EP0855699B1

EP0855699B1 - Multipulse-excited speech coder/decoder

Info

Publication number: EP0855699B1
Application number: EP98101335A
Authority: EP
Inventors: Toshiyuki Nomura
Original assignee: NEC Corp
Current assignee: NEC Corp
Priority date: 1997-01-27
Filing date: 1998-01-27
Publication date: 2004-04-28
Anticipated expiration: 2018-01-27
Also published as: DE69823398T2; JP3329216B2; JPH10207496A; DE69823398D1; EP0855699A2; CA2228183A1; CA2228183C; EP0855699A3

Description

BACKGROUND OF THE INVENTION

The present invention relates to a speech coder/decoder for high quality coding speech signal with designated parameters and corresponding speech coding/decoding methods according to the respective preamble of Claims 1, 5, 8 and 10.
As a usual controllable bit rate speech coder/decoder, a CDMA (Code Division Multiple Access) system is well known in the art. This system is disclosed in, for instance, "Enhanced Variable Rate Coded Speech Service Option 3 for Wideband Spread Spectrum Digital Systems", Standardization Recommendation Specifications, TIA/EIA/IS-127, committee TR45.5 published on 1.1.1997 by the Telecommunications Industry Association (Literature 1).
In this system, CELP (code excited linear prediction) coding system control parameters are set from a table, which is produced in advance from results of bit rate determination on the basis of input signal features, and the input signal is coded on the basis of the control parameters set in this way. This system also has a function of forcibly setting a bit rate on the basis of an external signal.
This type of speech coder/decoder will now be briefly described with reference to Fig. 11. In the illustrated speech coder/decoder, the bit rate is controlled on the basis of an external signal.
The illustrated speech coder/decoder comprises a speech coder and a speech decoder. The speech coder and speech decoder include respective coding parameter controllers 51 and 55. In the speech coder, a bit rate is given to the coding parameter controller 51. The coding parameter controller 51 selects control parameters corresponding to the given bit rate with reference to a table (not shown, but for instance a ROM (read only memory) with bit rate addresses), in which a plurality of control parameters for controlling the operation of a CELP coder 52 are stored, and provides the selected control parameters to the CELP coder 52. The control parameters are sub-frame length as a unit of excitation signal coding in CELP coding, and bit distribution.
An input signal (i.e., input speech signal) is supplied to a CELP coder 52. The CELP coder 52 computes linear prediction coefficients, which represent a spectral envelope characteristic of the input signal, by linear prediction analysis thereof for each predetermined frame. The CELP coder 52 also generates an excitation signal by driving a linear prediction synthesis filter corresponding to the spectral envelope characteristic, and codes the excitation signal on the basis of the bit distribution. The excitation signal is coded for each of a plurality of sub-frames, into which each frame is divided.
The excitation signal noted above is constituted by a periodic component representing the pitch period of the input signal, a residue signal, and gains of these components. The periodic component representing the pitch period of the input signal, is expressed as an adaptive codevector stored in a codebook called adaptive codebook. The residue component is expressed as a multi-pulse signal, which is disclosed in, for instance, J-P. Adoul et al, "Fast CELP Coding Based on Algebraic Coders" , Proc. ICASSP, pp. 1957-1960, 1987 (Literature 2). The excitation signal is generated by weight imparting the adaptive codevector and the multi-pulse signal by gain data stored in a gain codebook and adding together the results of the weight imparting. A reproduced signal can be synthesized by driving the linear prediction synthesis filter on the basis of the excitation signal.
The selection of the adaptive codevector, multi-pulse signal and gain is controlled such as to minimize error power as a result of acoustical weight imparting of an error signal, which represents an error between the reproduced signal and the input signal. The CELP coder 52 outputs indexes corresponding to the adaptive codevector, multi-pulse signal and gain, and an index representing the linear prediction coefficients, to a multiplexer 53.
The multiplexer 53 provides a bit stream which is obtained by converting the indexes corresponding to the adaptive codevector, multi-pulse signal, gain index and linear prediction coefficients for each frame. Data representing the bit rate is stored in a bit stream header.
In the speech decoder, a multiplexer 54 receives the bit stream, extracts bit stream header data representing the bit rate, and provides the extracted bit rate data to the coding parameter controller 55. Then, the multiplexer 54 extracts the indexes corresponding to the adaptive codevector, multi-pulse signal, gain and linear prediction coefficients from the bit stream for each frame, and provides the extracted data to a CELP decoder 56.
The coding parameter controller 55 executes a similar process to that in the coding parameter controller 51, then selects the control parameters on the basis of the supplied bit rate data, and provides the selected control parameters to the CELP decoder 56.
The CELP decoder 56 executes a decoding process using the indexes corresponding to the adaptive codevector, multi-pulse signal, gain and linear prediction coefficients as well as the sub-frame length and bit rate data. The excitation signal is obtained by weight imparting the adaptive codevector and multi-pulse signal with gain data held in the gain codebook and adding together the results of the weight imparting. In the CELP decoder 56, the reproduced signal is obtained by driving the linear prediction synthesis filter on the basis of the excitation signal.
As shown above, in the CELP coding system the bit rate is controlled by controlling the sub-frame length as a unit of excitation signal coding and the bit distribution.
In the prior art speech coder/decoder, however, the frame length as a unit of coding is fixed. Therefore, it is impossible to control coding delay, which is defined as time from the instant when a first input signal sample is supplied till the instant of start of the coding.
In addition, in the prior art coder/decoder it is necessary to provide in advance parameters which are necessary for generating the multi-pulse signal. Therefore, the system can serve its function only when a predetermined bit rate is given.
ABREU-SERNANDEZ V ET AL: 'A variable rate multipulse speech coder for CDMA cellular systems' WIRELESS PERSONAL COMMUNICATIONS, 1995, KLUWER ACADEMIC PUBLISHERS, NETHERLANDS, vol. 2, no. 3, pages 255-263 describes a variable rate speech coder with a variable bit rate, wherein the bit rate selection can be network-controlled and source-controlled, and a fixed frame length, and with a fixed frame length of 32 msec.
TAUMI S ET AL: 'LOW-DELAY CELP WITH MULTI-PULSE VQ AND FAST SEARCH FOR GSM EFR' ICASSP'96: IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, ATLANTA, GA, USA, vol. 1, 7 - 10 May 1996, pages 562-565 describes a speech coder with a 10 msec frame length, employing a novel multi-pulse excitation signal quantization method, where the pulse amplitudes are vector-quantized.
WO 97 15983 A describes a method and an apparatus which allow coding, manipulation and decoding of audio signals independently of the specific signal content.
EP 0 465 057 A1 describes an improved digital communication system which is improved for use with a wide-band signal by modifying the noise weighting filter.

