unit for 1/8 rate separates the LSP code and the algebraic codebook gain code included in the speech code by the 1st speech coding method And dequantizing each of them to output an inverse quantized value. The inverse quantized value of the LSP code among the inverse quantized values is quantized by the second speech coding method to obtain the LSP code of the second speech code. Step, random signal output from the sound source generator Generating a target signal by multiplying the inverse quantization value of the gain code and inputting the multiplication result to an LPC synthesis filter composed of the inverse quantization value of the LSP code, an arbitrary algebra in the second speech coding system Generating an algebraic synthesized signal using a code and an inverse quantized value of the LSP code of the second speech code, and an algebraic code in the second speech coding system that minimizes a difference between the target signal and the algebraic synthesized signal Seeking
The second speech code using the inverse quantization value of the LSP code of the second speech code, the inverse quantization value of the pitch lag code obtained by full-rate or half-rate speech code conversion, the obtained algebraic code and the target signal Obtaining a gain code of the encoding system, and outputting an LSP code, an algebraic code, a gain code, and the pitch lag code in the obtained second speech encoding system.
In the speech code conversion method of the present invention, when the rate of the first speech code is a full rate, the speech code conversion unit for full rate dequantizes each code constituting the first speech code, and among the inverse quantization values The step of obtaining the LSP code and pitch lag code of the second speech code by quantizing the inverse quantization values of the LSP code and pitch lag code by the second speech coding method, and using the inverse quantization value of the pitch gain code of the first speech code A step of obtaining an inverse quantized value of the pitch gain code of the second speech code by interpolation processing, and generating a pitch periodic composite signal using the LSP code, pitch lag code, and the inverse quantized value of the pitch gain of the second speech code And a step of reproducing an audio signal from the first audio code and generating a difference signal between the reproduced audio signal and the pitch periodic synthesized signal as a target signal, a second audio encoding method Generating an algebraic composite signal using an arbitrary algebraic code and an inverse quantized value of the LSP code of the second speech code, and a second speech encoding that minimizes a difference between the target signal and the algebraic composite signal A step of obtaining an algebraic code in the system, a step of obtaining a gain code of the second speech coding system using the LSP code of the second speech code, an inverse quantization value of the pitch lag code, the obtained algebraic code and the target signal, A step of outputting an LSP code, a pitch lag code, an algebraic code, and a gain code in the obtained second speech coding method.
According to a second aspect of the present invention, a first speech code obtained by encoding a speech signal with an LSP code, a pitch lag code, an algebraic code, a pitch gain code, and an algebraic codebook gain code based on the first speech coding scheme is provided. This is a speech code conversion device for converting to a second speech code based on a 2-speech coding scheme, and a speech code conversion unit for full rate, half rate, and 1/8 rate corresponding to the rate of the first speech code. The 1/8 rate speech code conversion unit separates the LSP code and the algebraic codebook gain code included in the speech code by the first speech coding method, and inverse quantizes each to obtain an inverse quantized value. From the inverse quantization unit that outputs, the LSP quantization unit that quantizes the inverse quantization value of the LSP code among the inverse quantization values by the second speech coding method and outputs the LSP code of the second speech code, and the sound source generation unit The inverse quantization value of the gain code is added to the output random signal. A target generation unit for generating a target signal by inputting the multiplication result to an LPC synthesis filter composed of an inverse quantization value of the LSP code, an arbitrary algebraic code in the second speech coding scheme, and the second speech code An algebraic code acquisition unit for generating an algebraic synthesized signal using a dequantized value of the LSP code of the second speech encoding method, and obtaining an algebraic code in the second speech coding scheme that minimizes a difference between the target signal and the algebraic synthesized signal; Second speech coding using the inverse quantization value of the LSP code of two speech codes, the inverse quantization value of the pitch lag code obtained by full-rate or half-rate speech code conversion, the obtained algebraic code and the target signal A gain code acquisition unit for obtaining a gain code of the system, a code that multiplexes and outputs the LSP code, the algebraic code, the gain code, and the pitch lag code in the obtained second speech coding system It has a heavy part.
In the speech code conversion device according to the present invention, the full-rate speech code conversion unit performs inverse quantization on each code constituting the first speech code, and among the inverse quantization values, an LSP code and an inverse quantization value of a pitch lag code Is quantized by the second speech coding method to obtain the LSP code of the second speech code and the pitch lag code, and the second speech by interpolation using the inverse quantization value of the pitch gain code of the first speech code. A pitch gain interpolation unit for obtaining a dequantized value of a pitch gain code of the code, an LSP code of the second speech code, a pitch lag code, and a pitch gain dequantized value of the pitch gain. A target generator that reproduces an audio signal from one audio code, and generates a difference signal between the reproduced audio signal and the pitch periodic synthesized signal as a target signal; an arbitrary in the second audio encoding method; An algebraic synthesized signal is generated using an inverse quantized value of the LSP code of the number code and the second speech code, and an algebraic code in the second speech coding system that minimizes the difference between the target signal and the algebraic synthesized signal An algebraic code obtaining unit for obtaining a gain code for obtaining a gain code of the second speech coding method using the LSP code of the second speech code, the inverse quantization value of the pitch lag code, the obtained algebraic code and the target signal A code multiplexing unit that multiplexes and outputs the LSP code, pitch lag code, algebraic code, and gain code in the obtained second speech encoding method.

According to the present invention, speech code conversion can be performed even between speech coding systems having different subframe lengths, sound quality deterioration can be reduced, and delay time can be reduced.
Further, according to the present invention, code conversion corresponding to the EVRC rate can be performed.

(A) Outline of the Present Invention FIG. 1 is a diagram for explaining the principle of a speech code conversion apparatus according to the present invention. The speech code CODE1 of encoding method 1 (G.729A) is changed to the speech code CODE2 of encoding method 2 (EVRC). The principle structure of the speech code converter in the case of converting is shown.
In the present invention, the LSP code, pitch lag code, and pitch gain code are converted from encoding method 1 to encoding method 2 in the quantization parameter area in the same manner as in prior art 2, and the reproduced speech and the pitch periodic composite signal are converted. Then, a target signal is created, and an algebraic code and algebraic codebook gain are obtained so that an error between the target signal and the algebraic synthesized signal is minimized. This is characterized in that the encoding method 1 is converted to the encoding method 2. The details of the conversion procedure will be described with reference to FIG.

When a speech code CODE1 of encoding method 1 (G.729A) is input, the code separation unit 101 separates the speech code CODE1 into parameter codes of LSP code Lsp1, pitch lag code Lag1, pitch gain code Gain1, and algebraic code Cb1, The data is input to the LSP code conversion unit 102, the pitch lag conversion unit 103, the pitch gain conversion unit 104, and the audio reproduction unit 105.
The LSP code converter 102 converts the LSP code Lsp1 into the LSP code Lsp2 of the encoding scheme 2. The pitch lag conversion unit 103 converts the pitch lag code Lag1 into the pitch lag code Lag2 of the encoding scheme 2. The pitch gain conversion unit 104 obtains a pitch gain inverse quantization value from the pitch gain code Gain1, and converts the pitch gain inverse quantization value into a pitch gain code Gp2 of the encoding scheme 2.

The audio reproduction unit 105 reproduces the audio signal Sp using the LSP code Lsp1, the pitch lag code Lag1, the pitch gain code Gain1, and the algebraic code Cb1, which are code components of the audio code CODE1. The target creation unit 106 creates a pitch periodicity synthesized signal of the encoding method 2 from the LSP code Lsp2, the pitch lag code Lag2, and the pitch gain code Gp2 of the speech encoding method 2. Thereafter, the target creating unit 106 creates the target signal Target by subtracting the pitch periodic synthesized signal from the reproduced audio signal Sp.
The algebraic code conversion unit 107 generates an algebraic synthesized signal using an arbitrary algebraic code in the speech coding scheme 2 and an inverse quantization value of the LSP code Lsp2 in the speech coding scheme 2, and generates a target signal Target and the algebraic synthesized signal. The algebraic code Cb2 of the speech coding scheme 2 that minimizes the difference between the two is determined.

