Foreword................................................................................................................................................ 4

1....... Scope........................................................................................................................................... 5

2....... References.................................................................................................................................... 5

3....... Definitions, symbols and abbreviations........................................................................................... 6

3.1......... Definitions............................................................................................................................................................................ 6

3.2......... Symbols................................................................................................................................................................................ 7

3.3......... Abbreviations.................................................................................................................................................................... 11

4....... Outline description...................................................................................................................... 11

4.1......... Functional description of audio parts............................................................................................................................ 11

4.2......... Preparation of speech samples........................................................................................................................................ 12

4.2.1........... PCM format conversion............................................................................................................................................ 12

4.3......... Principles of the GSM enhanced full rate speech encoder........................................................................................ 12

4.4......... Principles of the GSM enhanced full rate speech decoder........................................................................................ 14

4.5......... Sequence and subjective importance of encoded parameters................................................................................... 14

5....... Functional description of the encoder............................................................................................ 14

5.1......... Pre‑processing................................................................................................................................................................... 14

5.2......... Linear prediction analysis and quantization................................................................................................................. 15

5.2.1........... Windowing and auto‑correlation computation...................................................................................................... 15

5.2.2........... Levinson‑Durbin algorithm...................................................................................................................................... 16

5.2.3........... LP to LSP conversion................................................................................................................................................ 17

5.2.4........... LSP to LP conversion................................................................................................................................................ 18

5.2.5........... Quantization of the LSP coefficients...................................................................................................................... 19

5.2.6........... Interpolation of the LSPs.......................................................................................................................................... 20

5.3......... Open‑loop pitch analysis................................................................................................................................................. 20

5.4......... Impulse response computation....................................................................................................................................... 21

5.5......... Target signal computation............................................................................................................................................... 21

5.6......... Adaptive codebook search.............................................................................................................................................. 22

5.7......... Algebraic codebook structure and search..................................................................................................................... 23

5.8......... Quantization of the fixed codebook gain...................................................................................................................... 26

5.9......... Memory update................................................................................................................................................................. 27

6....... Functional description of the decoder............................................................................................ 27

6.1......... Decoding and speech synthesis...................................................................................................................................... 27

6.2......... Post‑processing................................................................................................................................................................. 29

6.2.1........... Adaptive post‑filtering............................................................................................................................................... 29

6.2.2........... Up‑scaling.................................................................................................................................................................... 30

7....... Variables, constants and tables in the C‑code of the GSM EFR codec............................................. 30

7.1......... Description of the constants and variables used in the C code................................................................................. 30

8....... Homing sequences....................................................................................................................... 33

8.1......... Functional description...................................................................................................................................................... 33

8.2......... Definitions.......................................................................................................................................................................... 33

8.3......... Encoder homing................................................................................................................................................................ 35

8.4......... Decoder homing................................................................................................................................................................ 35

8.5......... Encoder home state........................................................................................................................................................... 36

8.6......... Decoder home state.......................................................................................................................................................... 37

9....... Bibliography............................................................................................................................... 42

Annex A (informative):....... Change history....................................................................................... 43

 

This Technical Specification has been produced by the 3^rd Generation Partnership Project (3GPP).

The present document describes the detailed mapping between input blocks of 160 speech samples in 13‑bit uniform PCM format to encoded blocks of 244 bits and from encoded blocks of 244 bits to output blocks of 160 reconstructed speech samples within the digital cellular telecommunications system.

The contents of the present document are subject to continuing work within the TSG and may change following formal TSG approval. Should the TSG modify the contents of the present document, it will be re-released by the TSG with an identifying change of release date and an increase in version number as follows:

Version x.y.z

where:

x    the first digit:

1    presented to TSG for information;

2    presented to TSG for approval;

3    or greater indicates TSG approved document under change control.

y    the second digit is incremented for all changes of substance, i.e. technical enhancements, corrections, updates, etc.

z    the third digit is incremented when editorial only changes have been incorporated in the document.

The present document describes the detailed mapping between input blocks of 160 speech samples in 13‑bit uniform PCM format to encoded blocks of 244 bits and from encoded blocks of 244 bits to output blocks of 160 reconstructed speech samples. The sampling rate is 8 000 sample/s leading to a bit rate for the encoded bit stream of 12,2 kbit/s. The coding scheme is the so‑called Algebraic Code Excited Linear Prediction Coder, hereafter referred to as ACELP.