SUMMARY OF THE INVENTION

It is the object of the present invention to provide a speech coder/decoder capable of controlling not only the coding delay but also the coding delay and computed effort.
This object is achieved with the combination of features of independent claims 1, 5, 8, 10, respectively.
In the present invention, the speech coder comprises a coding parameter control circuit for generating control parameters, i.e., frame length, sub-frame length and bit distribution that are necessary for the coding, from given bit rate and coding delay data. The input speech signal is divided into frames on the basis of the given frame length. A multi-pulse signal coding parameter setting circuit sets parameters, which are necessary for generating a multi-pulse signal from the given bit rate and coding delay.
Since the coding parameter control circuits generates the frame length, sub-frame length and.bit distribution data, and the input speech signal is divided into frames on the basis of the generated frame length, it is possible to vary the frame length which is a unit of processing for the coding. It is thus possible to control the coding delay in addition to the bit rate.
Since the multi-pulse signal coding parameter setting circuit sets parameters necessary for the multi-pulse signal generation, it is possible to increase the bit rate range. That is, it is not necessary to set a bit rate in advance.
Other objects and features will be clarified from the following description with reference to attached drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a block diagram of a speech coder/decoder according to a first embodiment of the present invention;
Fig. 2 is a block diagram for explaining the CELP coding circuit shown in Fig. 1;
Fig. 3 is a block diagram for explaining the CELP decoding circuit shown in Fig. 1;
Fig. 4 is a block diagram of a speech coder/decoder according to a second embodiment of the present invention;
Fig. 5 is a block diagram for explaining the CELP coding circuit shown in Fig. 4;
Fig. 6 is a block diagram for explaining the CELP decoding circuit shown in Fig. 4;
Fig. 7 is a block diagram of a speech coder/decoder according to a third embodiment of the present invention;
Fig. 8 is a block diagram for explaining the CELP coding circuit shown in Fig. 7;
Fig. 9 is a block diagram of a speech coder/decoder according to a fourth embodiment of the present invention;
Fig. 10 is a block diagram for explaining the CELP coding circuit shown in Fig. 9; and
Fig. 11 is a block diagram of a prior art speech coder/decoder.