The algebraic codebook gain conversion unit 108 inputs an algebraic codebook output signal corresponding to the algebraic code Cb2 of the speech coding method 2 to an LPC synthesis filter composed of an inverse quantization value of the LSP code Lsp2, and inputs the algebraic synthesized signal , And an algebraic codebook gain is determined from the algebraic synthesized signal and the target signal, and an algebraic codebook gain code is generated using the algebraic codebook gain using a quantization table of encoding scheme 2.
The code multiplexing unit 109 multiplexes the LSP code Lsp2, the pitch lag code Lag2, the pitch gain code Gp2, the algebraic code Cb2, and the algebraic codebook gain code Gc2 obtained as described above as the speech code CODE2 of the encoding system 2. Output.

(B) First Embodiment FIG. 2 is a block diagram of a speech code conversion apparatus according to a first embodiment of the present invention. The same reference numerals are given to the same parts as in the principle diagram of FIG. In the present embodiment, G.729A is used as the speech encoding scheme 1 and EVRC is used as the speech encoding scheme 2. EVRC has three modes of full rate, half rate, and 1/8 rate. Here, only full rate is used.
Since the frame length of G.729A is 10 msec and the frame length of EVRC is 20 msec, the voice code for two frames of G.729A is converted into the voice code for one frame of EVRC. In the following, a case will be described in which the G.729A frame n and n + 1 frame speech codes shown in FIG. 3 (a) are converted into EVRC m frame speech codes shown in FIG. 3 (b). .

In FIG. 2, the voice code (line data) CODE1 (n) of the nth frame is input to the terminal # 1 from the G.729A encoder (not shown) via the transmission line. The code separation unit 101 separates the LSP code Lsp1 (n), the pitch lag code Lag1 (n, j), the gain code Gain1 (n, j), and the algebraic code Cb1 (n, j) from the speech code CODE1 (n). Each of the transform units 102, 103, 104 and the algebraic code inverse quantization unit 110 is input. Here, the subscript j in parentheses represents a subframe number (see FIG. 3A), and takes a value of 0 or 1.
The LSP code converter 102 includes an LSP inverse quantizer 102a and an LSP quantizer 102b. As described above, the frame length of G.729A is 10 msec, and the G.729A encoder quantizes the LSP parameter obtained from the input signal of the first subframe once every 10 msec. On the other hand, the frame length of EVRC is 20 msec, and the EVRC encoder quantizes the LSP parameter obtained from the input signal of the second subframe and the prefetch portion only once every 20 msec. That is, considering the same 20 msec as a unit, the G.729A encoder performs LSP quantization twice, whereas the EVRC encoder performs quantization only once. For this reason, the LSP code of two consecutive frames of G.729 cannot be converted into the EVRC LSP code as it is.

Therefore, in the first embodiment, only the LSP code in the odd frame ((n + 1) th frame) of G.729A is converted into the EVRC LSP code, and the LSP code in the even frame (nth frame) is not converted. It was. However, it is possible to convert the LSP code of the even frame into the LSP code of EVRC and not convert the LSP code of the odd frame.
When the LSP code Lsp1 (n) is input, the LSP dequantization unit 102a dequantizes the code and outputs an LSP dequantized value lsp1. Here, lsp1 is a vector composed of 10 coefficients. The LSP inverse quantization unit 102a performs the same operation as the inverse quantizer used in the G.729A decoder.

When the LSP inverse quantized value lsp1 of the odd frame is input to the LSP quantizing unit 102b, the LSP quantizing unit 102b quantizes it according to the EVRC LSP quantization method and outputs the LSP code Lsp2 (m). Here, the LSP quantization unit 102b is not necessarily the same as the quantizer used in the EVRC encoder, but at least the LSP quantization table uses the same table as the EVRC quantization table. . Note that the LSP inverse quantization value of the even frame is not used for LSP code conversion. Further, the LSP inverse quantization value lsp1 is used as a coefficient of the LPC synthesis filter in the audio reproduction unit 105 described later.
Next, the LSP quantization unit 102b decodes the converted LSP code Lsp2 (m) and the LSP inverse quantization value obtained by decoding the LSP code Lsp2 (m-1) of the previous frame. LSP parameters lsp2 (k) and (k = 0, 1, 2) in three subframes in the current frame are obtained from the converted values by linear interpolation. lsp2 (k) is used in the target generation unit 106 and the like described later. lsp2 (k) is a 10-dimensional vector.

The pitch lag conversion unit 103 includes a pitch lag inverse quantization unit 103a and a pitch lag quantization unit 103b. In G.729A, the pitch lag is quantized every 5 msec subframe. On the other hand, EVRC quantizes the pitch lag only once per frame. Considering 20 msec as a unit, G.729A quantizes four pitch lags, whereas EVRC quantizes only one pitch lag. Therefore, when converting a G.729A speech code into an EVRC speech code, it is not possible to convert all pitch lags in G.729A into EVRC pitch lags.
Therefore, in the first embodiment, the pitch lag code Lag1 (n + 1,1) in the last subframe (first subframe) of the (n + 1) th frame of G.729A is inverted by the pitch lag inverse quantization unit 103a of G.729A. The pitch lag lag1 is obtained by quantization, and this lag1 is quantized by the EVRC pitch lag quantization unit 103b to obtain the pitch lag code Lag2 (m) in the m-th frame second subframe. Further, the pitch lag quantization unit 103b performs pitch lag interpolation by the same method as the EVRC encoder / decoder. That is, the pitch lag interpolation value of each subframe is obtained by linear interpolation between the pitch lag inverse quantization value of the second subframe obtained by dequantizing Lag2 (m) and the pitch lag inverse quantization value of the second subframe of the previous frame. Find lag2 (k), (k = 0,1,2). The pitch lag interpolation value is used by the target generation unit 106 described later.

The pitch gain converter 104 includes a pitch gain inverse quantizer 104a and a pitch gain quantizer 104b. In G.729A, the pitch gain is quantized every 5 msec sub-frame, so when considering 20 msec as a unit, G.729A quantizes four pitch gains in one frame. On the other hand, EVRC quantizes three pitch gains in one frame. Therefore, when converting a G.729A speech code into an EVRC speech code, it is not possible to convert all pitch gains of G.729A into EVRC pitch gains. Therefore, in the first embodiment, gain conversion is performed by the method shown in FIG. That is, gp1 (0), gp1 (1), gp1 (2), gp1 (3) are set as the pitch gain of two consecutive frames of G.729A
gp2 (0) = gp1 (0)
gp2 (1) = (gp1 (1) + gp1 (2)) / 2
gp2 (2) = gp1 (3)
To synthesize pitch gain. The synthesized pitch gain gp2 (k) (k = 0, 1, 2) is scalar quantized using the EVRC pitch gain quantization table to obtain a pitch gain code Gp2 (m, k). The pitch gain gp2 (k) (k = 0, 1, 2) is used by the target generation unit 106 described later.

The algebraic code dequantization unit 110 dequantizes the algebraic code Cb (n, j), and inputs the obtained algebraic code dequantized value cb1 (j) to the speech reproduction unit 105.
The audio playback unit 105 creates G.729A playback audio Sp (n, h) in the nth frame and G.729A playback audio Sp (n + 1, h) in the n + 1th frame. Note that the method of creating reproduced audio is the same as the operation of the G.729A decoder, and has been described in the section of the prior art, and the description thereof is omitted here. The number of dimensions of the playback audio Sp (n, h) and Sp (n + 1, h) is 80 samples, the same as the frame length of G.729A (h = 1 to 80). The number of samples per unit. The audio reproducing unit 105 generates the reproduced audio Sp (n, h), Sp (n + 1, h) as shown in FIG. 5 as Sp (0, i), Sp (1, i), Sp (2, Divided into three vectors i) and output. i is 1 to 53 in the 0th and 1st subframes of EVRC, and 1 to 54 in the 2nd subframe.

