SA90110060B1 - Method of organizing excitation pulses in a linear predictive speech (LPC) coder - Google Patents

Abstract

Abstract: The present invention relates to a method for organizing the excitation pulses of a linear predictive coder (LPC) according to the principle of multipulse, i.e. a certain number of such pulses are placed at specified time points and with specified amplitudes. The time points and amplitudes of the predictive parameters (ak) and the predictive residue (dk) signal are determined by the correlation between a speech representative signal (y) and a composed synthesized signal (ý). This method can produce all possible temporal modes of excitation pulses within a given period of time. According to the proposed method, the possible time modes are divided into a number of phase modes (nf), and each phase mode is divided into a number of phases (f). These phases are empty for the first excitation pulse. When this pulse is placed, it loses the specific phase of this pulse and moves on to the next excitation pulses until all the pulses in the frame are placed. ,

Description

, 0 Method of regulating excitation pulses in a linear predictive coder (LPC): Full description BACKGROUND: The present invention relates to a method of regulating excitation pulses in an encoder (LPC) Linear predictive speech coder (LPC) (sha) works according to the principle of multiplexing and this speech encoder can be integrated into a mobile phone equipment for example for the purpose of compressing the JF speech signals sent from that machine. No encoders have been defined Linear predictive speech that operates according to the multiplicity principle aforementioned in US Patent Specification VIVE describing the linear predictive encoding of speech signals; and also from US Specification 774549776 that specifies how predictive parameters and residual signals can be configured Predictive residue signal Sie this speech encoder When an artifical speech signal is generated by a predictive residue signal 0 | linear, a number of predictive coefficients (a) are generated from the original signal which characterizes the artificial speech signal. Thus, through these coefficients, it is possible to form a speech signal that does not contain additional A1 increments, such as those usually found in natural speech, and therefore its conversion becomes unnecessary when transmitting and transmitting speech between a mobile station, for example, and a base station in a mobile radio. As for the bandwidth aoa, it is only appropriate to JB the predictive coefficients instead of the original speech signal which requires a much wider bandwidth. However, it may sometimes be difficult to understand the speech signal that is generated by a receiver and forms an artificial speech signal, due to the lack of agreement, Ye, between the speech pattern of the original signal and the artificial signal that was generated by prediction coefficients. These shortcomings are described in detail in US Patent Specification No. EEVYAYY (Patent Application No. 597714) and can be mitigated and to some extent overcome by introducing so-called excitation pulses (multiple pulses) when creating the Synthetic Speech version. In this case, the pattern of the original speech input is divided into a frame of time periods, within each period of which there is a specified number of 0 pulses with variable amplitudes and time-phase mode in terms of their dependence on the one hand on the prediction coefficients YAR.

x0 (ay) and its dependence on the other hand on the predictive residual (dy) between the speech input pattern and the speech transcript. Each of the impulses is allowed to influence the transcription of the speech pattern until the predictive residual is diminished to a real possible size. And since the generated excitation pulses are Janay with a relatively low stability, Al can be encoded and transmitted in a narrow space, as well as pall coefficients, which of course leads to a significant improvement in the quality of the speech signal that is regenerated. GENERAL DESCRIPTION OF THE INVENTION In the case of the aforementioned known methods, excitation pulses are generated within each time frame of the speech input pattern by weighing the residual signal from the (dy) and feedback and weighing the generated values of the excitation pulses each in a separate predictive filter. Then the output signals of the filters are connected. Then, the number of signal elements is maximized from the associated signal, forming the amplitude and phase coefficients of the excitation pulses. The main benefit of the multiple-pulse arithmetic approach for generating excitation pulses is that it can generate different types of sounds with a small number of pulses (eg A pulses per time frame). The pulse-finding algorithm is lle in determining the position of the pulses in the frame. It is possible to recreate sounds that are not distinct in tone, which usually require randomly placed pulses; And distinct sounds require a more clustered placement of impulses. Z The only shortcoming of the known pulse regulation method is that the coding in its sequential form in determining the pulse modes is complex in terms of both computational and storage operations. Beside that; This method requires a large number of 5 bits per pulse mode within the specified time interval 0. Bits in encoding words that are obtained from Ball compatible spline encoding algorithms are also susceptible to bit errors. A mistake in biting a JES codeword from a transmitter to a receiver can have bad effects for the pulse mode when trying to decode the codeword in the receiver. 0 The invention, Jad, is based on the fact that the number of pulse modes of the excitation pulses ve within the frame of the time period is so large that it is not possible to fail to determine the correct AA position of one or more of the excitation pulses within the frame and obtain A renewed speech signal of acceptable quality subsequent to encoding and transmission. father-in-law