PREFERRED EMBODIMENTS OF THE INVENTION

Referring to Fig. 1, a speech coder/decoder is shown, which comprises a speech coder and a speech decoder. The speech coder includes a coding parameter control circuit 11, a CELP coding circuit 12 and a multiplexer 13. The speech decoder includes a demultiplexer 14, a coding parameter control circuit 15 and a CELP decoding circuit 16.
In the speech coder, bit rate and coding delay are given as control data to the coding parameter control circuit 11. The coding parameter control circuit 11 calculates a frame length by subtracting an advance read length, which is necessary for an analytic processing in CELP coding, from the given bit rate and coding delay. For example, in a case where the coding delay is 25 ms and the advance read length of the linear prediction analysis is 5 ms, the frame length is 20 ms.
The coding parameter control circuit 11 selects, on the basis of the given bit rate, control parameters from a table, in which a plurality of control parameters for controlling the operation of the CELP coding circuit 12 are set on the basis of calculated frame length, and provides the selected control parameters to the CELP coding circuit 12. The selected control parameters are frame length, sub-frame length (of 5 ms, for instance) and bit distribution. The CELP coding circuit 12 codes the input signal (input speech signal) on the basis of frame length, sub-frame length and bit distribution that have been set.
The operation of the CELP coding circuit 12 will now be described by having reference also to Fig. 2.
The frame length F that has been set in the coding parameter control circuit 11, is supplied through an input terminal 213 to a frame dividing circuit 201 and a linear prediction coefficient quantizing circuit 204.
The sub-frame length S that has also been set in the coding parameter control circuit 11, is supplied through an input terminal 214 to a sub-frame dividing circuit 202, a linear prediction analysis circuit 203, the linear prediction coefficient quantizing circuit 204, an acoustical weight imparting signal generating circuit 205, an acoustical weight imparted reproduced signal generating circuit 206, a target signal generating circuit 208, an adaptive codebook retrieving circuit 209, a multi-pulse retrieving circuit 210 and a gain retrieving circuit 211.
The bit distribution to the parameters having been set in the coding parameter control circuit 11, is supplied through an input terminal 215 to the linear prediction coefficient quantizing circuit 204, adaptive codebook retrieving circuit 209, multi-pulse retrieving circuit 210 and gain retrieving circuit 211.
The frame dividing circuit 201 divides the input signal on the basis of the frame length F having been set, and provides each frame of input signal to the sub-frame dividing circuit 202.
The sub-frame dividing circuit 202 divides each frame on the basis of the sub-frame length S having been set, and provides each sub-frame of input signal to the linear prediction analysis circuit 203 and acoustical weight imparting signal providing circuit 205.
The linear prediction analysis circuit 203 executes linear prediction analysis of signal (sub-frame signal) provided from the sub-frame dividing circuit 202 on the basis of the sub-frame length S having been set for each sub-frame, and provides linear prediction coefficients a(i) (i = 1, ..., Np) to the linear prediction coefficient quantizing circuit 204, acoustical weight imparting signal providing circuit 205, acoustical weight imparted reproduced signal generating circuit 206, adaptive ccdebook retrieving circuit 209 and multi-pulse retrieving circuit 210. Np is the degree number of the linear prediction analysis, for instance 10. The linear prediction analysis may be a self-correlation process or a covariance process, and is detailed in Furui, "Digital Speech Processing", Tokai University Publishing Association (Literature 3).
The linear prediction coefficient quantizing circuit 204 executes collective quantization of the linear prediction coefficients obtained for the individual sub-frames on the basis of the frame length F and sub-frame length S having been set for each frame. In order to reduce the bit rate, this quantization is executed for only the last sub-frame in the frame and using interpolated values of the quantized values of the pertinent and immediately preceding frames as the quantized values of the other sub-frames. This quantization and interpolation are executed after conversion of the linear prediction coefficient into corresponding line spectrum pair (LSP). The conversion of the linear prediction coefficient into LSP is described in, for instance, Sugamura et al, "Speech Data Compression in Linear Spectrum Pair (LSP) Speech Analysis Synthesis Systems", The Transactions of Institute of Electronics and Communication Engineers of Japan, J64-A, pp. 599-606, 1981 (Literature 4). The LSP quantization may be executed in a well-known manner; for instance, it is disclosed in Japanese Laid-Open Patent Publication No. 4-171500 (Literature 5), and it is not described here. The linear prediction coefficient quantizing circuit 204 converts the quantized LSP into corresponding linear prediction coefficients, and provides the result as quantized linear prediction coefficient a'(i) (i = 1, ..., Np) to the acoustical weight imparting signal providing circuit 205, acoustical weight imparted reproduced signal generating circuit 206, an adaptive codebook retrieving circuit 209 and multi-pulse retrieving circuit 210.
