KR20110086919A - Transcoding method and transcoding apparatus for smv and amr speech coding schemes - Google Patents

Abstract

PURPOSE: A mutual encoding method for applying the encoder of an SMV and AMR is provided to maintain the voice quality by a delay time and a small amount of calculation in using a difference communication network. CONSTITUTION: An encoding device generates a SMV parameter by encoding a received SMV(Selectable Mode Vocoder) bit stream(S320). The encoding device performs an LSP(Line Spectral Pairs) conversion(S330). If a transmission rate is set to 1/4 or 1/8, the encoding device sets a pitch delay as 0(S340,S3350). If the transmission rate is set to 1 or 1/2, the encoding device performs a pitch delay and ACB(Adaptive Code Book) conversion(S360). The encoding device performs a fixing code book conversion(S370).

Description

Transcoding method and transcoding apparatus for SMV and AMR speech coding schemes

The present invention relates to a method of integration and efficient interworking between data networks using different communications. In particular, the present invention relates to SMV and AMR through effective mutual coding between Selectable Mode Vocoder (SMV) and Adaptive Multi-Rate (AMR) speech coding techniques. It relates to a method and apparatus for interworking.

Various communication networks use a standard voice coding method according to each communication network. For example, a general wired telephone network uses an ITU G.711 speech coder of a pulse code modulation (PCM) method, and an ITU G.723 and an ITU G. CCE of a Code Excited Linear Prediction (CELP) method in a VoIP network. 729.A speech encoders are used, and CELP-based adaptive multi rate (AMR) and enhanced variable rate codec (EVRC) speech encoders are used as standards.

The basic SMV and AMR voice encoders are as follows.

1 is a block diagram conceptually illustrating a conventional SMV voice encoder.

The SMV voice encoder 100 shown in FIG. 1 has multiple transmission rates of 8.55 kbps (Rate 1), 4.0 kbps (Rate 1/2), 2.0 kbps (Rate 1/4), and 0.8 kbps (Rate 1/8). Coder in bitrate mode. These modes are determined by network conditions, and different modes can be selected for each frame, so that the average transmission rate and sound quality can be properly balanced. The rate for each frame is determined by a Rate Decision Algorithm (RDA), and the excitation signal for the LPC synthesis filter is calculated in different ways according to the determined rate. Excitation signal calculation of Rate 1 and Rate 1/2 is done by eX-CELP method, and rate 1/4 excitation signal is generated by Random Number Generator, and then multiplies the gain every 2ms subframe and frequency Obtained by passing through a Frequency Shaping Filter. The rate 1/8 multiplies the signal obtained by the random number generator with one gain value to generate the excitation signal.

2 is a block diagram conceptually illustrating a conventional AMR speech encoder.

The AMR speech encoder 200 supports eight modes from 4.75 kbps to 12.2 kbps, and linear prediction analysis is performed in common in all modes. Then, the excitation signal is modeled by adding the adaptive codebook vector and the fixed codebook vector by the gain of each codebook, and using the ACELP (Algebraic Code Excited Linear Prediction) method.

However, since there are various voice coding techniques as described above, efficient interworking of different communication networks has been an important issue. However, since different voice coder standards are not compatible with each other, a process of decoding a packet coded by the voice coder A to match the voice coder B is necessary.

There are two methods for doing this, tandem and mutual encoding. The tandem method continuously repeats the decoding / decoding process, so the delay time is long, the calculation amount is high, and the sound quality is degraded. The purpose of the inter-coding method is to reduce computation and latency by maintaining sound quality similar to that of the tandem method.

However, SMV is based on eX-CELP (Extended Code Excited Linear Prediction) as a voice coder for CDMA 2000 systems, and AMR is an ACELP (Algebraic Code-Excited) as a voice coder widely used in GSM and WCDMA mobile communications. Based on the Linear Prediction method. Therefore, a terminal supporting the AMR technique cannot use a network supporting the SMV technique. However, in recent years, the roaming (roaming) technology that can be used anywhere in the world by using their own terminal has been in the spotlight. In order to implement such a roaming technique, it is necessary to ensure compatibility between voice encoding techniques used in different communication networks.

Therefore, a terminal and the like that support both the SMV and AMR techniques are required a method and apparatus that enables mutual encoding between the AMR and SMV techniques while maintaining the sound quality without increasing the amount of computation.

It is an object of the present invention to provide an intercoding method for reducing computation and delay time while minimizing sound degradation in the intercoding method between AMR and SMV.

It is another object of the present invention to provide an apparatus for mutual encoding between an SMV and an AMR speech encoder, which applies a line spectral pairs (LSP) transform, a pitch delay transform, a fast fixed codebook vector search for pulse rescanning, and a rate determination method.