According to the known methods, the correct phase positions of the Jala excitation pulses are calculated one frame and the following frames of the speech signal; The pulses are organized based on the combined processing of the speech signal parameters (the predictive residual, the residual signal, and the excitatory impulse coefficients in the previous frames). to the method of the present invention; Some specific constraints are placed in phase mode when ky is set to 0

Cs pulse mode; By rejecting a specified number of pre-determined phase positions for these pulses, the phase position of the excitation pulse is actually calculated. It is successively after calculating the placement of the first pulse inside the frame; After placing this pulse in the calculated phase position; Rejects said phase position for subsequent pulses within the frame. This rule is best applied to all 0 pulse modes in the frame. The purpose of the invention (all) is to provide a method for identifying and reporting the positions of Jada excitation pulses within one time interval and the next interval of speech input mode for a linear predictive encoder that requires less encoding complexity and smaller bandwidth and which helps to reduce bit error. In the subsequent re-encoding process before transmission, vo and the proposed method can be applied to a speech encoder operating da of the principle of multiple pulses with the correlation of the original speech signal and the impulse response to an artificial signal of a linear prediction encoder LS This method can be applied Lad On the so-called RPE- speech encoder in which many excitation pulses take their position within the time period at the same time. V) Simplified assembly diagram of the speech coding device - the well-known linear predictive coding; Figures (7)-(7c) Temporal diagrams covering some of the signals that occur in the speech coding device according to Figure (1); Figure (1) a schematic diagram explaining The principle on which the invention is based; vo Figures (4a)-(;k) more detailed diagrams to show all the ash of the invention; Figure (0) an assembly diagram showing a part of the speech encoder operating according to the innovative principle; © > Figures (1), (M) Flow diagram of the speech encoder shown in Figure (0) o Detailed description id row 0 is a simplified assembly diagram of a speech encoder-linear predictive encoder (1) a form of the multipulse principle. This encoder is described in detail in a well-known Wide Works Spec 7 (USP Requisition No. 4071714. In addition to the analog-to-digital converter, an analog speech signal occurs (111). Examples include a loudspeaker on the input of the prediction analyzer 0m Computer-Linear Prediction Encoder and Residual Signal Generator which is Lad (Lad). 011) prediction analyzer respectively. The prediction coefficients characterize the artifact signal; while the prediction coefficients dy and the remainder signal the residual signal explains the error between the artifact signal and the original speech signal via the transmitter input. 0 two signals 3 and it works under one From a number of frame periods (VY) the actuator receives the mutual sequential excitation time decided by the frame signal (test frame) to emit a specified number of excitation 0 pulses during each of the said time periods. Each of the mentioned pulses decides the value of the frame MP device. The Counseling Pulse Coefficients are tedious; dalam, and their time mode is Ap, their peak is Sea, where they are multiplied after that by connecting them to the prediction coefficients before starting transmission from (VY) encoding: radio transmission, for example. Tarashala two predictive filters with the same impulse response to a weight (071) including excitation operator Vo.

Ap Depending on the prediction coefficients, p during a given calculation or the computational phase amped; ody and signal (y) are correlative signals that work to achieve correlation between the original weighted signal alse includes: the weighted artifact (9) each time an excitation pulse is generated. For each link process, one of which gives the smallest error (0 < i > 1) my cA; for the elements of the pulse “candidates” (q) the number of the “candidate” selected in the my generator, and the mode temporal App Quadratic or smallest absolute value. The amplitude is calculated as 0 eta from the desired signal in eA, the excitation signal. The contribution from the extruded pulse is then subtracted by the coherence signal so that a new series of 'candidates' can be obtained; The method repeats as many times as many times as the desired number of excitation pulses within the frame.These steps are described in detail in the aforementioned US patent specification. xe which (A) Asie Aad and the excitation pulses in order and the number of excitation pulses in this dy