  The target generation unit 106 generates a target signal Target (k, I) that is used as a reference signal in the algebraic code conversion unit 107 and the algebraic code length gain conversion unit 108. FIG. 6 is a configuration diagram of the target generation unit 106. The adaptive codebook 106a outputs N sample signals acb (k, i) (i = 0 to N−1) corresponding to the pitch lag lag2 (k) obtained by the pitch lag code conversion unit 103. Here, k is the EVRC subframe number, N is the EVRC subframe length, and is 53 for the 0th and 1st subframes and 54 for the 2nd subframe. Hereinafter, the suffix i is 53 or 54 unless otherwise specified. Reference numeral 106e denotes an adaptive codebook update unit.

  Gain multiplication section 106b multiplies adaptive codebook output acb (k, i) by pitch gain gp2 (k) and inputs the result to LPC synthesis filter 106c. The LPC synthesis filter 106c is configured with an inverse quantization value lsp2 (k) of the LSP code, and outputs an adaptive codebook synthesis signal syn (k, i). The computing unit 106d obtains the target signal Target (k, i) by subtracting the adaptive codebook synthesized signal syn (k, i) from the reproduction signal Sp (k, i) divided into three. Target (k, i) is used in an algebraic code converter 107 and an algebraic codebook gain converter 108 described later.

  The algebraic code conversion unit 107 performs exactly the same processing as the EVRC algebraic code search. FIG. 7 is a configuration diagram of the algebraic code converter 107. The algebraic codebook 107a outputs an arbitrary pulsed sound source signal that can be generated by a combination of pulse positions and polarities shown in Table 3. That is, when the algebraic codebook 107a is instructed to output a pulsed excitation signal corresponding to a predetermined algebraic code from the error evaluation unit 107b, the algebraic codebook 107a sends the pulsed excitation signal corresponding to the instructed algebraic code to the LPC synthesis filter 107c. input. The LPC synthesis filter 107c composed of the dequantized value lsp2 (k) of the LSP code creates and outputs an algebraic synthesized signal alg (k, i) when an algebraic codebook output signal is input. The error evaluation unit 107b calculates the cross-correlation value Rcx of the algebraic synthesized signal alg (k, i) and the target signal Target (k, i) and the autocorrelation value Rcc of the algebraic synthesized signal, and normalizes the square of Rcx with Rcc. The algebraic code Cb2 (m, k) having the largest normalized cross-correlation value Rcx · Rcx / Rcc obtained by the conversion is searched and output.

  The algebraic codebook gain converter 108 has the configuration shown in FIG. The algebraic codebook 108a generates a pulsed excitation signal corresponding to the algebraic code Cb2 (m, k) obtained by the algebraic code converter 107 and inputs it to the LPC synthesis filter 108b. The LPC synthesis filter 108b composed of the dequantized value lsp2 (k) of the LSP code creates and outputs an algebraic synthesized signal gan (k, i) when an algebraic codebook output signal is input. The algebraic codebook gain calculation unit 108c obtains the cross-correlation value Rcx between the algebraic synthesized signal gan (k, i) and the target signal Target (k, i) and the autocorrelation value Rcc of the algebraic synthesized signal, and then calculates Rcx. The algebraic codebook gain gc2 (k) (= Rcx / Rcc) is obtained by normalization with Rcc. The algebraic codebook gain quantization unit 108d performs scalar quantization on the algebraic codebook gain gc (k) 2 using the EVRC algebraic codebook gain quantization table 108e. In EVRC, 5 bits (32 patterns) are assigned per subframe as quantization bits of algebraic codebook gain. Therefore, the table value closest to gc2 (k) is searched from among these 32 table values, and the index value at that time is set as the converted algebraic codebook gain code Gc2 (m, k).

After conversion of the pitch lag code, pitch gain code, algebraic code, and algebraic codebook gain code is completed for one EVRC subframe, the adaptive codebook 106a (FIG. 6) is updated. In the initial state, all signals with an amplitude of 0 are stored in the adaptive codebook 106a. When the subframe conversion process is completed, adaptive codebook updating section 106e in FIG. 6 discards the oldest temporal signal in the adaptive codebook by the subframe length, shifts the remaining signals by the subframe length, and converts them. The latest excitation signal immediately after is stored in the adaptive codebook. Here, the latest excitation signal is a periodic excitation signal corresponding to the converted pitch lag code lag2 (k), pitch gain gp2 (k), algebraic code Cb2 (m, k), algebraic codebook gain gc2 (k ) Is a sound source signal synthesized with a noisy sound source signal.
By the above, EVRC LSP code Lsp2 (m), pitch lag code Lag2 (m), pitch gain code Gp2 (m, k), algebraic code Cb2 (m, k), algebraic codebook gain code Gc2 (m, k) If it is obtained, the code multiplexing unit 109 multiplexes these codes and outputs them as a speech code CODE2 (m) of the encoding scheme 2 as one.

In the first embodiment, since the LSP code, pitch lag code, and pitch gain code are code-converted in the quantization parameter area, the analysis error is small compared with the case where the reproduced speech is analyzed again by LPC analysis and pitch analysis, and the sound quality is deteriorated. Less parameter conversion is possible. Further, since the reproduced speech is not subjected to LPC analysis or pitch analysis again, the problem of delay due to code conversion, which has been a problem in the prior art 1, can be solved.
On the other hand, for the algebraic code and the algebraic codebook gain code, a coding method which has been a problem in the prior art 2 is created by creating a target signal from reproduced speech and converting the target signal to minimize an error from the target signal. Even when the configurations of the algebraic codebooks 1 and 2 are greatly different, code conversion with little deterioration in sound quality is possible.

(C) Second Embodiment FIG. 9 is a block diagram of a speech code conversion apparatus according to a second embodiment of the present invention. Components identical with those of the first embodiment of FIG. The second embodiment differs from the first embodiment in that (1) the algebraic codebook gain conversion unit 108 of the first embodiment is removed, and an algebraic codebook gain quantization unit 110 is provided instead. ) In addition to the LSP code, pitch lag code, and pitch gain code, the algebraic codebook gain code is also subjected to code conversion in the quantization parameter area.

The second embodiment differs from the first embodiment only in the algebraic codebook gain code conversion method. The algebraic codebook gain code conversion method of the second embodiment will be described below.
In G.729A, the algebraic codebook gain is quantized every 5 msec sub-frame, so when considering 20 msec as a unit, G.729A quantizes four algebraic codebook gains in one frame. On the other hand, EVRC quantizes three algebraic codebook gains in one frame. Therefore, when converting a G.729A speech code to an EVRC speech code, it is not possible to convert all G.729A algebraic codebook gains to EVRC algebraic codebook gains. Therefore, in the second embodiment, gain conversion is performed by the method shown in FIG. That is, the algebraic codebook gains of two consecutive frames of G.729A are gc1 (0), gc1 (1), gc1 (2), and gc1 (3),
gc2 (0) = gc1 (0)
gc2 (1) = (gc1 (1) + gc1 (2)) / 2
gc2 (2) = gc1 (3)
To synthesize the algebraic codebook gain. The synthesized algebraic codebook gain gc2 (k) (k = 0,1,2) is scalar quantized using the EVRC algebraic codebook gain quantization table, and the algebraic codebook gain code Gc2 (m, k) is obtained. Ask.

In the second embodiment, since the LSP code, pitch lag code, pitch gain code, and algebraic codebook gain code are code-converted in the quantization parameter area, the analysis error is compared with the case where the reproduced speech is subjected to LPC analysis and pitch analysis again. Parameter conversion with small sound quality degradation is possible. Further, since the reproduced speech is not subjected to LPC analysis or pitch analysis again, the problem of delay due to code conversion, which has been a problem in the prior art 1, can be solved.
On the other hand, for algebraic codes, a target signal is generated from reproduced speech and converted so that an error from the target signal is minimized, so that encoding method 1 and encoding method 2 which are problems in conventional technique 2 are obtained. Even if the configurations of the algebraic codebooks are greatly different, code conversion with little deterioration in sound quality is possible.

(D) Third Embodiment FIG. 11 is an overall configuration diagram of a speech code conversion device according to a third embodiment. The third embodiment shows an example in which EVRC speech codes are converted to G.729A speech codes. In FIG. 11, when a speech code is input from an EVRC encoder, a rate determination unit 201 determines an EVRC rate. Since the EVRC speech code includes rate information indicating whether the rate is a full rate, a half rate, or a 1/8 rate, the rate determination unit 201 uses this information to determine the EVRC rate. Then, the rate determination unit 201 switches the switches S1 and S2 according to the rate, and selectively inputs the EVRC speech code to the predetermined rate speech code conversion units 202, 203, and 204, and is output from the rate speech code conversion unit. The G.729A speech code is sent to the G.729A decoder.