A first aspect of the present invention for achieving the above object is, for converting a bitstream coded by the Selectable Mode Vocoder (SMV) speech coding scheme to a bitstream coded by the Adaptive Multi-Rate (AMR) speech coding technique It is about a method. The method according to the present invention decodes a bitstream encoded by the SMV speech coding scheme to determine SMV parameters including SMV LSP (Line Spectral Pairs), SMV pitch delay, SMV rate, adaptive codebook gain, and fixed codebook gain. SMV parameter decoding step for generating, LSP conversion step for converting SMV LSP into AMR LSP to be used in AMR speech coding, determining whether the transmission rate is 1 or 1/2, and if positive, SMV pitch delay and adaptive codebook gain, respectively Convert the SMV fixed codebook vector to the AMR fixed codebook vector using a pitch delay / adaptive codebook transform step that transforms the AMR pitch delay and adaptive codebook gain and sets the AMR pitch delay to zero, the SMV adaptive codebook vector and the adaptive codebook gain. A fixed codebook transform step, and AMR including AMR LSP, AMR pitch delay, codebook gain, and AMR fixed codebook vector A bit stream encoded by an audio encoding method using SMV la meter includes AMR parameter coding step of encoding the AMR speech coding scheme. In particular, the LSP conversion step uses the SMV LSP corresponding to the fourth subframe of the nth frame of the decoded packet as it is as the fourth subframe of the nth frame of the AMR LSP, and the (n-1) th of the SMV LSPs. Interpolating the SMV LSPs corresponding to the fourth subframe of each of the frame and the n th frame and using the AMR LSPs corresponding to the second subframe of the n th frame of the AMR LSP. In addition, the pitch delay / adaptive codebook conversion step includes predicting the open loop pitch delay past value of the pitch delay with the pitch delay P _AMR of the AMR when the transmission rate is 1 or 1/2, and the closed loop pitch delay P of the P _AMR and SMV. Determining whether the difference of _SMV is greater than a predetermined threshold, retrieving the pitch delay of AMR if positive, and setting closed loop pitch delay P _SMV to pitch delay P _AMR if negative, and pitch delay P _AMR An ACB search step for searching an Adaptive Code Book (ACB). Further, the fixed codebook conversion step may include generating an object signal for fixed codebook retrieval from the SMV adaptive codebook vector and the adaptive codebook gain, the long _term prediction residual signal res _LTP (n), and the correlation between the object signal and the impulse response d ( correlation between n) for use by the reference vector b (n) the operating step, the step of selecting from a large value of the reference vector to extract the initial pulse array for the location search, the object signal and the impulse response of FIG d (n) and the impulse Computing the search criteria Q _k of the initial pulse arrays using the autocorrelation matrix Φ ( i , j ) between the responses, and removing one pulse per two adjacent tracks among the pulses belonging to the pulse array. removing pulses, change the items Q _k reflects the removed pulse, and a removing pulse to maximize the search criterion Q _k is changed pulse To 2 pulses of selecting the track with re-search step, and a second pulse from the pulse array obtained by the re-search step reflect the pulse removed while removing one by one pulse, and change the items Q _k, the pulse that maximizes the modified search criterion Q _k A single pulse rescan step of selecting. In particular, the two-pulse rescanning step further includes the step of simplifying the retrieval by modifying the correlation d (n) between the target signal and the impulse response and the autocorrelation matrix Φ ( i , j ) between the impulse responses to include a preselected sign. .

A second aspect of the present invention for achieving the above object relates to a method for converting a bitstream encoded by the AMR speech encoding technique into a bitstream encoded by the SMV speech encoding technique. According to a second aspect of the present invention, there is provided a method of decoding an AMR parameter to generate AMR parameters including an AMR LSP, an AMR pitch delay, an adaptive codebook gain, and a fixed codebook gain by decoding a bitstream encoded by an AMR speech coding technique. LSP conversion step for converting AMR LSP into SMV LSP for use in SMV speech coding, adaptive codebook gain, fixed codebook gain, noise to signal ratio (NSR), and AMR parameter decoding A rate determining step of determining whether the transmission rate to be used for SMV encoding using a signal is 1, 1/2, 1/4, or 1/8, and if the transmission rate is 1/4 or 1/8, excitation is performed using a random number generator. Generating an excitation signal, if the rate is 1 or 1/2, pitch delay / adaptation to convert the AMR pitch delay and adaptive codebook gain to SMV pitch delay and adaptive codebook gain, respectively Performing a fixed codebook transform operation that converts the AMR fixed codebook vector to an SMV fixed codebook vector using the drainbook transform operation and the AMR adaptive codebook vector and adaptive codebook gain, and SMV LSP, SMV pitch delay, codebook gain, and SMV fixed And an SMV parameter encoding step of encoding a bitstream encoded by the AMR speech encoding technique using the SMV parameters including the codebook vector. In particular, the LSP conversion step includes using the AMR LSP corresponding to the fourth subframe of the nth frame of the decoded packet as it is as the fourth subframe of the nth frame of the SMV LSP. In addition, the rate determining step may be performed by referring to the adaptive codebook gain and the fixed codebook gain. A second classification step of classifying the unvoiced and the voiced sound using the magnitude of the adaptive codebook gain for the current frame that is not silent, and a third classification step of determining whether the previous frame class of the frame classified as the voiced sound has been unvoiced In the case where the frame does not change, the fourth classification step for determining whether the normal state or abnormal state according to the change of the adaptive codebook gain and the pitch delay, and the transmission rate of the frame differentially selected according to the classification result of the classification steps Steps. Further, the pitch delay / adaptive codebook conversion operation may include predicting the open loop pitch delay past value of the pitch delay as the SMV pitch delay, determining whether the difference between the SMV pitch delay and the closed loop pitch delay of the AMR is greater than a predetermined threshold value, Retrieving the pitch delay of the SMV if positive, setting the closed loop pitch delay to the pitch delay if negative, and ACB retrieving to retrieve the adaptive codebook using the pitch delay.