The present document also specifies the conversion between A‑law or m-law (PCS 1900) PCM and 13‑bit uniform PCM. Performance requirements for the audio input and output parts are included only to the extent that they affect the transcoder performance. This part also describes the codec down to the bit level, thus enabling the verification of compliance to the part to a high degree of confidence by use of a set of digital test sequences. These test sequences are described in GSM 06.54 [7] and are available on disks.

In case of discrepancy between the requirements described in the present document and the fixed point computational description (ANSI‑C code) of these requirements contained in GSM 06.53 [6], the description in GSM 06.53 [6] will prevail.

The transcoding procedure specified in the present document is applicable for the enhanced full rate speech traffic channel (TCH) in the GSM system.

In GSM 06.51 [5], a reference configuration for the speech transmission chain of the GSM enhanced full rate (EFR) system is shown. According to this reference configuration, the speech encoder takes its input as a 13‑bit uniform PCM signal either from the audio part of the Mobile Station or on the network side, from the PSTN via an 8‑bit/A‑law or m-law (PCS 1900) to 13‑bit uniform PCM conversion. The encoded speech at the output of the speech encoder is delivered to a channel encoder unit which is specified in GSM 05.03 [3]. In the receive direction, the inverse operations take place.

The following documents contain provisions which, through reference in this text, constitute provisions of the present document.

·       References are either specific (identified by date of publication, edition number, version number, etc.) or non‑specific.

·       For a specific reference, subsequent revisions do not apply.

·       For a non-specific reference, the latest version applies.  In the case of a reference to a 3GPP document (including a GSM document), a non-specific reference implicitly refers to the latest version of that document in the same Release as the present document.

[1]                          GSM 01.04: "Digital cellular telecommunications system (Phase 2+); Abbreviations and acronyms".

[2]                          GSM 03.50: "Digital cellular telecommunications system (Phase 2+); Transmission planning aspects of the speech service in the GSM Public Land Mobile Network (PLMN) system".

[3]                          GSM 05.03: "Digital cellular telecommunications system (Phase 2+); Channel coding".

[4]                          GSM 06.32: "Digital cellular telecommunications system (Phase 2+); Voice Activity Detection (VAD)".

[5]                          GSM 06.51: "Digital cellular telecommunications system (Phase 2+); Enhanced Full Rate (EFR) speech processing functions General description".

[6]                          GSM 06.53: "Digital cellular telecommunications system (Phase 2+); ANSI‑C code for the GSM Enhanced Full Rate (EFR) speech codec".

[7]                          GSM 06.54: "Digital cellular telecommunications system (Phase 2+); Test vectors for the GSM Enhanced Full Rate (EFR) speech codec".

[8]                          ITU‑T Recommendation G.711 (1988): "Coding of analogue signals by pulse code modulation Pulse code modulation (PCM) of voice frequencies".

[9]                          ITU‑T Recommendation G.726: "40, 32, 24, 16 kbit/s adaptive differential pulse code modulation (ADPCM)".

3.1        Definitions

For the purposes of the present document, the following terms and definitions apply:

adaptive codebook: adaptive codebook contains excitation vectors that are adapted for every subframe. The adaptive codebook is derived from the long term filter state. The lag value can be viewed as an index into the adaptive codebook.

adaptive postfilter: this filter is applied to the output of the short term synthesis filter to enhance the perceptual quality of the reconstructed speech. In the GSM enhanced full rate codec, the adaptive postfilter is a cascade of two filters: a formant postfilter and a tilt compensation filter.

algebraic codebook: fixed codebook where algebraic code is used to populate the excitation vectors (innovation vectors).The excitation contains a small number of nonzero pulses with predefined interlaced sets of positions.

closed‑loop pitch analysis: this is the adaptive codebook search, i.e., a process of estimating the pitch (lag) value from the weighted input speech and the long term filter state. In the closed‑loop search, the lag is searched using error minimization loop (analysis‑by‑synthesis). In the GSM enhanced full rate codec, closed‑loop pitch search is performed for every subframe.

direct form coefficients: one of the formats for storing the short term filter parameters. In the GSM enhanced full rate codec, all filters which are used to modify speech samples use direct form coefficients.

fixed codebook: fixed codebook contains excitation vectors for speech synthesis filters. The contents of the codebook are non‑adaptive (i.e., fixed). In the GSM enhanced full rate codec, the fixed codebook is implemented using an algebraic codebook.