Full-rate speech code converter FIG. 12 is a block diagram of the full-rate speech code converter 202. Since the EVRC frame length is 20 msec and the G.729A frame length is 10 msec, the EVRC 1 frame (m-th frame) voice code is changed to 2 G.729A frames (n-th and n + 1-th frames). Convert to speech code.
The mth frame speech code (line data) CODE1 (m) is input to terminal # 1 from an EVRC encoder (not shown) via a transmission line. Code separation unit 301 is LSP code Lsp1 (m), pitch lag code Lag1 (m), pitch gain code Gp1 (m, k), algebraic code Cb1 (m, k), algebraic codebook gain from speech code CODE1 (m) The code Gc1 (m, k) is separated and input to the inverse quantization units 302 to 306. Here, k is a subframe number of EVRC, and takes a value of 0, 1, or 2.

The LSP inverse quantization unit 302 obtains an inverse quantization value lsp1 (m, 2) of the LSP code Lsp1 (m) of subframe number 2. Note that the LSP inverse quantization unit 302 uses the same quantization table as the EVRC decoder. Next, the LSP inverse quantization unit 302 performs the inverse quantization value lsp1 (m−1,2) of subframe number 2 obtained in the same manner in the previous frame (the m−1th frame) and the inverse quantization value lsp1. The inverse quantization values lsp1 (m, 0) and lsp1 (m, 1) of subframe numbers 0 and 1 are obtained by linear interpolation using (m, 2), and the inverse quantization values lsp1 (m , 1) is input to the LSP quantization unit 307. The LSP quantization unit 307 quantizes the inverse quantization value lsp1 (m, 1) using the quantization table of coding scheme 2 (G.729A) to obtain the LSP code Lsp2 (n) of coding scheme 2 In addition, the LSP inverse quantization value lsp2 (n, 1) is obtained. Similarly, the LSP inverse quantization unit 302 inputs the inverse quantization value lsp1 (m, 2) of subframe number 2 to the LSP quantization unit 307, and the LSP code Lsp2 (n + 1) of encoding scheme 2 And its LSP inverse quantization value lsp2 (n + 1, 1). Here, it is assumed that the LSP quantization unit 302 uses the same quantization table as G.729A.
Next, the LSP quantization unit 307 performs linear interpolation between the inverse quantization value lsp2 (n-1,1) obtained in the previous frame (the (n-1) th frame) and the inverse quantization value lsp2 (n, 1) of the current frame. Thus, the inverse quantization value lsp2 (n, 0) of subframe number 0 is obtained. Further, the inverse quantization value lsp2 (n + 1,0) of subframe 0 is obtained by linear interpolation between the inverse quantization value lsp2 (n, 1) and the inverse quantization value lsp2 (n + 1,1). These inverse quantized values lsp2 (n, j) are used for creating a target signal and converting algebraic codes and gain codes.

  The pitch lag inverse quantization unit 303 obtains an inverse quantization value lag1 (m, 2) of the pitch lag code Lag1 (m) of subframe number 2, and uses the inverse quantization value lag1 (m, 2) and the m−1th frame. The inverse quantization values lag1 (m, 0) and lag1 (m, 1) of the subframe numbers 0 and 1 are obtained by linear interpolation of the obtained inverse quantization value lag (m-1, 2) of the subframe number 2. Next, the pitch lag inverse quantization unit 303 inputs the inverse quantization value lag1 (m, 1) to the pitch lag quantization unit 308, and the pitch lag quantization unit 308 obtains the quantization table of coding scheme 2 (G.729A). The pitch lag code Lag2 (n) of the encoding scheme 2 corresponding to the inverse quantized value lag1 (m, 1) is obtained and the inverse quantized value lag2 (n, 1) is obtained. Similarly, the pitch lag inverse quantization unit 303 inputs the inverse quantization value lag1 (m, 2) to the pitch lag quantization unit 308, and the pitch lag quantization unit 308 obtains the pitch lag code Lag2 (n + 1) and vice versa. The quantized value lag2 (n + 1, 1) is obtained. Here, the pitch lag quantization unit 308 uses the same quantization table as G.729A.

Next, the pitch lag quantization unit 308 performs linear interpolation between the inverse quantization value lag2 (n-1,1) obtained in the previous frame (n-1 frame) and the inverse quantization value lag2 (n, 1) of the current frame. Thus, the inverse quantization value lag2 (n, 0) of subframe 0 is obtained. Further, the inverse quantization value lag2 (n + 1, 0) of subframe 0 is obtained by linear interpolation between the inverse quantization value lag2 (n, 1) and the inverse quantization value lag2 (n + 1, 1). These dequantized values lag2 (n, j) are used for creating a target signal and converting a gain code.
The pitch gain inverse quantization unit 304 obtains inverse quantization values gp1 (m, k) of the three pitch gain codes Gp1 (m, k) (k = 0, 1, 2) of the EVRC mth frame, and the pitch gain Input to the interpolation unit 309. The pitch gain interpolation unit 309 uses the inverse quantized value gp1 (m, k), and the pitch gain inverse quantized value gp2 (n, j) (j = 0, 1) of coding scheme 2 (G.729A), gp2 (n + 1, j) (j = 0,1)
(1) gp2 (n, 0) = gp1 (m, 0)
(2) gp2 (n, 1) = (gp1 (m, 0) + gp1 (m, 1)) / 2
(3) gp2 (n + 1, 0) = (gp1 (m, 1) + gp1 (m, 2)) / 2
(4) gp2 (n + 1, 1) = gp1 (m, 2)
Is obtained by interpolation. Note that the pitch gain dequantized value gp2 (n, j) is not directly required for gain code conversion, but is used for generating a target signal.

The audio playback unit 310 inputs the inverse quantization values lsp1 (m, k), lag1 (m, k), gp1 (m, k), cb1 (m, k), gc1 (m, k) of each EVRC code Thus, EVRC playback sound SP (k, i) of 160 samples in total in the m-th frame is created, and these playback signals are converted into two G.729A playback signals Sp (n, h), Sp ( n + 1, h) and output. Here, the method of creating the playback audio is the same as that of the EVRC decoder and is well known, and thus the description thereof is omitted.
The target generation unit 311 has the same configuration as the target generation unit (see FIG. 6) of the first embodiment, and the target signal Target (n, h) used in the algebraic code conversion unit 312 and the algebraic codebook gain conversion unit 313. , Target (n + 1, h). That is, the target generation unit 311 first obtains an adaptive codebook output corresponding to the pitch lag lag2 (n, j) obtained by the pitch lag quantization unit 308, and multiplies it by the pitch gain gp2 (n, j) to generate a sound source signal. Create Next, the excitation signal is input to an LPC synthesis filter composed of LSP dequantized values lsp2 (n, j) to create an adaptive codebook synthesis signal syn (n, h). Thereafter, the target code Target (n, h) is obtained by subtracting the adaptive codebook synthesized signal syn (n, h) from the reproduced speech Sp (n, h) created by the speech reproducing unit 310. Similarly, a target signal Target (n + 1, h) of the (n + 1) th frame is generated.

  The algebraic code converting unit 312 has the same configuration as the algebraic code converting unit (see FIG. 7) of the first embodiment, and performs exactly the same processing as the G.729A algebraic codebook search. First, an algebraic code book signal generated by the combination of pulse position and polarity shown in FIG. 18 is input to an LPC synthesis filter composed of LSP dequantized values lsp2 (n, j) to create an algebraic synthesized signal. Next, the cross-correlation value Rcx between the algebraic synthesized signal and the target signal and the autocorrelation value Rcc of the algebraic synthesized signal are calculated, and the normalized cross-correlation value Rcx · Rcx / obtained by normalizing the square of Rcx with Rcc. Search for the algebraic code Cb2 (n, j) with the largest Rcc. Similarly, an algebraic code Cb2 (n + 1, j) is obtained.