A third aspect of the present invention for achieving the above object relates to an apparatus for converting a bitstream encoded by the SMV speech encoding technique to a bitstream encoded by the AMR speech encoding technique. An apparatus according to a third aspect of the present invention includes a processor for executing instructions that can be executed by a computer, and a memory coupled to the processor, the memory for storing instructions that can be executed by the computer, and can be executed by the computer. Instructions are adapted to execute the method according to the first aspect of the invention.

A fourth aspect of the present invention for achieving the above object relates to an apparatus for converting a bitstream encoded by the AMR speech encoding technique to a bitstream encoded by the SMV speech encoding technique. An apparatus according to the fourth aspect of the present invention includes a processor for executing instructions executable by a computer, and a memory coupled to the processor, the memory for storing instructions executable by the computer, and executable by the computer. Instructions are adapted to execute the method according to the second aspect of the invention.

According to the present invention, parameters extracted by applying the LSP transform method, the pitch delay, the adaptive codebook transform method, and the fixed codebook transform method in encoding from SMV to AMR are transmitted to the AMR encoder. In encoding from AMR to SMV, parameters extracted by applying a transmission rate, a type determination method, and an excitation signal conversion method of the AMR decoded signal are transmitted to the SMV encoder so as to be applicable to the mode of the SMV. Therefore, by using the present invention, the SMV and AMR mutual encoders can be used to maintain sound quality performance with a small amount of computation and delay time in using different communication networks.

1 is a block diagram conceptually illustrating a conventional SMV voice encoder.
2 is a block diagram conceptually illustrating a conventional AMR speech encoder.
3 is a flowchart illustrating a method of mutual encoding from SMV to AMR according to a first aspect of the present invention.
4 is a diagram illustrating an LSP conversion operation performed in the mutual encoding method and apparatus according to the first and third aspects of the present invention.
5 is a flowchart illustrating a pitch delay conversion operation from SMV to AMR performed in the mutual encoding method and apparatus according to the first and third aspects of the present invention.
6 is a block diagram conceptually illustrating an apparatus for SM-to-AMR encoding according to a third aspect of the present invention.
7 is a flow diagram illustrating a method of intercoding from AMR to SMV in accordance with a second aspect of the present invention.
8 is a block diagram conceptually illustrating an apparatus for mutual encoding from AMR to SMV according to a fourth aspect of the present invention.
FIG. 9 is a diagram for describing a rate determining process included in the mutual encoding method illustrated in FIG. 7.

In order to fully understand the present invention, the operational advantages of the present invention, and the objects achieved by the practice of the present invention, reference should be made to the accompanying drawings which illustrate preferred embodiments of the present invention and the contents described in the accompanying drawings.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In addition, in order to clearly describe the present invention, parts irrelevant to the description are omitted, and the same reference numerals in the drawings indicate the same members.

Throughout the specification, when a part is said to "include" a certain component, it means that it may further include other components, without excluding the other components unless otherwise stated. In addition, the terms "... unit", "... unit", "module", "block", etc. described in the specification mean a unit for processing at least one function or operation, which means hardware, software, or hardware. And software.

3 is a flowchart illustrating a method of mutual encoding from SMV to AMR according to a first aspect of the present invention.

First, the SMV bitstream is received (S310). Then, the received SMV bitstream is decoded to generate an SMV parameter used in the encoding process (S320). At this time, an excitation signal of the SMV is used to calculate an object signal to be used in the AMR codebook search.

LSP conversion step (S330) is performed the same regardless of the rate. When the LSP conversion is performed, it is determined whether the transmission rate of the SMV is 1 or 1/2 (S340). The reason for confirming the transmission rate of the SMV is that different mutual encoding processes should be performed according to the four transmission rates of the SMV.

If the transmission rate is 1/4 or 1/8, the pitch delay is set to 0 (S350). On the other hand, if the transmission rate is 1 or 1/2, the pitch delay and adaptive codebook conversion operation is performed (S360). As such, when the pitch delay and the adaptive codebook are converted, a fixed codebook conversion operation (S370) is performed. When all AMR parameters for AMR speech encoding are determined as described above, AMR speech encoding is performed using the determined AMR parameters to generate an AMR bitstream.

An apparatus for implementing the mutual encoding method shown in FIG. 3 is shown in FIG. 6.

6 is a block diagram conceptually illustrating an apparatus for SM-to-AMR encoding according to a third aspect of the present invention.

The mutual encoding apparatus 600 shown in FIG. 6 includes an SMV parameter decoder 610, an LSP converter 620, a rate determiner 630, a pitch delay and adaptive codebook converter 640, a fixed codebook converter 650, and An AMR parameter encoder is included (660).