An index representing the quantized LSP is supplied through an output terminal 216 to the multiplexer 13. Linear prediction synthesis filter Hs(z) is expressed by formula (1).
In the acoustical weight imparting signal generating circuit 205, an acoustical weight imparting filter Hw(z) expressed by formula (2) is formed using the linear prediction coefficients, and is driven by sub-frame input signal to generate an acoustical weight imparted signal. This acoustical weight imparted signal is provided to the target signal generating circuit 208. where R1 and R2 are weight imparting coefficients to control the extent of the acoustical weight imparting and, for instance, R1 = 0.6 and R2 = 0.9.
The acoustical weight imparted reproduced signal generating circuit 206 drives the linear prediction synthesis filter and the acoustical weight imparting synthesis filter of the preceding frame with the excitation signal of the preceding sub-frame which is obtained through a sub-frame buffer 207, and provides data representing the states of the two filters after the driving to the target signal generating circuit 208.
The target signal generating circuit 208 receives the data representing the states of the linear prediction synthesis filter and acoustical weight imparting filter from the acoustical weight imparting reproduced signal generating circuit 206, generates a zero input response of a filter which is constituted by the two filters connected in cascade, subtracts the zero input response thus generated from the acoustical weight imparted signal, and provides the resultant difference as the target signal to the adaptive codebook retrieving circuit 209 and multi-pulse retrieving circuit 210 as well as to a gain retrieving circuit 211.
The adaptive codebook retrieving circuit 209 updates a codebook, called adaptive codebook and holding past excitation signals, on the basis of the excitation signal of the immediately preceding sub-frame that is obtained through the sub-frame buffer 207, and then selects an adaptive codevector corresponding to pitch d from the adaptive codebook. When the pitch d is shorter than the sub-frame length, an adaptive codevector is formed by repeatedly connecting excitation signal segments each corresponding to delay d, separated one after another from past excitation signal stored in the adaptive codebook, until reaching of the sub-frame length. The reproduced signal SAd(n) is formed by driving the linear prediction synthesis filter and acoustical weight imparting filter in zero states thereof with the adaptive codevector Ad(n) thus formed, and selects pitch d which minimizes the error Ed between the target signal X(n) and the reproduced signal SAd(n), given by formula (3). where L is the sub-frame length set by the coding parameter control circuit 11. The adaptive codebook retrieving circuit 209 further provides the selected pitch d through the output terminal 216 to the multiplexer 13, and also provides the selected adaptive codevector Ad(n) and the reproduced signal SAd(n) thereof to the gain retrieving circuit 211. The adaptive codebook retrieving circuit 209 provides the reproduced signal SAd(n) to the gain retrieving circuit 211 and provides the reproduced signal SAd(n) to the multi-pulse retrieving circuit 210.
The multi-pulse retrieving circuit 210 forms a multi-pulse signal constituted by a plurality of non-zero pulses. The position of each pulse is selected from a plurality of pulse position candidates predetermined for each pulse. Each pulse is a polarity pulse. For example, in 8-kHz sanpling with a sub-frame length of 5 ms (i.e., with a sample number N of 40), the multi-pulse excitation signal is constituted by P (for instance 5) pulses. The position of each of the P pulses is selected from M(p) (p = 1, ..., P-1, for instance 8) pulse position candidates. The multi-pulse retrieving circuit 210 is holding a plurality of combinations of pulse number P and M(p) pulse position candidates, and selects a combination of pulse number P and M(p) pulse position candidates on the basis of a bit distribution designated by a coding parameter control circuit 11. The multi-pulse retrieving circuit 210 also forms multi-pulse signal Cj(n) by using the selected pulse number P (equal to the number of channels) and M pulse position candidates of each channel, and selects a multi-pulse signal Cj(n) which minimizes formula (4). where X'(n) is a subtracted signal of the reproduced signal SA(n) of the adaptive codevector from the target signal X(n) and given by formula (5).
Formula (4) can be minimized with reducing the computational effort extent, for instance by using method as described in Japanese Patent Application No. 7-318071 (Literature 6). The multi-pulse retrieving circuit 210 provides the selected multi-pulse signal Cj(n) and reproduced signal SCj(n) thereof to the gain retrieving circuit 211, and provides corresponding index j through the output terminal 216 to the multiplexer 13.
The gain retrieving circuit 211 quantizes the gains GA and GC by using the reproduced signal SAd(n) of the adaptive codevector, reproduced signal SCj(n) of the multi-pulse signal and target signal X(n) such as to minimize formula (6).