The gain conversion unit 313 performs gain conversion using the target signal Target (n, h), the pitch lag lag2 (n, j), the algebraic code Cb2 (n, j), and the LSP inverse quantization value lsp2 (n, j). The conversion method is the same as the gain quantization in the G.729A encoder. The procedure is shown below.
(1) A set of table values (pitch gain, algebraic codebook gain correction coefficient γ) is extracted from the G.729 gain quantization table.
(2) The signal X is generated by multiplying the adaptive codebook output by the pitch gain table value.
(3) The signal Y is generated by multiplying the algebraic codebook output by the correction coefficient γ and the predicted gain value g ′.
(4) A signal obtained by adding the signal X and the signal Y is input to an LPC synthesis filter composed of the LSP dequantized value lsp2 (n, j) to create a synthesized signal Z.
(5) The error power E between the target signal and the synthesized signal Z is calculated.
(6) The processing of (1) to (5) is performed for all the table values of the gain quantization table, the table value that minimizes the error power E is determined, and the index is set as the gain code Gain2 (n, j). To do. Similarly, from the target signal Target (n + 1, h), pitch lag lag2 (n + 1, j), algebraic code Cb2 (n + 1, j), LSP dequantized value lsp2 (n + 1, j) Gain code Gain2 (n + 1, j) is obtained.

  After that, the code multiplexing unit 314 multiplexes the LSP code Lsp2 (n), the pitch lag code Lag2 (n), the algebraic code Cb2 (n, j), and the gain code Gain2 (n, j) to divide the G.729A nth frame. Output the voice code CODE2 (n). The code multiplexing unit 314 multiplexes the LSP code Lsp2 (n + 1), the pitch lag code Lag2 (n + 1), the algebraic code Cb2 (n + 1, j), and the gain code Gain2 (n + 1, j). The voice code CODE2 (n + 1) in the (n + 1) th frame of G.729A is output. As described above, according to the third embodiment, an EVRC (full rate) speech code can be converted into a G.729A speech code.

Half-rate speech code conversion unit The full-rate and half-rate encoders / decoders differ only in the size of each quantization table, and have substantially the same configuration. Therefore, the half-rate speech code conversion unit 203 can be configured in the same manner as the full-rate speech code conversion unit 202 described above, and can similarly convert a half-rate speech code into a G.729A speech code.

1/8 Rate Speech Code Conversion Unit FIG. 13 is a configuration diagram of the 1/8 rate speech code conversion unit 204. The 1/8 rate is used for non-voice segments such as silence and background noise. The information transmitted at the 1/8 rate is 16 bits in total including the LSP code (8 bits / frame) and the gain code (8 bits / frame), and the excitation signal is not transmitted because it is randomly generated inside the encoder / decoder.
In FIG. 13, when the speech code CODE1 (m) in the mth frame of EVRC (1/8 rate) is input, the code separation unit 401 separates the LSP code Lsp1 (m) and the gain code Gc1 (m). The LSP inverse quantization unit 402 and the LSP quantization unit 403 convert the EVRC LSP code Lsp1 (m) into the G.729A LSP code Lsp2 (n), as in the case of the full rate in FIG. The LSP inverse quantization unit 402 obtains an LSP code inverse quantization value lsp1 (m, k), and the LSP quantization unit 403 outputs an LSP code Lsp2 (n) of G.729A and an LSP code inverse quantization value. Find lsp2 (n, j).

The gain inverse quantization unit 404 obtains a gain inverse quantization value gc1 (m, k) of the gain code Gc1 (m). At the 1/8 rate, only the gain for the noisy sound source signal is used, and the gain for the periodic sound source (pitch gain) is not used.
At the 1/8 rate, the sound source signal is randomly generated inside the encoder / decoder. Therefore, also in the 1/8 rate speech code converter, the sound source generator 405 generates a random signal in the same manner as the EVRC encoder / decoder, and the signal adjusted so that the amplitude of the random signal has a Gaussian distribution. Output as sound source signal Cb1 (m, k). In addition, the method similar to EVRC is used about the generation method of a random signal, and the adjustment method to a Gaussian distribution.

The gain multiplier 406 multiplies the sound source signal Cb1 (m, k) by the gain dequantized value gc1 (m, k) and inputs it to the LPC synthesis filter to input the target signals Target (n, h), Target (n + 1) , h). Note that the LPC synthesis filter 407 is configured with an LSP code inverse quantization value lsp1 (m, k).
Algebraic code conversion section 408 performs algebraic code conversion in the same manner as in the full rate case of FIG. 12, and outputs an algebraic code Cb2 (n, j) of G.729A.
The EVRC 1/8 rate is used for non-speech sections with almost no periodicity such as silence and noise, so there is no pitch lag code. Therefore, a pitch lag code for G.729A is generated by the following method. The 1/8 rate speech code converter 204 extracts the pitch lag codes for G.729A obtained by the pitch lag conversion units 303 and 308 of the full rate or half rate speech code conversion units 202 and 203 and stores them in the pitch lag buffer 409. When the 1/8 rate is selected in the current frame (nth frame), the pitch lag code Lag2 (n, j) in the pitch lag buffer 409 is output. However, the stored contents of the pitch lag buffer are not changed. On the other hand, when the 1/8 rate is not selected in the current frame, the pitch lag code for G.729A obtained by the pitch lag converters 303 and 308 of the speech code converters 202 and 203 at the selected rate (full rate or half rate) is It is stored in the buffer 409.

Gain conversion section 410 performs gain code conversion in the same manner as in the full rate case of FIG. 12, and outputs gain code Gc2 (n, j).
After that, the code multiplexing unit 411 multiplexes the LSP code Lsp2 (n), the pitch lag code Lag2 (n), the algebraic code Cb2 (n, j), and the gain code Gain2 (n, j), and the nth frame of G.729A Output the voice code CODE2 (n). The code multiplexing unit 411 multiplexes the LSP code Lsp2 (n + 1), the pitch lag code Lag2 (n + 1), the algebraic code Cb2 (n + 1, j), and the gain code Gain2 (n + 1, j). The voice code CODE2 (n + 1) in the (n + 1) th frame of G.729A is output. As described above, EVRC (1/8 rate) speech code can be converted to G.729A speech code.

(E) Fourth Embodiment FIG. 14 is a block diagram of a voice code conversion apparatus according to the fourth embodiment, which can cope with a line error in the voice code. The first embodiment of FIG. Are given the same reference numerals. The difference is that (1) the line error detection unit 501 is provided, (2) the LSP inverse quantization unit 102a, the pitch lag inverse quantization unit 103a, the gain inverse quantization unit 104a, and the algebraic code inverse quantization unit 110. Instead, an LSP code correction unit 511, a pitch lag correction unit 512, a gain code correction unit 513, and an algebraic code correction unit 514 are provided.

  When the input speech xin is input to the encoding method 1 (G.729A) encoder 500, the encoder 500 generates the encoding method 1 speech code sp1. The speech code sp1 is input to the speech code conversion device through a transmission line 502 of a wireless line or a wired line (Internet or the like). Here, if a line error ERR is mixed before being input to the speech code converter, the speech code sp1 is transformed into a speech code sp ′ containing a line error. The pattern of line error ERR depends on the system and can take various patterns such as random bit errors and burst errors. If no error is mixed, sp1 ′ and sp1 have exactly the same sign. The speech code sp1 ′ is input to the code separation unit 101 and separated into an LSP code Lsp1 (n), a pitch lag code Lag1 (n, j), an algebraic code Cbi (n, j), and a gain code Gain1 (n, j). . The voice code sp1 'is input to the line error detecting unit 501, and the presence or absence of a line error is detected by a known method. For example, a line error can be detected by adding a CRC code to the voice code sp1.