YAR

0 from which the my ¢ Ary pulse was selected first (it gave the smallest error) and then the my Ag etc pulse within the frame. : In the previously known method of calculating the amplitude, Ap, and phase mode, my, for each excitation pulse; my, = 0, was calculated for that pulse which gives the maximum value of 010/th; Then calculate the amplitude, Ap, correlated 0 where om is the cross-correlation vectors between the signals ,z and .37 according to above; and 00070 is the autocorrelation matrix of the impulse response of the prediction filters. Any my status, whatever it is, is accepted only when the above conditions are met. The indicator means the stage in which the calculation of the excitation pulse occurs according to what is: mentioned above. According to the invention; The frame illustrated by Figure (Y) is divided in the manner shown in Figure (9). From 0 it is assumed that the frame has VY = N mode. In this case; Modes 11 Search vector (©). The entire frame is divided into so-called partial assemblies 9010-010©168; Each sub-assembly contains a certain number of phases. For example; If the whole frame contains VY = 17 and placed according to the figure (), four Jali sub-assemblies are obtained, each sub-assembly has three different phases. Each sub-assembly has a specific position within the whole frame; This position is referred to as phase mode. 1 Then each mode <n <N(n 0) belongs to a given subassembly shally (0 > ne > Nong the determinant 65 >? > 0) in said subassembly. In general, the 8 modes become < < n 0) in the overall search vector that includes modes 27 F+{Unified N¢1),...,0 x term, (1- ... spirit a and n=0, .. , (N-1) In addition, it is also possible to apply the following relation (...) (1) aspiration 5 f=nMOD The graph of Figure ( ) shows the distribution of © phases and partial assemblies np on the search vector vo given that includes modes 17. In this case 1712, F = 3 and Np = 4 m, the innovative method aims to limit the search for the pulse to the modes that do not belong to the occupied phase 5, for the excitation pulses whose positions 8 were calculated in Previous stages

: 0 The order or sequence of the arithmetic cycle number specified for the excitation pulse is marked as p according to the above; Then the proposed method will result in the following computational stages for a time frame period: -1 Calculate the desired signal Ya 7- Calculate the cross-correlation vector 1 0 0 *- Calculate the autocorrelation matrix ij for ;- When it is p=1 Find omy, i.e., the pulse mode that gives Tan max HD0/sh<:0/:e in unoccupied phases £ : 0 - #- Calculate the amplitude App of the detected pulse mode my = create cross-correlation vector 0 ; 0 7- Calculate da ngs f for relation (1) above; A—carry out steps 4-7 above when ppl and Figures (4a)-(;k ) diagrams of the proposed method. Figures (14)-(© e) show an example in which the number of modes within one frame is (N = Yi and the number of phases; = 17 and the number of phase modes 6 = Np. It is assumed that the phases are not It is initially occupied at 0-1. It is also assumed that the arithmetic phases above 1-a are given the position .my=5 and a circle is placed around this pulse position in the figure ig) This gives the phase (1) in The special phase modes are 0-0,1,2,3,4,5; the symmetric pulse modes are 0-1,5,9,13,17,21 Wh of relation (1) above. (1) and corresponding pulse modes when calculating the mode of the next excitation pulse (0-2). It is assumed that x. is the arithmetic stage; of p=2 results in mp=T that would be my=9 if given the maximum value of only even though that gives a busy phase. Phase position 7-Z“ gives phase (7) in each of the phase positions 0-0,...5; it means that the pulse modes 0=3,7,11,15,22 will be occupied. YY YY 14 AY Ao AY before the start of the next arithmetic stage 0 (0-3). Yo It is assumed that the above arithmetic stages 1-; of p=3 are given my=12 and that Arithmetic steps 4-8 yield the last mode my=22 so all modes are inside CY YA