The SMV parameter decoder 610 performs a parameter decoding process on the encoded packet of the SMV, and selects parameters to be converted during the mutual encoding process. Thereafter, the SMV parameter decoder 610 calculates an object signal to be used in the AMR codebook search process using the excitation signal of the SMV. Different intercoding processes are performed according to the four data rates of the SMV, but the LSP conversion process is performed the same regardless of the data rates. The pitch delay and adaptive codebook converter 640 only operates when Rate 1 and Rate 1/2 of the SMV. The reason is that the rate delay and the adaptive codebook conversion process are not performed because Rate 1/4 and Rate 1/8 have no information about the pitch. In addition, the fixed codebook converter 650 operates at all rates. The AMR parameter encoder 660 generates an AMR bitstream using the outputs of the LSP converter 620 and the fixed codebook converter 650.

4 is a diagram illustrating an LSP conversion operation performed in the mutual encoding method and apparatus according to the first and third aspects of the present invention. In Fig. 4, the thick parts represent the window weights.

In FIG. 4, the SMV and the AMR have the same frame size and the number of subframes. A weighted window is used for the fourth subframe of the SMV and the second and fourth subframes of the AMR. In addition, only the LSPs of the second and fourth subframes are transmitted in the 12.2 kbps mode of AMR, and only the LSPs corresponding to the fourth subframe are quantized and transmitted in all other SMV and AMR modes. Therefore, the LSP to be used in the AMR speech encoder can be converted using the LSP of the SMV as it is or through simple linear interpolation. Equation 1 is used only at 12.2 kbps of AMR, and Equation 2 is used at all data rates.

5 is a flowchart illustrating a pitch delay conversion operation from SMV to AMR performed in the mutual encoding method and apparatus according to the first and third aspects of the present invention.

Pitch delay plays an important role in adaptive codebook retrieval in CELP, so even small errors have a big impact on sound quality. Therefore, when the decoded pitch delay in the SMV packet is used as it is in the AMR decoding process, sound quality deterioration occurs.

The pitch delay converter (see 640 of FIG. 6) predicts the pitch delay using the open loop pitch delay past value (S510). Then, the predicted pitch delay is compared with the pitch delay of the decoded SMV through the parameter decoding process (S520). If the difference between P _SMV and P _AMR is greater than the threshold, an open loop pitch delay search of the AMR is performed to obtain a more accurate pitch delay (S540). On the other hand, if the difference between P _SMV and P _AMR is smaller than the threshold value, the closed loop pitch delay of the SMV is selected as the open loop pitch delay of the AMR, and thus the open loop pitch delay search process is omitted (S530).

Referring back to FIG. 6, the fixed codebook vector converter 650 performs an operation requiring the largest amount of computation in the encoding process of the two speech encoders. The fast fixed codebook retrieval apparatus used in the present invention is a repetitive pulse position retrieval apparatus using a reference vector. That is, the fixed codebook vector converter 650 according to the present invention first performs two-pulse rescan using a pulse array extracted from a reference vector and then performs a single pulse rescan.

Equation 3 shows a fixed codebook object signal.

In Equation 3, x ( n ) is an adaptive codebook object signal, y ( n ) is a filtered adaptive codebook vector, and g _p is an unquantized adaptive codebook gain. g _p is a measure of how well the adaptive codebook vector models the excitation signal. Therefore, a frame with a high g _p has a relatively low contribution of the fixed codebook, whereas a frame with a low g _p has a relatively important contribution with a fixed codebook.

See also the following equations (4) and (5).

In Equations 4 and 5, d ( n ) is a correlation between the target signal x ₂ ( n ) and the impulse response h ( n ), and Φ ( i , j ) is an autocorrelation matrix between h ( n ).

The fixed codebook then selects pulse positions that maximize the following equation (6), which is a search reference.

The purpose of the search is to determine the code vector with the optimized pulse position. To obtain the optimized pulse position, first, the position of the pulses for the initial search is obtained by using Equation 7, which represents the reference vector.

In Equation 7, res _LTP ( n ) is a long _term prediction residual signal.

Then, the pulses that can be placed in each track are selected from the larger values in b ( n ) to produce an initial pulse array for position search. For 2 pulse rescan and single pulse rescan, the search criteria Q _k of the initialized pulses must be calculated.

To simplify the search, the signal d ( n ) and the matrix Φ ( i , j ) are modified to include a preselected sign. The modified signals d ' ( n ) and Φ' ( i , j ) are defined by Equations 8 and 9, where s _b ( n ) is the sign of b ( n ).

Then, R and E of the molecules of Equation 6 are obtained using Equations 8 and 9 as shown in Equations 10 and 11.

Once R and E are found, two-pulse rescan is performed first. That is, after removing two pulses by one pulse from two adjacent tracks, a new pulse combination is searched for in the track from which the pulses are removed. This selects to maximize the search criterion Q _k to search for previously searched pulses in two adjacent tracks. After two pulses are removed, the numerator and denominator of the search criterion Q _k '' after the pulse is removed are recalculated using Equations 12 and 13.