When an error-free LSP code Lsp1 (n) is input, the LSP correction unit 511 performs the same processing as the LSP inverse quantization unit 102a of the first embodiment and outputs an LSP inverse quantization value lsp1. On the other hand, when the correct Lsp code of the current frame cannot be received due to a line error or frame loss, the LSP inverse quantization value lsp1 is output using the Lsp code of the last four good frames received.
Pitch lag correction unit 512 outputs a dequantized value lag1 of the pitch lag code of the received current frame unless a line error or frame disappears. If there is a line error or frame loss, the dequantized value of the pitch lag code of the last good frame received is output. Generally, it is known that the pitch lag changes smoothly in the voiced portion. Therefore, in the voiced portion, there is almost no deterioration in sound quality even when the pitch lag of the previous frame is substituted as described above. Further, it is known that the pitch lag changes greatly in the unvoiced part, but since the contribution rate of the adaptive codebook in the unvoiced part is small (the pitch gain is small), there is almost no deterioration in sound quality due to the above-described method.

When there is no line error or frame loss, the gain code correction unit 513 uses the gain code Gain1 (n, j) of the received current frame to the pitch gain gp1 (j) and the algebraic codebook gain gc1 as in the first embodiment. Find (j). On the other hand, if there is a line error or frame loss, the gain code of the current frame cannot be used.
gp1 (n, 0) = α ・ gp1 (n-1,1)
gp1 (n, 1) = α ・ gp1 (n-1,0)
gc1 (n, 0) = β ・ gc1 (n-1,1)
gc1 (n, 1) = β ・ gc1 (n-1,0)
Attenuate the gain of the previous subframe stored in step (a) to obtain and output the pitch gain gp1 (n, j) and the algebraic codebook gain gc1 (n, j). Here, α and β are constants of 1 or less.

  The algebraic code correction unit 514 outputs the dequantized value cbi (j) of the received algebraic code Cb1 (n, j) when there is no line error or frame loss. In addition, when there is a line error or frame loss, the dequantized value of the algebraic code of the last received good frame stored is output.