A

etc. obtained (App,mp)s (Agy,my) excitation pulses (it) busy. Figure 0 shows

Jada Figures (5 and)-(; k) show another example in which 14-25 and 15 and 5-c 21 are added; That is, the number of phases, each mode is one phase. The pulse is positioned in the same way as in Figures (4a)-(5e) and at the end five excitation pulses are obtained. Thus, the maximum number of excitation pulses that he reached is equal to the number of phases within one phase mode. They are separately ng, Dg, Figures (4 and)-(;k) together, while the modes of the resulting phases are encoded. Compatible coding can be used to encode the phase. And each of the 0 modes before the phase transmission process is encoded by a code word by itself. 0 . . (5) According to one embodiment, the known speech actuator circuit may be modified in the manner shown in Fig. (VY) which shows that part of the speech actuator comprising excitation generation circuits on the special filter (VYY). The predictive residual, L, and excitatory signals are produced (118) (177) by FC gates simultaneously with the frame signal (17), and (VY) represents the signal (1"©). 9. Wi signals that are associated in correlation generator (YY) (1?1) filters 0 the artificial speech signal. From correlation generator 3, Ji is obtained the real speech signal; while y, according to the above. An arithmetic operation is performed in gs 0 on the © signal that includes the components (YY0) through which the pulse mode 0, which gives a maximum (VYV) excitation ase my, is performed according to the above in addition to To the pulse mode Arp Obtaining the capacitance To the phase generator (VTY) Produced by the excitation generator my Ap The excitation pulse parameters Y. m which App are sent from the values ng This generator calculates the current phases )£ and phase modes (VY) according to the relation (VY) delivered from the exciter © f=(m-1) MOD 1+1 ne= (m-1) DIV F+1 Number of possible phases. =F where vo from an actuator including a read memory functioning to store instructions (VY) The generator of phases for calculating phases and phase mode may be composed according to the above relation. Cc q with phase and phase mode. This encoder having the same structure (TY) then supplies the basic 0 encoder to the known encoder but operates in phase and phase mode rather than pulse modes and on the receiver side; The phases and phase positions are decoded as the amy decoder computes, according to the relation: my, then encodes the phase position mp=(ng,-1). F+fp °, which gives a clear identification of the excitation pulse mode. The correlation generator (YY) and the excitation generator (YO) phase 5 is also supplied to the correlation generator: busy. No values for the signal e fi are calculated to store this phase, taking into account that this phase is included in these positions that belong to all phases, such as the previous ones calculated for the sequence q, if they are the analyzed and occupied positions are: ye q=n. F+f; The ally generator takes all previous phases occupied within the frame. 0 = 0, .. (Ng -1) where J * 0. Cig takes into account the occupied phases when making a comparison between the signals (VV) Excitation | s When all the pulse positions for one frame have been computed and processed and when the next frame is about to start all phases are vacant again to receive the first pulse in the new 10th frame. sel yl The figure shows (1) and (b) the flow diagram shown in Figure (©) of the aforementioned (FY) (YTV) US specification which has been modified to include phase limits. Interspersed between assemblies of a phase generator my An, (in Assembly place (7748) which is concerned with calculating the output signals Assembly (748?a) which is concerned with the calculations_ done in the ep generator and the position indicator (1) oY lists the phases” and then assembly (748*b) (1) For the relation Wb, 5 and (Yo) and (VY) encoding generators (17) are calculated for the relation Wb. Which q obtained is assigned a vector 1n that is used when testing the value (177) and (YYO) in the two generators in order to confirm whether the symmetrical phase mode is given by Ol Bru on it * which gives the maximum value (between (YA) ("7 A)s (IY +A) This test is performed on LIA assemblies in busy phase or ve (11) assemblies () and (10))_and in_assemblies (1318) and (41) b) (Between_assemblies and (c) in the (YA) generator (17+ A) the instructions given by the assemblies are executed.))?15(f)

YAR

. 1 in the excitation generator (B14) and (FIA) the interconnection (761); while the instructions given by the assemblies (YY) i.e. the phase of the indicator © are carried out according to the aforementioned, where then of is done first; The signal bm] is calculated if it is equal to tp 1. A test is performed to confirm whether the mode of the router is for phase © in the router exactly; no correlation calculations are made according to g* which means that the phase Busy for this indicator 0 Comparisons in assembly (99). On the other hand when Bally (Yr) for instructions from assembly and subsequent computations are done as mentioned before. JA This indicates phase ug=0 will still All occupied phases during all computed sequences are connected to a time-slot frame (YoY) is set complete, but it will be empty at the start of the time-slot frame.So further to the directed assembly ben to zero before each new frame analysis. With different excitation pulses within the frame, each my, when encoding the modes, is also encoded, thus dividing the encoding of the modes into two separate code words fy and the ng phase of the a phase mode have a different meaning interchangeably. The bits in the encoder's words have a different meaning interchangeably; Thus the degree of sensitivity to bit error will also vary. This difference is advantageous for coding a correction or error detection channel. 1 The aforementioned constraint in excitation pulse regulation means that encoding of the pulse modes occurs at a lower bit rate than when coding the modes in multiple pulses without the said constraint. This means that the search algorithm will be less complicated than without this limitation or limitation, and the innovative method will undeniably include some limitations when regulating the impulses. It will not always be possible to achieve an accurate determination of the pulse position but this limitation will not be of great importance when compared with the aforementioned advantages. pulse by pulse until the frame of the time period fills there is an encoder and it works with a pattern regulation of 155 7 speech of another type described in the European patent application a pulse in which the temporal distance between the pulses remains constant rather than being variable The innovative method can also be applied to a speech encoder of this type.