Then, in the next step, the numerator and denominator are corrected by adding the amount of change in the newly added pulse. Select the pulse that maximizes the new Q _k in the track to which the removed pulse belongs, and R and E are modified by equations (14) and (15).

In searching for a new pulse, v _h ( n ) and R _hv ( m ) are calculated by Equations 16 and 17 during execution for all selectable cases in the track.

A single pulse rescan is performed with the pulse array obtained through the two pulse rescan. At each iteration, the most appropriate position of each pulse in the track is retrieved while maintaining the position of other pulses previously searched in the track. To find a new pulse, first remove one pulse position and then calculate the Q _k value after it has been removed. The numerator and denominator of the search criterion Q _k ' after one pulse position is removed are calculated by the following equations (18) and (19).

In the next step, the numerator and denominator are modified by adding the amount of change in the newly added pulse. Select the pulse that maximizes the new Q _k in the track to which the removed pulse belongs. R and E are modified by equations (20) and (21).

Hereinafter, a description will be given of a technique of mutual encoding from AMR to SMV voice encoder.

7 is a flow diagram illustrating a method of intercoding from AMR to SMV in accordance with a second aspect of the present invention.

When the AMR bitstream is received (S700), the AMR bitstream is decoded to generate AMR parameters including AMR LSP, AMR pitch delay, adaptive codebook gain, and fixed codebook gain (S710). Then, the AMR LSP is converted into the SMV LSP to be used in the SMV speech coding technique among the AMR parameters (S720). Then, using the adaptive codebook gain, the fixed codebook gain, the noise to signal ratio (NSR), and the speech signal reconstructed in the AMR parameter decoding step, the transmission rates for the SMV encoding are 1, 1/2, 1 One of / 4 and 1/8 is determined (S730). The transmission rate determination method will be described later in detail with reference to FIG.

When the transmission rate is determined, it is determined whether the corresponding determination rate is 1 or 1/2 or 1/4 or 1/8 (S740). If the transmission rate is 1/4 or 1/8, an excitation signal is generated using a random number generator (S750). On the other hand, if the data rate is 1 or 1/2, the AMR pitch delay and the adaptive codebook gain are converted into the SMV pitch delay and the adaptive codebook gain, respectively (S760).

In addition, the AMR fixed codebook vector is converted into an SMV fixed codebook vector using the AMR adaptive codebook vector and the adaptive codebook gain (S770). When the SMV parameter is obtained in this manner, the obtained SMV parameter is encoded to generate an SMV bitstream (S780).

8 is a block diagram conceptually illustrating an apparatus for mutual encoding from AMR to SMV according to a fourth aspect of the present invention.

The mutual encoding apparatus 800 shown in FIG. 8 includes an AMR parameter decoder 810, an LSP converter 820, a rate determiner 830, a pitch delay and adaptive codebook converter 840, a fixed codebook converter 950, A signal converter 860, and an SMV parameter encoder 870 (660).

The AMR parameter decoder 810 performs a parameter decoding process on the AMR encoded packet to select the parameters to be transformed in the inter-encoding process. Then, the LSP converter 820 converts the decoded signal into an LSP and delivers it to the SMV parameter encoder 870. The rate determiner 830 determines the rate through class classification, and transmits it to the pitch delay and the adaptive codebook converter 840 and the excitation signal converter 860 according to the rate. The configuration and operation of pitch delay and adaptive codebook converter 804 and fixed codebook converter 805 are similar to sebum delay from SMV to AMR speech coder and adaptive codebook converter 640 and fixed codebook converter 650, thus simplifying the specification. Duplicate description is omitted for the sake of brevity.

The intercoder from AMR to SMV speech coder can be obtained by direct conversion with the LSP decoded in AMR and is converted at all transmission rates of SMV. Unlike AMR, however, SMV determines the data rate and type for each frame, and requires an excitation signal and a quantization process. In order to obtain the data rate and type, information on the excitation signal of the AMR and the speech signal reconstructed by the AMR decoding process are used. Rate 1 and Rate 1/2 perform fixed codebook search after obtaining the closed loop pitch delay and adaptive codebook gain through the adaptive codebook search method of SMV based on the open loop pitch delay obtained through pitch conversion. At rates 1/4 and Rate 1/8, the excitation signal converter 860 generates an excitation signal using a random number generator.

In AMR and SMV speech encoders, LSP conversion has the same frame and subframe size and the same number of subframes. Therefore, the LSP converter 820 applies the fourth subframe decoded in the AMR as shown in Equation 22 below.

FIG. 9 is a diagram for describing a rate determining process included in the mutual encoding method illustrated in FIG. 7.

In order to determine the transmission rate of the SMV, a reconstructed speech signal decoded into a CELP parameter (adaptive codebook gain, fixed codebook gain, noise to signal ratio (NSR)) and AMR packet is used in the AMR packet. In SMV, it is classified into six types of silence, noise-like, unvoiced, onset, abnormal voiced, normal voiced, but mutual encoder except noise. It is classified into five frame classes.