・ Additional notes
(Supplementary note 1) In a speech code conversion method for converting a speech code obtained by encoding using the first speech encoding method into a speech code of the second speech encoding method,
From the audio code by the first audio encoding method, a plurality of code components necessary to reproduce the audio signal are separated,
Dequantize the sign of each component and output the dequantized value,
Quantize the dequantized value of the code component other than the algebraic code and convert it to the code component of the speech code of the second speech coding scheme,
Play audio from each dequantized value,
Dequantizing each code component of the second speech coding scheme to obtain a dequantized value of the second speech coding scheme;
Generating a target signal using the reproduced speech and each inverse quantization value of the second speech encoding method;
Using the target signal to obtain the algebraic code of the second speech coding scheme,
Outputting each code component of the second speech coding method as a speech code;
A speech code conversion method characterized by the above.
(Appendix 2) Detects the presence or absence of transmission line errors,
If the transmission path error does not occur, the separated code component is used, and if the transmission path error has occurred, the dequantized value is output using the past normal code component. The speech code conversion method according to appendix 1.
(Supplementary Note 3) A first speech code obtained by encoding a speech signal with an LSP code, a pitch lag code, an algebraic code, and a gain code based on the first speech coding method is used as a second speech code based on the second speech coding method. In the voice code conversion method for converting to
The LSP code, pitch lag code, and gain code of the first speech code are inversely quantized, and the inverse quantized values are quantized by the second speech encoding method to obtain the LSP code, pitch lag code, and gain code of the second speech code. ,
A pitch periodic synthesized signal is generated using the dequantized values of the LSP code, pitch lag code, and gain code of the second audio encoding method, and the audio signal is reproduced from the first audio code, and the reproduced audio signal is reproduced. And a difference signal between the pitch periodicity synthesis signal and the target signal,
An algebraic composite signal is generated using an arbitrary algebraic code in the second speech coding system and an inverse quantization value of the LSP code constituting the second speech code;
Obtaining an algebraic code in the second speech coding scheme that minimizes the difference between the target signal and the algebraic synthesized signal;
Outputting the LSP code, pitch lag code, algebraic code, and gain code in the second speech encoding method;
A speech code conversion method characterized by the above.
(Supplementary note 4) A signal obtained by multiplying an adaptive codebook output signal corresponding to the inverse quantized value of the pitch lag code of the second speech coding scheme by a gain corresponding to the gain code of the second speech coding scheme Is input to the LPC synthesis filter based on the inverse quantization value of the LSP code of the second speech coding system, and the output signal is the pitch periodic synthesis signal,
The speech code conversion method according to supplementary note 3, characterized by:
(Supplementary Note 5) An algebraic codebook output signal corresponding to the arbitrary algebraic code of the second speech coding scheme is input to an LPC synthesis filter based on an inverse quantization value of the LSP code of the second speech coding scheme, The output signal is the algebraic composite signal,
The speech code conversion method according to supplementary note 3, characterized by:
(Supplementary Note 6) The gain code of the first speech encoding method is a combination of pitch gain and algebraic codebook gain, and among the inverse quantized values obtained by dequantizing the gain code A pitch gain code of the second speech code is obtained by quantizing the pitch gain inverse quantization value by the second speech coding method;
The speech code conversion method according to supplementary note 3, characterized by:
(Supplementary note 7) An algebraic codebook output signal corresponding to the obtained algebraic code of the second speech coding scheme is input to an LPC synthesis filter based on an inverse quantization value of the LSP code of the second speech coding scheme,
Obtain the algebraic codebook gain from the output signal and the target signal,
Quantizing the algebraic codebook gain to obtain an algebraic codebook gain based on the second speech coding scheme;
The speech code conversion method according to appendix 6, wherein:
(Supplementary Note 8) The gain code of the first speech encoding method is a combination of pitch gain and algebraic codebook gain, and the pitch gain dequantized value obtained by dequantizing the gain code And the algebraic codebook gain dequantized values are respectively quantized by the second speech coding method to obtain the pitch gain code and the algebraic codebook gain code of the second speech code.
The speech code conversion method according to supplementary note 3, characterized by:
(Supplementary note 9) The first speech code obtained by encoding the speech signal with the LSP code, the pitch lag code, the algebraic code, the pitch gain code, and the algebraic codebook gain code based on the first speech coding method is used as the second speech coding method. In the speech code conversion method for converting to the second speech code based on
Each code constituting the first speech code is inversely quantized, and among the inversely quantized values, the LSP code and the inverse quantized value of the pitch lag code are quantized by the second speech coding method, and the LSP code and pitch lag of the second speech code Find the sign
Obtain the inverse quantization value of the pitch gain code of the second speech code by interpolation using the inverse quantization value of the pitch gain code of the first speech code,
A pitch periodic composite signal is generated using the LSP code of the second speech code, the pitch lag code, and the inverse quantization value of the pitch gain, and the speech signal is reproduced from the first speech code, and the reproduced speech signal and Generating a difference signal of the pitch periodic composite signal as a target signal;
An algebraic synthesized signal is generated using an arbitrary algebraic code in the second speech coding system and an inverse quantization value of the LSP code of the second speech code;
Obtaining an algebraic code in the second speech coding scheme that minimizes the difference between the target signal and the algebraic synthesized signal;
The second speech code combining the pitch gain and the algebraic codebook gain by the second speech coding method using the LSP code of the second speech code, the inverse quantization value of the pitch lag code, the obtained algebraic code and the target signal. Find the gain sign of
Outputting an LSP code, a pitch lag code, an algebraic code, and a gain code in the obtained second speech encoding method;
A speech code conversion method characterized by the above.
(Supplementary Note 10) In a speech code conversion device that converts a speech code obtained by encoding using the first speech encoding method into a speech code of the second speech encoding method,
Code separating means for separating a plurality of code components necessary for reproducing a sound signal from a sound code by the first sound coding method;
An inverse quantization unit that inversely quantizes the sign of each component and outputs an inverse quantized value;
A quantization unit that quantizes a dequantized value of a code component other than the algebraic code output from each dequantization unit and converts the quantized value into a code component of a speech code of a second speech coding scheme;
An audio reproduction unit for reproducing audio from each dequantized value;
Inverse quantization means for inversely quantizing each code component of the second speech encoding method to obtain an inverse quantization value of the second speech encoding method;
Target generation means for generating a target signal using the reproduced sound output from the sound reproduction unit and each dequantized value output from the dequantization means of the second speech encoding method;
An algebraic code obtaining unit for obtaining an algebraic code of the second speech coding method using the target signal;
Code multiplexing means for outputting each code component of the second speech coding method as a speech code;
A speech code conversion device comprising:
(Additional remark 11) The 1st audio | voice code which encoded the audio | voice signal with the LSP code, the pitch lag code, the algebraic code, and the gain code based on the 1st audio | voice coding system is used as the 2nd audio | voice code based on the 2nd audio | voice coding system In the speech code conversion device for converting to
The LSP code, pitch lag code, and gain code of the first speech code are dequantized, and these dequantized values are quantized by the second speech coding method and converted to the LSP code, pitch lag code, and gain code of the second speech code. Conversion part,
An audio reproduction unit for reproducing an audio signal from the first audio code;
A pitch periodicity synthesized signal is generated using the dequantized values of the LSP code, pitch lag code, and gain code of the second speech coding method, and the speech signal reproduced by the speech playback unit and the pitch periodicity synthesized signal A target signal generator for generating a difference signal as a target signal;
An algebraic composite signal is generated using an arbitrary algebraic code in the second speech coding system and an inverse quantization value of the LSP code of the second speech code, and a difference between the target signal and the algebraic composite signal is minimized. An algebraic code acquisition unit for obtaining an algebraic code in the second speech encoding method,
A code multiplexing unit that multiplexes and outputs the LSP code, pitch lag code, algebraic code, and gain code in the obtained second speech coding method;
A speech code conversion device comprising:
(Supplementary Note 12) The target signal generator
An adaptive codebook for generating a periodic excitation signal according to an inverse quantization value of the pitch lag code of the second speech coding system;
A gain multiplier that multiplies the adaptive codebook output signal by a gain according to the gain code of the second speech encoding method;
An LPC synthesis filter that is created based on the dequantized value of the LSP code of the second speech coding system, and that receives the output signal of the gain multiplier and outputs the pitch periodic synthesis signal,
Means for outputting a difference signal between the audio signal reproduced by the audio reproduction unit and the pitch periodic synthetic signal as a target signal;
The speech code conversion device according to appendix 11, comprising:
(Supplementary Note 13) The algebraic code acquisition unit includes:
An algebraic codebook that outputs a noisy sound source signal according to an arbitrary algebraic code of the second speech coding scheme;
An LPC synthesis filter that is created based on the dequantized value of the LSP code of the second speech coding system, and that receives the algebraic codebook output signal and outputs the algebraic synthesized signal,
Means for obtaining an algebraic code in the second speech coding system in which a difference between the target signal and the algebraic synthesized signal is minimized;
The speech code converter according to appendix 11, characterized by comprising:
(Supplementary note 14) If the gain code of the first speech encoding method is encoded by combining a pitch gain and an algebraic codebook gain,
An inverse quantization unit for inversely quantizing the gain code to generate a pitch gain inverse quantization value and an algebraic codebook gain inverse quantization value;
Means for quantizing a pitch gain dequantized value among the dequantized values by a second speech coding method and converting the quantized value into a pitch gain code of a second speech code;
The speech code conversion apparatus according to claim 11, comprising:
(Supplementary note 15) The speech code converter further includes:
An LPC synthesis filter created based on the inverse quantization value of the LSP code of the second speech encoding method,
An algebraic codebook gain determining unit that determines an algebraic codebook gain from an output signal when the algebraic codebook output signal corresponding to the obtained algebraic code is input to the LPC synthesis filter and the target signal, and the algebraic codebook gain An algebraic codebook gain code generator for generating an algebraic codebook gain code based on the second speech coding method by quantizing
15. The speech code conversion device according to supplementary note 14, characterized by comprising:
(Supplementary note 16) If the gain code of the first speech encoding method is encoded by combining a pitch gain and an algebraic codebook gain,
An inverse quantization unit for inversely quantizing the gain code to generate a pitch gain inverse quantization value and an algebraic codebook gain inverse quantization value;
The pitch gain inverse quantization value and the algebraic codebook gain inverse quantization value obtained by the inverse quantization are respectively quantized by the second speech coding method and converted into the pitch gain code and the algebraic codebook gain of the second speech code. means,
The speech code conversion apparatus according to claim 11, comprising:
(Additional remark 17) The said audio | voice reproduction | regeneration part reproduces | regenerates an audio | voice signal using the dequantization value of the LSP code of the 1st audio | voice code, the pitch lag code, and the gain code which were dequantized by the said conversion part. The speech code conversion device according to claim 11.
(Supplementary Note 18) A first speech code obtained by coding a speech signal with an LSP code, a pitch lag code, an algebraic code, a pitch gain code, and an algebraic codebook gain code based on the first speech coding method In the speech code conversion device for converting to the second speech code based on
Each code constituting the first speech code is inversely quantized, and among the inversely quantized values, the LSP code and the inverse quantized value of the pitch lag code are quantized by the second speech coding method, and the LSP code and pitch lag of the second speech code A conversion unit for converting into a code,
A pitch gain interpolation unit that generates an inverse quantized value of the pitch gain code of the second speech code by an interpolation process using an inverse quantized value of the pitch gain code of the first speech code;
An audio signal reproduction unit for reproducing an audio signal from the first audio code;
A pitch periodic composite signal is generated using the LSP code of the second speech code, the pitch lag code, and the inverse quantized value of the pitch gain, and the reproduced speech signal output from the speech signal playback unit and the pitch periodic composite signal A target signal generator for generating a difference signal as a target signal;
An algebraic synthesized signal is generated using an arbitrary algebraic code in the second speech coding system and an inverse quantization value of the LSP code of the second speech code, and the difference between the target signal and the algebraic synthesized signal is minimized. An algebraic code acquisition unit for obtaining an algebraic code in the second speech encoding method;
A combination of pitch gain and algebraic codebook gain according to the second speech coding method using the inverse quantization value of the LSP code of the second speech code, the pitch lag code and algebraic code of the second speech code, and the target signal. A gain code acquisition unit for acquiring a gain code of two audio codes;
A code multiplexing unit that multiplexes and outputs the LSP code, pitch lag code, algebraic code, and gain code in the obtained second speech encoding method;
A speech code conversion device comprising:
(Supplementary note 19) The target signal generator
An adaptive codebook for generating a periodic excitation signal according to an inverse quantization value of the pitch lag code of the second speech coding system;
A gain multiplier that multiplies the adaptive codebook output signal by a gain corresponding to the pitch gain code of the second speech encoding method;
An LPC synthesis filter that is created based on the dequantized value of the LSP code of the second speech coding system, and that receives the output signal of the gain multiplier and outputs the pitch periodic synthesis signal,
Means for outputting a difference signal between the audio signal reproduced by the audio reproduction unit and the pitch periodic synthetic signal as a target signal;
The speech code conversion device according to appendix 18, comprising:
(Supplementary note 20) The algebraic code acquisition unit
An algebraic codebook that outputs a noisy sound source signal according to an arbitrary algebraic code of the second speech coding scheme;
An LPC synthesis filter that is created based on the dequantized value of the LSP code of the second speech coding system, and that receives the algebraic codebook output signal and outputs the algebraic synthesized signal,
Means for obtaining an algebraic code in a second speech coding scheme in which a difference between the target signal and the algebraic synthesized signal is minimized;
The speech code converter according to appendix 18, characterized by comprising:

As described above, according to the present invention, the LSP code, pitch lag code, and pitch gain code are code-converted in the quantization parameter area, or the LSP code, pitch lag code, pitch gain code, and algebraic codebook gain code are quantized. Since the code conversion is performed in the parameter area, the parameter conversion can be performed with less analysis error and less deterioration in sound quality as compared with the case where the reproduced sound is subjected to the LPC analysis and the pitch analysis again.
Further, according to the present invention, since the reproduced speech is not subjected to LPC analysis or pitch analysis again, the problem of delay due to code conversion, which has been a problem in the prior art 1, can be solved.

According to the present invention, for the algebraic code and the algebraic codebook gain code, the target signal is created from the reproduced speech and converted so that the error between the target signal and the algebraic synthesized signal is minimized. Even if the configurations of the algebraic codebooks of coding scheme 1 and coding scheme 2 which are problematic in 2 are greatly different, code conversion with little deterioration in sound quality is possible.
Further, according to the present invention, speech code conversion can be performed between the G.729A encoding method and the EVRC encoding method.
Furthermore, according to the present invention, code conversion corresponding to the EVRC rate can be performed.