11 While a specific embodiment of the present invention has been described and illustrated, it should be recognized that the invention is not limited to it since those skilled in the art can make modifications to it. The present application includes any of the modifications that fall within the scope of the invention described in

ASN this specification and claims to protect it according to the protections

YAR

Claims (1)

VY : 0 Protections 1-1 Method for regulating Seal excitation pulses Predictive coding: linear predictive coder (Ipc) Y and for encoding the regulating information where Led forms a synthesized speech signal v from an original speech signal including: 1 0) Determine the number of predictive parameters characterizing said original speech signal ° within a time frame interval 1 Calculate a residual signal representing Usd between said original speech signal 7 and said synthesized speech signal A within said time frame interval and generate a format array aun excitation pulses 9 time frame interval based on said residual signal Ve and said predictive parameters. 1© signal Yo weighted by the weight of said residual VY signal with said predictive parameters. Vr (3) Generation of a synthesized speech signal weighed 3, by the weight of a graphic signal Jidi speech-representative signal Vt Time mode amplitude of one of the excitation pulses 3_5S4dll pulses Vo Co with predictive parameters [355%]: (e) Associate each number of modulation stages 1 mentioned, weighted speech-representative signal 7 mentioned xa Y, weighted synthesized speech signal YA mentioned Y. To specify the difference signal 14 for each mentioned dal yall. Ye (9) Determine the candidate for the excitation pulse for each of the mentioned Jal gall 2 representing the amplitude of Ap and time mode my From the aforementioned connection to that stage, determining the minimum value TY of the difference signal mentioned between the difference signals of all TY filters candidates, and selecting a candidate candidate that corresponds to the mentioned minimum value of Yt to obtain the Ap amplitude, the time mode my of one of said excitation pulses Yo; and the frequency of the procedure for determining the pulse candidate for a desirable number of YAR pulses 0 7 Excitation excitation pulses in a frame ignoring specific excitation pulses in Tv previous modulation stages. YA (g) Divide the total number of possible time modes n of excitation pulses within the said timeframe Ye into eng phase modes and each phase mode includes a number of phases © such that 7 is aren where F is the total number of phases? In a given phase mode Dp m (h) Determination of amplitude and time mode according to steps (d) to (f) of the first and subsequent excitation vy 8 pulses between time modes “which have corresponding phases in each phase mode rr without being preoccupied with the time positions of the previous excitation pulses until vi finds a number of predetermined excitation pulses within the aforementioned time frame interval Yo. } na (i) encode each np specific phase mode separately to form independent code words. (b) Coding the specific phases mentioned together to form a single coding word. 1 ”- method according to protection Cus) phase, a and phase mode ng, corresponds to an amplitude and time mode my Y Define for a given excitation pulse p calculated according to the relation: ng, = (m,-1) Mod F+1 v (m,-1) Div F+1 1 =,£° where the value of phase ,5 is mentioned only in AS modes np within the time frame time frame interval 1 mentioned determines which time mode of the excitation pulse following the given excitation pulse p should be suppressed and this procedure A is repeated for each excitation pulse until Obtaining the ae excitation pulses Ase salt pulses 4 in the frame. 1 *- A method according to the claim ¥ also includes: Y Generate a test vector From the number © of the pulse phases Gad One phase mode Dp Between: A group of phase modes of a frame that includes Alla Availability of each phase within the aforementioned time frame. ¢ Determination of a phase in the mentioned test vector that corresponds to the specified time mode according to step 0 (fo) YAR Ve Determines whether said specified phase is available for a particular phase mode based on test prompt x 0 mentioned test vector 7; shad if a mode is available le if the specified mode is not available; another A is selected; 4 and if the specified phase mentioned is available; Steps (e) and (g) are performed, respectively, for 1 11 post pulse; 1” and update said test vector. 01 ¢— method according to claim 0 also includes generation of test vector from number 1 for phases Y is the pulse in one phase mode m” between de gana the phase modes of a frame representing the availability status of each phase And within each mode in the mentioned timeframe. 1 Determine a phase in the aforementioned test vector that corresponds to the specified time mode according to 0 for step (h). x Le determines if the specified phase mentioned is available for a particular phase mode based on the test prompt test vector 7 mentioned. A If the specified phase is not available, select Le if the phase is available for another mode. 4 If the specified phase mentioned is available; Steps (f) and (e) are performed, respectively; To put 01 next pulse. 11 and update the aforementioned test vector. YAY |