In the primary classification step 901, speech and silence are first classified using information required in the AMR packet (pitch delay, NSR, adaptive codebook gain, fixed codebook gain). In this case, since the pitch delay of unvoiced or unvoiced sound tends to be large, the noise-to-signal ratio (NSR) is more than a predetermined value or the pitch delay is large by referring to the adaptive codebook gain and the fixed codebook gain which have been subjected to appropriate linear processes. Classify as silent. This is because the pitch delay of unvoiced or silent sounds tends to be large.

In the second classification step 902, the speech section is classified into an unvoiced sound and a voiced sound using the magnitude of the adaptive codebook gain.

In the third classification step 903, when the frame class of the last frame in the voiced sound section is an unvoiced sound, it is classified as an onset. Otherwise, it is classified as a voiced sound.

In the fourth classification step 904, the voiced sound is classified into the normal voiced sound and the abnormal voiced sound in the frame by the adaptive codebook gain and the pitch delay change. Finally, the class classification is determined as the best average transmission rate versus voice quality.

As such, when the transmission rate and the class are determined, the SMV open loop pitch value is replaced with the pitch delay decoded in the AMR packet, and is similar to the method shown in FIG. 5. In other words, the open circuit analysis process is omitted to obtain a more accurate open circuit pitch delay value when it is smaller than the threshold value when the value is smaller than the threshold value by comparing with the previous past pitch delay value. The determined open circuit pitch delay is used for the closed loop search of the SMV.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and equivalent other embodiments are possible.

Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

According to the present invention, it is possible to maintain the sound quality performance by using the SMV and AMR mutual encoder with a small amount of computation and delay time in using different communication networks.

100: SMV voice encoder
200: AMR voice encoder
400: SMV to AMR LSP Conversion
500: pitch delay conversion unit from SMV to AMR
600: SMV to AMR Voice Encoder
610: SMV parameter decoder
620: LSP Converter
640: pitch delay and adaptive codebook converter
650: fixed codebook converter
660: AMR parameter encoder
800: AMR to SMV Encoder
810: AMR Parameter Decoder
820: LSP Converter
830: rate determiner
840: pitch delay and adaptive codebook converter
850: Fixed Codebook Finder
860: excitation signal converter
870: SMV parameter encoder

Claims (18)