It is principle explanatory drawing of the speech code converter of this invention. 1 is a configuration diagram of a speech code conversion device according to a first embodiment of the present invention. FIG. It is a frame configuration diagram of G.729A and EVRC. It is pitch gain code conversion explanatory drawing. It is explanatory drawing of the number of samples of the sub-frame in G.729A and EVRC. It is a block diagram of a target production | generation part. It is a block diagram of an algebraic code converter. It is a block diagram of an algebraic codebook gain converter. FIG. 5 is a configuration diagram of a speech code conversion device according to a second embodiment of the present invention. It is conversion explanatory drawing of an algebraic codebook gain code. FIG. 10 is an overall configuration diagram of a speech code conversion device according to a third embodiment. It is a block diagram of the full-rate speech code converter. It is a block diagram of the 1/8 rate speech code converter. FIG. 10 is a configuration diagram of a speech code conversion device according to a fourth embodiment. It is a block diagram of an encoder of ITU-T recommendation G.729A system. It is quantization method explanatory drawing. It is a block diagram of an adaptive codebook. It is an algebraic codebook explanatory diagram of G.729A. It is explanatory drawing of the sample point allocated to each pulse system group. [Fig. 7] Fig. 7 is a block diagram of a G.729A decoder. It is a block diagram of the encoder of EVRC. It is explanatory drawing of the relationship between the frame of EVRC, the LPC analysis window, and the pitch analysis window. It is a principle diagram of a conventional typical speech code conversion method. This is a speech encoding apparatus according to prior art 2. 3 is a detailed speech encoding apparatus of conventional technique 2;

Explanation of symbols

201 rate determination unit 202 full-rate speech code conversion unit 203 half-rate speech code conversion unit 204 1/8 rate speech code conversion unit 401 code separation unit 402 LSP inverse quantization unit 403 LSP quantization unit 404 gain inverse quantization unit 405 Sound source generation unit 406 Gain multiplication unit 407 LPC synthesis filter 408 Algebraic code conversion unit 409 Pitch lag buffer 410 Gain conversion unit 411 Code multiplexing unit

Claims (4)

The first speech code obtained by encoding the speech signal with the LSP code, the pitch lag code, the algebraic code, the pitch gain code, and the algebraic codebook gain code based on the first speech coding method is based on the second speech coding method. In a voice code conversion method for converting into two voice codes,
A full-rate, half-rate, and 1 / 8-rate speech code conversion unit is provided corresponding to the rate of the first speech code,
When the rate of the first speech code is 1/8 rate, the speech code conversion unit for 1/8 rate is
Separating the LSP code and the algebraic codebook gain code included in the speech code by the first speech coding method and dequantizing each to output an inverse quantization value,
Among the inverse quantization values, the inverse quantization value of the LSP code is quantized by the second speech encoding method to obtain the LSP code of the second speech code,
The random signal output from the sound source generator is multiplied by the inverse quantization value of the gain code, the multiplication result is input to an LPC synthesis filter composed of the inverse quantization value of the LSP code, and a target signal is generated,
An algebraic synthesized signal is generated using an arbitrary algebraic code in the second speech coding system and an inverse quantization value of the LSP code of the second speech code;
Obtaining an algebraic code in the second speech coding scheme that minimizes the difference between the target signal and the algebraic synthesized signal;
The second speech code using the inverse quantization value of the LSP code of the second speech code, the inverse quantization value of the pitch lag code obtained by full-rate or half-rate speech code conversion, the obtained algebraic code and the target signal The gain code of the conversion method,
Outputting the LSP code, the algebraic code, the gain code and the pitch lag code in the obtained second speech encoding method;
A speech code conversion method characterized by the above. When the rate of the first speech code is full rate, the speech code conversion unit for full rate is
Each code constituting the first speech code is inversely quantized, and among the inversely quantized values, the LSP code and the inverse quantized value of the pitch lag code are quantized by the second speech coding method, and the LSP code and pitch lag of the second speech code Find the sign
Obtain the inverse quantization value of the pitch gain code of the second speech code by interpolation using the inverse quantization value of the pitch gain code of the first speech code,
A pitch periodic composite signal is generated using the LSP code of the second speech code, the pitch lag code, and the inverse quantization value of the pitch gain, and the speech signal is reproduced from the first speech code, and the reproduced speech signal and Generating a difference signal of the pitch periodic composite signal as a target signal;
An algebraic synthesized signal is generated using an arbitrary algebraic code in the second speech coding system and an inverse quantization value of the LSP code of the second speech code;
Obtaining an algebraic code in the second speech coding scheme that minimizes the difference between the target signal and the algebraic synthesized signal;
Using the LSP code of the second speech code, the inverse quantization value of the pitch lag code, the obtained algebraic code and the target signal, a gain code of the second speech coding method is obtained,
Outputting an LSP code, a pitch lag code, an algebraic code, and a gain code in the obtained second speech encoding method;
The speech code conversion method according to claim 1, wherein: The first speech code obtained by encoding the speech signal with the LSP code, the pitch lag code, the algebraic code, the pitch gain code, and the algebraic codebook gain code based on the first speech coding method is based on the second speech coding method. In a voice code conversion device that converts two voice codes,
Corresponding to the rate of the first speech code, it is equipped with a speech code conversion unit for full rate, half rate, 1/8 rate, the speech code conversion unit for 1/8 rate,
An inverse quantization unit that separates an LSP code and an algebraic codebook gain code included in the speech code according to the first speech coding scheme and dequantizes each to output an inverse quantization value;
An LSP quantization unit that quantizes the inverse quantization value of the LSP code among the inverse quantization values by the second speech encoding method and outputs the LSP code of the second speech code;
A target generation unit that generates a target signal by multiplying a random signal output from a sound source generation unit by an inverse quantization value of the gain code, and inputs the multiplication result to an LPC synthesis filter including an inverse quantization value of an LSP code ,
An algebraic synthesized signal is generated using an arbitrary algebraic code in the second speech coding system and an inverse quantization value of the LSP code of the second speech code, and the difference between the target signal and the algebraic synthesized signal is minimized. An algebraic code acquisition unit for obtaining an algebraic code in the second speech encoding method;
The second speech code using the inverse quantization value of the LSP code of the second speech code, the inverse quantization value of the pitch lag code obtained by full-rate or half-rate speech code conversion, the obtained algebraic code and the target signal Gain code acquisition unit for obtaining the gain code of the conversion method,
A code multiplexing unit that multiplexes and outputs the LSP code, the algebraic code, the gain code, and the pitch lag code in the obtained second speech encoding method;
A speech code conversion device comprising: The full-rate speech code converter is
Each code constituting the first speech code is inversely quantized, and among the inversely quantized values, the LSP code and the inverse quantized value of the pitch lag code are quantized by the second speech coding method, and the LSP code and pitch lag of the second speech code An inverse quantization unit for obtaining a code,
A pitch gain interpolation unit that obtains an inverse quantization value of the pitch gain code of the second speech code by an interpolation process using an inverse quantization value of the pitch gain code of the first speech code;
A pitch periodic composite signal is generated using the LSP code of the second speech code, the pitch lag code, and the inverse quantization value of the pitch gain, and the speech signal is reproduced from the first speech code, and the reproduced speech signal and A target generator for generating a difference signal of the pitch periodic composite signal as a target signal;
An algebraic synthesized signal is generated using an arbitrary algebraic code in the second speech coding system and an inverse quantization value of the LSP code of the second speech code, and the difference between the target signal and the algebraic synthesized signal is minimized. An algebraic code acquisition unit for obtaining an algebraic code in the second speech encoding method;
A gain code obtaining unit for obtaining a gain code of a second speech coding method using the LSP code of the second speech code, the inverse quantization value of the pitch lag code, the obtained algebraic code and the target signal;
A code multiplexing unit that multiplexes and outputs the LSP code, pitch lag code, algebraic code, and gain code in the obtained second speech encoding method;
4. The speech code conversion apparatus according to claim 3, further comprising:

Images