The gain retrieving circuit 211 further forms an excitation signal by using the quantized gain, adaptive codevector and multi-pulse signal, provides the excitation signal thus formed through the sub-frame buffer 207 to the acoustical weight imparted reproduced signal generating circuit 206 and adaptive codebook retrieving circuit 209, and an index corresponding to the gain through the output terminal 216 to the multiplexer 13.
Referring now back to Fig. 1, the multiplexer 13 provides a bit stream obtained by conversion from the indexes representing the quantized LSP, pitch, multi-pulse signal and quantized gains for each signal. The bit rate and coding delay data are provided in a header of the bit stream.
In the speech decoder, the bit stream is supplied to the demultiplexer 14. The demultiplexer 14 provides the bit rate and coding delay data present in the bit stream header to the coding parameter control circuit 15, and then it extracts the indexes of the quantized LSP, pitch, multi-pulse signal and quantized gains from the bit stream for each frame, and provides them to the CELP decoding circuit 16.
The coding parameter control circuit 15 executes an operation similar to that in the coder side coding parameter control circuit 11; i.e., it selects control parameters on the basis of the input bit rate and coding delay data, and provides the selected control parameters to the CELP decoding circuit 16.
The operation of the CELP decoding circuit will now be described by having reference also to Fig. 3.
The indexes representing the quantized LSP, pitch, multi-pulse signal and quantized gains, are supplied through an input terminal 227 to a linear prediction coefficient decoding circuit 221, an adaptive codebook decoding circuit 222, a multi-pulse signal decoding circuit 223 and a gain decoding circuit 224.
The frame length data set by the coding parameter control circuit 15 is supplied through an input terminal 228 to the linear prediction coefficient decoding circuit 221 and a frame unifying circuit 226.
The sub-frame length data set by the coding parameter control circuit 15 is supplied through an input terminal 229 to the linear prediction coefficient decoding circuit 221, adaptive codebook decoding circuit 222, multi-pulse signal decoding circuit 223 and gain decoding circuit 224 and also to a reproduced signal synthesizing circuit 225 and the frame unifying circuit 226.
The bit distribution data set by the coding parameter control circuit 15 is supplied through an input terminal 230 to the linear prediction coefficient decoding circuit 221, adaptive codebook decoding circuit 222, multi-pulse signal decoding circuit 223 and gain decoding circuit 224.
The linear prediction coefficient decoding circuit 221 receives the index representing the quantized LSP for each frame, and provides quantized linear prediction coefficient a'(i) (i = 1, ..., Np) restored by decoding for each sub-frame to the reproduced signal synthesizing circuit 225.
The adaptive codebook decoding circuit 222 restores the adaptive codevector by decoding from the pitch data supplied for each sub-frame. The multi-pulse decoding circuit 223 provides the multi-pulse signal restored by decoding from the indexes supplied for each sub-frame to the gain decoder 224.
The gain decoding circuit 224 restores the gains by decoding from the indexes supplied for each sub-frame, forms an excitation signal by using the adaptive codevector, multi-pulse signal and gains, and provides the excitation signal thus formed to the reproduced signal synthesizing circuit 225.
The reproduced signal synthesizing circuit 225 forms a reproduced signal by driving the linear prediction synthesis filter Hs(z) with the excitation signal for each sub-frame, and provides the reproduced signal thus formed to the frame unifying circuit 226. The linear prediction synthesis filter Hs(z) is expressed by formula (1) noted above. The frame unifying circuit 226 connects together successively supplied sub-frame reproduced signals for the frame length, and provides the resultant reproduced signal for each frame to be output at terminal 231.
A different embodiment of the speech coder/decoder according to the present invention will now be described with reference to Fig. 4.
The illustrated coder/decoder comprises a speech coder and a speech decoder. The speech coder includes a coding parameter control circuit 31, a CELP coding circuit 32, a multi-pulse signal coding parameter setting circuit 33 and a multiplexer 13. The speech decoder includes a demultiplexer 14, a coding parameter setting circuit 34, a CELP decoding circuit 35 and a multi-pulse signal coding parameter setting circuit 36.
In the speech coder, the coding parameter control circuit 31 receives the bit rate and coding delay as control data, and calculates the frame length by subtracting advance read length, which is necessary for an analysis process in CELP coding, from the given bit rate and coding delay. On the basis of the calculated frame length, the coding parameter control circuit 31 selects control parameters from a table, in which a plurality of control parameters for controlling the operation of the CELP coding circuit 32 are stored, on the basis of the supplied bit rate, and provides the selected control parameters to the CELP coding circuit 32. The coding parameter control circuit 31 further provides the bit number distributed to the sub-frame length and multi-pulse signal to the multi-pulse signal coding parameter setting circuit 33.