fractional lags: set of lag values having sub‑sample resolution. In the GSM enhanced full rate codec a sub‑sample resolution of 1/6th of a sample is used.

frame: time interval equal to 20 ms (160 samples at an 8 kHz sampling rate).

integer lags: set of lag values having whole sample resolution.

interpolating filter: FIR filter used to produce an estimate of sub‑sample resolution samples, given an input sampled with integer sample resolution.

inverse filter: this filter removes the short term correlation from the speech signal. The filter models an inverse frequency response of the vocal tract.

lag: long term filter delay. This is typically the true pitch period, or a multiple or sub‑multiple of it.

Line Spectral Frequencies: (see Line Spectral Pair).

Line Spectral Pair: transformation of LPC parameters. Line Spectral Pairs are obtained by decomposing the inverse filter transfer function A(z) to a set of two transfer functions, one having even symmetry and the other having odd symmetry. The Line Spectral Pairs (also called as Line Spectral Frequencies) are the roots of these polynomials on the z-unit circle).

LP analysis window: for each frame, the short term filter coefficients are computed using the high pass filtered speech samples within the analysis window. In the GSM enhanced full rate codec, the length of the analysis window is 240 samples. For each frame, two asymmetric windows are used to generate two sets of LP coefficients. No samples of the future frames are used (no lookahead).

LP coefficients: Linear Prediction (LP) coefficients (also referred as Linear Predictive Coding (LPC) coefficients) is a generic descriptive term for describing the short term filter coefficients.

open‑loop pitch search: process of estimating the near optimal lag directly from the weighted speech input. This is done to simplify the pitch analysis and confine the closed‑loop pitch search to a small number of lags around the open‑loop estimated lags. In the GSM enhanced full rate codec, open‑loop pitch search is performed every 10 ms.

residual: output signal resulting from an inverse filtering operation.

short term synthesis filter: this filter introduces, into the excitation signal, short term correlation which models the impulse response of the vocal tract.

perceptual weighting filter: this filter is employed in the analysis‑by‑synthesis search of the codebooks. The filter exploits the noise masking properties of the formants (vocal tract resonances) by weighting the error less in regions near the formant frequencies and more in regions away from them.

subframe: time interval equal to 5 ms (40 samples at an 8 kHz sampling rate).

vector quantization: method of grouping several parameters into a vector and quantizing them simultaneously.

zero input response: output of a filter due to past inputs, i.e. due to the present state of the filter, given that an input of zeros is applied.

zero state response: output of a filter due to the present input, given that no past inputs have been applied, i.e., given the state information in the filter is all zeroes.

3.2        Symbols

For the purposes of the present document, the following symbols apply:

                    The inverse filter with unquantized coefficients

                    The inverse filter with quantified coefficients

   The speech synthesis filter with quantified coefficients

                          The unquantized linear prediction parameters (direct form coefficients)

                          The quantified linear prediction parameters

                          The order of the LP model

                   The long‑term synthesis filter

                    The perceptual weighting filter (unquantized coefficients)

                  The perceptual weighting factors

                  Adaptive pre‑filter

                          The nearest integer pitch lag to the closed‑loop fractional pitch lag of the subframe

                           The adaptive pre‑filter coefficient (the quantified pitch gain)

   The formant postfilter

                        Control coefficient for the amount of the formant post‑filtering

                         Control coefficient for the amount of the formant post‑filtering

                  Tilt compensation filter

                          Control coefficient for the amount of the tilt compensation filtering

          A tilt factor, with being the first reflection coefficient

                  The truncated impulse response of the formant postfilter

                        The length of

                    The auto‑correlations of

           The inverse filter (numerator) part of the formant postfilter

     The synthesis filter (denominator) part of the formant postfilter

                     The residual signal of the inverse filter

                   Impulse response of the tilt compensation filter

                The AGC‑controlled gain scaling factor of the adaptive postfilter

                          The AGC factor of the adaptive postfilter

                 Pre‑processing high‑pass filter

,       LP analysis windows

                    Length of the first part of the LP analysis window

                   Length of the second part of the LP analysis window

                   Length of the first part of the LP analysis window

                  Length of the second part of the LP analysis window

                 The auto‑correlations of the windowed speech

                 Lag window for the auto‑correlations (60 Hz bandwidth expansion)