A method for converting a bitstream encoded by a Selectable Mode Vocoder (SMV) speech encoding technique into a bitstream encoded by an Adaptive Multi-Rate (AMR) speech encoding technique,
SMV parameter decoding for generating SMV parameters including SMV Line Spectral Pairs (SMV LSP), SMV pitch delay, SMV rate, adaptive codebook gain, and fixed codebook gain by decoding the bitstream encoded by the SMV speech coding technique. step;
An LSP conversion step of converting the SMV LSP into an AMR LSP to be used in an AMR speech coding technique;
Determine whether the transmission rate is 1 or 1/2, and if it is positive, convert the SMV pitch delay and the adaptive codebook gain into an AMR pitch delay and an adaptive codebook gain, respectively, and if negative, a pitch delay that sets the AMR pitch delay to zero. Adaptive codebook conversion step;
A fixed codebook conversion step of converting an SMV fixed codebook vector into an AMR fixed codebook vector using an SMV adaptive codebook vector and the adaptive codebook gain; And
AMR parameter encoding step of encoding a bitstream encoded by SMV speech coding using the AMR parameters including the AMR LSP, the AMR pitch delay, the codebook gain, and the AMR fixed codebook vector. Characterized in that. The method of claim 1, wherein the LSP conversion step,
Using the SMV LSP corresponding to the fourth subframe of the n th frame among the decoded packets as it is as the fourth subframe of the n th frame of the AMR LSP; And
Interpolating the SMV LSP corresponding to the (n-1) th frame and the fourth subframe of each of the nth frame of the SMV LSP and using the AMR LSP corresponding to the second subframe of the nth frame of the AMR LSP. Characterized in that. The method of claim 1, wherein the pitch delay / adaptive codebook conversion step is performed when the transmission rate is 1 or 1/2,
Predicting the open loop pitch delay past value of the pitch delay by the pitch delay PAMR of the AMR;
Determining whether the difference between the closed loop pitch delay PSMV of the PAMR and SMV is greater than a predetermined threshold, retrieving the pitch delay of AMR if positive, and setting the closed loop pitch delay PSMV to the pitch delay PAMR if negative; And
And searching for an adaptive code book (ACB) using the pitch delay PAMR. The method of claim 1, wherein the fixed codebook conversion step,
Generating an object signal for fixed codebook retrieval from the SMV adaptive codebook vector and adaptive codebook gain;
Calculating a reference vector b (n) using a long _term prediction residual signal res _LTP (n) and a correlation d (n) between the target signal and an impulse response;
Extracting an initial pulse array for position search by selecting from a large value of the reference vector;
Calculating a search criterion Q _k of the initial pulse arrays using the correlation d (n) between the target signal and the impulse response and the autocorrelation matrix Φ ( i , j ) between the impulse response;
The operation of removing one pulse per two adjacent tracks among the pulses belonging to the pulse array is performed to remove two pulses, change the search criteria Q _k to reflect the removed pulses, and maximize the changed search criteria Q _k . A two-pulse rescan step of selecting a pulse from the track to which the removed pulse belongs; And
And a single pulse rescan step of changing a search criterion Q _k by reflecting the removed pulses while removing pulses one by one from the pulse array obtained in the two pulse rescan step, and selecting a pulse that maximizes the changed search criterion Q _k . How to feature. The method of claim 4, wherein the two-pulse rescan step,
The correlation d (n) between the target signal and the impulse response and the autocorrelation matrix Φ ( i , j ) between the impulse response further comprises modifying to include a preselected code to simplify the search. . A method for converting a bitstream encoded by AMR speech encoding into a bitstream encoded by SMV speech encoding,
An AMR parameter decoding step of decoding the bitstream encoded by the AMR speech coding technique to generate AMR parameters including AMR LSP, AMR pitch delay, adaptive codebook gain, and fixed codebook gain;
An LSP conversion step of converting the AMR LSP into an SMV LSP to be used in an SMV speech coding technique;
By using the adaptive codebook gain, the fixed codebook gain, the noise to signal ratio (NSR), and the speech signal reconstructed in the AMR parameter decoding step, the transmission rate to be used for SMV encoding is 1, 1/2, 1 A rate determining step of determining which of / 4 and 1/8;
Generating an excitation signal using a random number generator when the data rate is 1/4 or 1/8;
If the transmission rate is 1 or 1/2, using the pitch delay / adaptive codebook transform operation of converting the AMR pitch delay and the adaptive codebook gain into the SMV pitch delay and the adaptive codebook gain, and the AMR adaptive codebook vector and the adaptive codebook gain, respectively, Performing a fixed codebook transform operation for converting the AMR fixed codebook vector to an SMV fixed codebook vector; And
A SMV parameter encoding step of encoding a bitstream encoded by an AMR speech encoding scheme using an SMV speech encoding scheme using SMV parameters including the SMV LSP, the SMV pitch delay, the codebook gain, and the SMV fixed codebook vector. Characterized in that. The method of claim 6, wherein the LSP conversion step,
And using the AMR LSP corresponding to the fourth subframe of the nth frame among the decoded packets as it is as the fourth subframe of the nth frame of the SMV LSP. The method of claim 6, wherein the rate determination step,
A first classification step of classifying as silence if the noise-to-signal ratio (NSR) of the current frame is greater than or equal to a predetermined threshold or the deviation of the AMR pitch delay is large by referring to the adaptive codebook gain and the fixed codebook gain;
A second classification step of classifying the unvoiced and the voiced sound using the magnitude of the adaptive codebook gain for the current frame that is not silence;
A third classification step of determining whether the class of the previous frame of the frame classified as the voiced sound is unvoiced;
A fourth classification step of determining whether the frame is in a normal state or an abnormal state according to the change of the adaptive codebook gain and the pitch delay in the case where the frame does not change; And
And selecting a transmission rate of the frame differentially according to a classification result of the classification steps. The method of claim 6, wherein the pitch delay / adaptive codebook conversion operation is performed by:
Predicting the open loop pitch delay past value of the pitch delay as the SMV pitch delay;
Determining whether the difference between the SMV pitch delay and the closed loop pitch delay of the AMR is greater than a predetermined threshold, retrieving the pitch delay of the SMV if positive, and setting the closed loop pitch delay to the pitch delay if negative; And
And searching for an adaptive codebook using the pitch delay. An apparatus for converting a bitstream encoded by the SMV speech encoding technique into a bitstream encoded by the AMR speech encoding technique,
A processor for executing instructions executable by a computer; And
Instructions coupled to the processor, including a memory for storing instructions executable by the computer, the instructions executable by the computer,