The multi-pulse signal coding parameter setting circuit 33 computes pulse number P, pulse position candidate number M(p) of each pulse and position candidates thereof, necessary for the multi-pulse excitation signal coding, from supplied sub-frame length N and bit number Y of the multi-pulse signal. The pulse position candidates of each pulse are set such that a sequence of 0, 2, 3, ..., N-1 is interleaved with the pulse number P, as disclosed in Literature 2 noted above. For example, in a case where the sub-frame length is set to 40 (i.e., a sample number N of 40) and the bit number Y of the multi-pulse signal is set to 20, the pulse number P is 5 and the pulse position candidate number M(p) is 8. An example of pulse position candidates in this case is shown in Table 1 below.

TABLE 1

PULSE No.	PULSE POSITION CANDIDATES
0	0, 5, 10, 15, 20, 25, 30, 35
1	1, 6, 11, 16, 21, 26, 31, 36
2	2, 7, 12, 17, 22, 27, 32, 37
3	3, 8, 13, 18, 23, 28, 33, 38
4	4, 9, 14, 19, 24, 29, 34, 39

The CELP coding circuit 32 codes the input signal on the basis of the frame length, sub-frame length and bit distribution that are set by the coding parameter control circuit 31, and also the pulse number P, pulse position candidate number M(p) of each pulse and position candidates thereof that are set by the multi-pulse signal coding parameter setting circuit 33.
The operation of the CELP coding circuit 32 will now be described with reference to Fig. 5.
The CELP coding circuit 32 is the same as the CELP coding circuit described before in connection with Fig. 2 except for the operation of the multi-pulse retrieving circuit. For this reason, only the operation of the multi-pulse retrieving circuit 401 will be described.
The multi-pulse retrieving circuit, designated at 401 in Fig. 5, generates the multi-pulse signal Cj(n) on the basis of the pulse number P and M(p) pulse position candidates of each pulse, set by the multi-pulse generation parameter setting circuit 33 and supplied through an input terminal 217, and selects a multi-pulse signal Cj(n) that minimizes formula (4) noted above. As described before, in the minimization of formula (4) the computational effort extent can be reduced by using the manner described in Literature 6.
The multi-pulse retrieving circuit 401 provides the selected multi-pulse signal Cj(n) and reproduced signal SCj(n) thereof to the gain retrieving circuit 211 and also provides corresponding index j through the output terminal 216 to the multiplexer 13. As described before in connection with Fig. 1, the multiplexer 13 provides a bit stream.
Referring back to Fig. 4, in the speech decoder the bit stream is received by the demultiplexer 14. As described before in connection with Fig. 1, the demultiplexer 14 provides the bit rate and coding delay data present in the bit stream header to the coding parameter control circuit 34, then extracts the indexes representing the quantized LSP, pitch and multi-pulse signal from the bit stream for each frame, and provides the extracted indexes to the CELP decoding circuit 35.
The coding parameter setting circuit 34 executes an operation similar to that in the coding parameter control circuit 31, thus selecting the control parameters and providing the same to the CELP decoding circuit 35.
The multi-pulse coding parameter setting circuit 36 executes an operation similar to that in the coding side multi-pulse generation parameter setting circuit 33, thus computing the pulse number representing the multi-pulse excitation signal, pulse position candidate number of each pulse and position candidates thereof, and providing the computed data to the CELP decoding circuit 35.
The operation of the CELP decoding circuit 35 will now be described with reference also to Fig. 6.
The CELP decoding circuit 35 is the same as the CELP decoding circuit described before in connection with Fig. 3 except for the operation of the multi-pulse decoding circuit. For this reason, only the operation of the multi-pulse decoding circuit 402 will be described.
The multi-pulse decoding circuit, designated at 402 in Fig. 6, receives the sub-frame length set by the coding parameter control circuit 34 through the input terminal 229, receives the pulse number, pulse position candidate number of each pulse and position candidates thereof set by the multi-pulse coding parameter setting circuit 36 through an input terminal 232, and restores the multi-pulse signal by decoding from the indexes supplied for each sub-frame.
A further embodiment of the speech code= according to the present invention will now be described with reference to Fig. 7.
The illustrated speech coder includes a coding parameter control circuit 61, a CELP coding circuit 62 and a multiplexer 13. The coding parameter control circuit 61 executes an operation similar to that in the coding parameter control circuit 11 described before in connection with Fig. 1, thus setting the frame length, sub-frame length and bit distribution from the supplied bit rate and coding delay data. The coding parameter control circuit 61 computes permissible multi-pulse signal coding computational effort extent, to which computational effort can be paid for the multi-pulse signal coding, from the supplied computational effort extent data. This computation can be executed by storing in advance data of computational effort extents necessary for the coding of other parameters and subtracting these stored computational effort extents frcm the supplied computational effort extent. The coding parameter control circuit 61 provides frame length, sub-frame length, bit distribution and permissible multi-pulse coding computational effort extent as control parameters to the CELP coding circuit 62.
The CDLP coding circuit 62 codes the input signal on the basis of the supplied frame length, sub-frame length, bit distribution and permissible multi-pulse signal coding computational effort extent data.
The operation of the CELP coding circuit 62 will now be described by having reference also to Fig. 8.
The CELP coding circuit 62 is the same as the CELP coding circuit described before in connection with Fig. 2 except for the operation of the multi-pulse retrieving circuit. For this reason, only the multi-pulse retrieving circuit will be described.
The multi-pulse retrieving circuit, designated at 301 in Fig. 8, executes an operation similar to that in the multi-pulse retrieving circuit 210 described before in connection with Fig. 2, thus selecting a multi-pulse signal Cj(n) that minimizes formula (4) noted above. In this case, the computational effort paid for the coding of the multi-pulse signal, is preliminarily selected such that it does not exceed the permissible multi-pulse coding computational effort extent data supplied through an input terminal 218. This preliminary selection can be realized by selection of a high value of E1 given by formula (9).
The multi-pulse retrieving circuit 301 provides the selected multi-pulse signal Cj(n) and reproduced signal SCj(n) thereof to the gain retrieving circuit 211, and also provides corresponding index j through the output terminal 216 to the multiplexer 13.
A still further embodiment of the speech coder according to the present invention will now be described with reference to Fig. 9.
The illustrated speech coder includes a coding parameter control circuit 71, a multi-pulse generation parameter setting circuit 33, a CELP coding circuit 72 and a multiplexer 13.
The coding parameter control circuit 71 executes an operation similar to that in the coding parameter control circuit 31 described before in connection with Fig. 4, thus setting frame length, sub-frame length and bit distribution from the supplied bit rate and coding delay data. The coding parameter control circuit 71 computes permissible multi-pulse signal coding computational effort extent, which is paid for the coding of multi-pulse signal, from the supplied computational effort extent data. The coding parameter control circuit 71 provides the frame length, sub-frame length, bit distribution and permissible multi-pulse signal coding computational effort extent to the CELP coding circuit 72. The coding parameter control circuit 71 provides sub-frame length and bit number distributed to the multi-pulse signal to the multi-pulse generation parameter setting circuit 33.
The CELP coding circuit 72 codes the input signal on the basis of the frame length, sub-frame length, bit distribution and permissible multi-pulse signal coding computational effort extent set by the coding parameter setting circuit 71 and the pulse number P, pulse position candidate number M(p) of each pulse and position candidates thereof set by the multi-pulse signal generation parameter setting circuit 33.
The operation of the CELP coding circuit 72 will now be described by having reference also to Fig. 10.
The CELP coding circuit 72 is the same as the CELP coding circuit described before in connection with Fig. 5 except for the operation of the multi-pulse retrieving circuit. For this reason, only the operation for the multi-pulse retrieving circuit 501 will be described.
The multi-pulse retrieving circuit, designated at 501 in Fig. 10, executes an operation similar to that in the multi-pulse retrieving circuit 401 described before in connection with Fig. 5, thus selecting a multi-pulse signal Cj(n) that minimizes Formula (4) noted above. In this case, the computational effort paid for the coding of multi-pulse signal, is preliminarily set such that it does not exceed permissible multi-pulse signal coding computational effort extent supplied through an input terminal 218. The multi-pulse retrieving circuit 501 also provides the selected multi-pulse signal Cj(n) and reproduced signal SCj(n) thereof to the gain retrieving circuit 211, and also provide corresponding index j through the output terminal 216 to the multiplexer 13.
As has been described in the foregoing, according to the present invention the frame length as a unit of processing for coding is made variable, permitting generation of parameters necessary for the coding of multi-pulse signal from given bit rate and coding delay data. Thus, it is possible to control not only the bit rate but also the coding delay and computational effort. According to the present invention, it is thus possible to use the same coder/decoder when it is desired to make the coding delay to be as short as possible for a television conference system or the like or when it is desired to make the bit rate to be as low as possible rather than the coding delay for speech mail or like purposes. This permits scale reduction of the coder/decoder.
Changes in construction will occur to those skilled in the art and various apparently different modifications and embodiments may be made without departing from the scope of the present invention. The matter set forth in the foregoing description and accompanying drawings is offered by way of illustration only. It is therefore intended that the foregoing description be regarded as illustrative rather than limiting.