An SMV parameter decoding operation for decoding the bitstream encoded by the SMV speech coding technique to generate SMV parameters including an SMV LSP, an SMV pitch delay, a transmission rate, an adaptive codebook gain, and a fixed codebook gain;
An LSP conversion operation of converting the SMV LSP into an AMR LSP to be used in an AMR speech coding technique;
Determine whether the transmission rate is 1 or 1/2, and if it is positive, convert the SMV pitch delay and the adaptive codebook gain into an AMR pitch delay and an adaptive codebook gain, respectively, and if negative, a pitch delay that sets the AMR pitch delay to zero. Adaptive codebook conversion operation;
A fixed codebook transform operation of converting an SMV fixed codebook vector into an AMR fixed codebook vector using an SMV adaptive codebook vector and the adaptive codebook gain; And
Perform an AMR parameter encoding operation of encoding a bitstream encoded by an SMV speech encoding technique using AMR parameters including the AMR LSP, the AMR pitch delay, the codebook gain, and the AMR fixed codebook vector; Apparatus characterized in that it is adapted to. The method of claim 10, wherein the LSP conversion operation,
Using the SMV LSP corresponding to the fourth subframe of the nth frame among the decoded packets as it is as the fourth subframe of the nth frame of the AMR LSP; And
Interpolating the SMV LSP corresponding to the (n-1) th frame and the fourth subframe of each of the nth frame of the SMV LSP and using the AMR LSP corresponding to the second subframe of the nth frame of the AMR LSP. Device characterized in that. The method of claim 10, wherein the pitch delay / adaptive codebook conversion operation is performed when the transmission rate is 1 or 1/2,
Predicting the open loop pitch delay past value of the pitch delay by the pitch delay PAMR of the AMR;
Determining whether the difference between the closed loop pitch delay PSMV of the PAMR and the SMV is greater than a predetermined threshold, retrieving the pitch delay of the AMR if positive, and setting the closed loop pitch delay PSMV to the pitch delay PAMR if negative; And
And an ACB search operation for searching an adaptive codebook using the pitch delay PAMR. The method of claim 10, wherein the fixed codebook conversion operation,
Generating an object signal for fixed codebook retrieval from the SMV adaptive codebook vector and the adaptive codebook gain;
Calculating a reference vector b (n) using a long _term prediction residual signal res _LTP (n) and a correlation d (n) between the target signal and an impulse response;
Extracting an initial pulse array for position search by selecting from a large value of the reference vector;
Calculating a search criterion Q _k of the initial pulse arrays using the correlation d (n) between the target signal and the impulse response and the autocorrelation matrix Φ ( i , j ) between the impulse response;
The operation of removing one pulse per two adjacent tracks among the pulses belonging to the pulse array is performed to remove two pulses, change the search criteria Q _k to reflect the removed pulses, and maximize the changed search criteria Q _k . A two-pulse rescan operation for selecting a pulse from the track to which the removed pulse belongs; And
And a single pulse rescan operation to change the search criterion Q _k by reflecting the removed pulses while removing pulses one by one from the pulse array obtained in the two-pulse rescan operation, and selecting a pulse that maximizes the changed search criterion Q _k . Characterized in that the device. The method of claim 13, wherein the two-pulse rescan operation,
The correlation d (n) between the target signal and the impulse response and the autocorrelation matrix Φ ( i , j ) between the impulse response further comprises modifying to include a preselected code to simplify the search. . An apparatus for converting a bitstream encoded by AMR speech encoding into a bitstream encoded by SMV speech encoding,
A processor for executing instructions executable by a computer; And
Instructions coupled to the processor, including a memory for storing instructions executable by the computer, the instructions executable by the computer,
An AMR parameter decoding operation for decoding the bitstream encoded by the AMR speech coding technique to generate AMR parameters including AMR LSP, AMR pitch delay, adaptive codebook gain, and fixed codebook gain;
An LSP conversion operation of converting the AMR LSP into an SMV LSP to be used in an SMV speech coding technique;
By using the adaptive codebook gain, the fixed codebook gain, the noise-to-signal ratio (NSR), and the speech signal reconstructed in the AMR parameter decoding operation, the transmission rates to be used for SMV encoding are 1, 1/2, 1/4, and 1 /. A rate determining operation of determining which of eight;
Generating an excitation signal using a random number generator when the data rate is 1/4 or 1/8;
When the transmission rate is 1 or 1/2, using a pitch delay / adaptive codebook conversion operation and an AMR adaptive codebook vector and the adaptive codebook gain, which convert the AMR pitch delay and adaptive codebook gains into SMV pitch delay and adaptive codebook gains, respectively. Performing a fixed codebook conversion operation of converting the AMR fixed codebook vector into an SMV fixed codebook vector; And
Performing an SMV parameter encoding operation for encoding a bitstream encoded by an AMR speech encoding scheme using an SMV speech encoding scheme using SMV parameters including the SMV LSP, the SMV pitch delay, the codebook gain, and the SMV fixed codebook vector Apparatus characterized in that it is adapted to. The method of claim 15, wherein the LSP conversion operation,
And using the AMR LSP corresponding to the fourth subframe of the nth frame among the decoded packets as it is as the fourth subframe of the nth frame of the SMV LSP. The method of claim 15, wherein the rate determining operation,
A first classification operation of classifying as silence if the noise-to-signal ratio (NSR) of the current frame is greater than or equal to a predetermined threshold or the deviation of the AMR pitch delay is large with reference to the adaptive codebook gain and the fixed codebook gain;
A second classification operation of classifying the unvoiced and the voiced sound using the magnitude of the adaptive codebook gain for the current frame that is not silent;
A third classification operation of determining whether the class of the previous frame of the frame classified as the voiced sound is unvoiced;
A fourth classification operation of determining whether the frame is in a normal state or an abnormal state according to the change of the adaptive codebook gain and the pitch delay in the case of a frame in which no change occurs; And
And selecting a transmission rate of the frame differentially according to a classification result of the classification operations. The method of claim 15, wherein the pitch delay / adaptive codebook conversion operation,
Predicting the open loop pitch delay past value of the pitch delay as the SMV pitch delay;
Determining whether the difference between the SMV pitch delay and the closed loop pitch delay of the AMR is greater than a predetermined threshold, retrieving the pitch delay of the SMV if positive, and setting the closed loop pitch delay to the pitch delay if negative; And
And an ACB search operation for searching for an adaptive codebook using the pitch delay.

Images