Claims

A speech coder for coding an input speech signal, comprising speech coding means (12; 32; 62; 72) including
a frame divider (201) partitioning an input speech signal into frames with a frame length F,

means (203, 204) generating linear prediction coefficients of the partitioned input speech signal,

means (209, 210; 301; 401; 501) generating a reproduced speech signal by driving a linear prediction synthesis filter on the basis of an input speech signal excitation signal, the linear prediction synthesis filter being supplied with the output from the linear prediction coefficients generating means (203, 204),

means (211) generating, from the output of the reproduced speech signal generating means (209, 210; 301; 401; 501), said input speech signal excitation signal, expressed in the form of a plurality of pulses,
characterized in that
the speech coder comprises a control circuit (11; 31; 61; 71), the input of which are designated control data and the output of which is at least one variable control parameter including the variable frame length F, generated on the basis of the designated control data, and
the generated control parameters are input to the speech coding means (12; 32; 62; 72), in particular the frame length F is input to the frame divider (201).
A speech coder according to claim 1, wherein the control data input to the control circuit (11; 31; 61; 71) comprise a designated bit rate and a coding delay.
A speech coder according to claim 2, wherein the control data input to the control circuit (11; 31; 61; 71) comprise a computational effort extent/ computation amount.
A speech coder according to claim 2 or 3, comprising a parameter setting circuit (33), input of which is given by control parameters output from the control circuit (31; 71), and the output of which are calculated setting parameters which are input to at least one input terminal (217, 218) of the speech coding means (32; 72).
A speech decoder for restoring a reproduced speech signal from received coded speech data, comprising speech decoding means (16; 35) including
means (221) restoring linear prediction coefficients from the received coded speech data,

means (225) restoring a reproduced speech signal by driving a linear prediction synthesis filter on the basis of an excitation signal, the linear prediction synthesis filter being supplied with the output from the linear prediction coefficients restoring means (221),

means (224) generating, from the output of the reproduced speech signal restoring means (225), said excitation signal,

a frame unifying circuit (226) generating the decoded speech signal by connecting together the reproduced speech signal,
characterized in that
the speech decoder comprises a control circuit (15; 34), the input of which are designated control data and the output of which is at least one variable control parameter including the variable frame length F, generated on the basis of the designated control data, and
the generated control parameters are input to the speech decoding means (16; 35), in particular the frame length F is input to the frame unifying circuit (226).
A speech decoder according to claim 5, wherein the control data input to the control circuit (15; 34) comprise a designated bit rate and a coding delay.
A speech decoder according to claim 6, wherein the control data input to the control circuit (15; 34) comprise a computational effort extent/ computation amount.
A speech coding method for coding an input speech signal, comprising the steps of
partitioning the input speech signal into frames with a frame length F,

generating linear prediction coefficients of the partitioned input speech signal,

generating a reproduced speech signal by driving a linear prediction synthesis filter on the basis of an input speech signal excitation signal, the linear prediction synthesis filter being supplied with the generated linear prediction coefficients,

generating, from the generated reproduced speech signal, said input speech signal excitation signal, expressed in the form of a plurality of pulses,

outputting the coded speech data,
characterized by
generating at least one variable control parameter including the variable frame length F on the basis of designated control data, and
using the generated frame length F in the step of partitioning the input speech signal.
A speech coding method according to claim 8, wherein the control data used to calculate the control parameters comprise a designated bit rate and a coding delay.
A speech decoding method for restoring a reproduced speech signal from received coded speech data, comprising the steps of
restoring linear prediction coefficients from the received coded speech data,

restoring a reproduced speech signal by driving a linear prediction synthesis filter on the basis of an excitation signal, the linear prediction synthesis filter being supplied with the restored linear prediction coefficients,

generating, from the reproduced speech signal, said excitation signal,

connecting together the reproduced speech signal,

outputting the decoded speech signal,
characterized by
generating at least one variable control parameter including the variable frame length F, generated on the basis of designated control data, and
using the generated frame length F in the step of connecting together the reproduced speech signal.