                         The bandwidth expansion in Hz

                          The sampling frequency in Hz

              The modified (bandwidth expanded) auto‑correlations

               The prediction error in the ith iteration of the Levinson algorithm

                          The ith reflection coefficient

                      The jth direct form coefficient in the ith iteration of the Levinson algorithm

                 Symmetric LSF polynomial

                 Antisymmetric LSF polynomial

                  Polynomial  with root  eliminated

                  Polynomial  with root  eliminated

                          The line spectral pairs (LSPs) in the cosine domain

                           An LSP vector in the cosine domain

                      The quantified LSP vector at the ith subframe of the frame n

                         The line spectral frequencies (LSFs)

                  A th order Chebyshev polynomial

        The coefficients of the polynomials and

      The coefficients of the polynomials  and

                     The coefficients of either  or

                  Sum polynomial of the Chebyshev polynomials

                           Cosine of angular frequency

                        Recursion coefficients for the Chebyshev polynomial evaluation

                        The line spectral frequencies (LSFs) in Hz

The vector representation of the LSFs in Hz

,   The mean‑removed LSF vectors at frame n

,      The LSF prediction residual vectors at frame n

                     The predicted LSF vector at frame n

       The quantified second residual vector at the past frame

                         The quantified LSF vector at quantization index k

                   The LSP quantization error

LSP‑quantization weighting factors

                        The distance between the line spectral frequencies  and

                   The impulse response of the weighted synthesis filter

                      The correlation maximum of open‑loop pitch analysis at delay k

   The correlation maxima at delays

The normalized correlation maxima  and the corresponding delays

 The weighted synthesis filter

          The numerator of the perceptual weighting filter

    The denominator of the perceptual weighting filter

                          The nearest integer to the fractional pitch lag of the previous (1st or 3rd) subframe

                    The windowed speech signal

                 The weighted speech signal

                    Reconstructed speech signal

                   The gain‑scaled post‑filtered signal

                Post‑filtered speech signal (before scaling)

                   The target signal for adaptive codebook search

,         The target signal for algebraic codebook search

           The LP residual signal

                    The fixed codebook vector

                   The adaptive codebook vector

    The filtered adaptive codebook vector

                  The past filtered excitation

                    The excitation signal

                     The emphasized adaptive codebook vector

                   The gain‑scaled emphasized excitation signal

                       The best open‑loop lag

                     Minimum lag search value

                    Maximum lag search value

                   Correlation term to be maximized in the adaptive codebook search

                       The FIR filter for interpolating the normalized correlation term

                 The interpolated value of  for the integer delay k and fraction t

                       The FIR filter for interpolating the past excitation signal  to yield the adaptive codebook vector

                       Correlation term to be maximized in the algebraic codebook search at index k

                        The correlation in the numerator of  at index k

                     The energy in the denominator of  at index k

          The correlation between the target signal  and the impulse response , i.e., backward filtered target

                         The lower triangular Toepliz convolution matrix with diagonal  and lower diagonals

         The matrix of correlations of

                    The elements of the vector d

                 The elements of the symmetric matrix

                         The innovation vector

                         The correlation in the numerator of

                         The position of the i th pulse

                         The amplitude of the i th pulse

                       The number of pulses in the fixed codebook excitation

                      The energy in the denominator of

         The normalized long‑term prediction residual

                    The sum of the normalized  vector and normalized long‑term prediction residual

                  The sign signal for the algebraic codebook search

                 Sign extended backward filtered target

               The modified elements of the matrix , including sign information

,             The fixed codebook vector convolved with

                  The mean‑removed innovation energy (in dB)

                          The mean of the innovation energy

                  The predicted energy

    The MA prediction coefficients

                  The quantified prediction error at subframe k

                       The mean innovation energy

                   The prediction error of the fixed‑codebook gain quantization

                      The quantization error of the fixed‑codebook gain quantization

                     The states of the synthesis filter

                 The perceptually weighted error of the analysis‑by‑synthesis search

                          The gain scaling factor for the emphasized excitation

                         The fixed‑codebook gain

                         The predicted fixed‑codebook gain

                          The quantified fixed codebook gain

                         The adaptive codebook gain

                        The quantified adaptive codebook gain

  A correction factor between the gain  and the estimated one

                     The optimum value for

                       Gain scaling factor

3.3        Abbreviations

For the purposes of the present document, the following abbreviations apply. Further GSM related abbreviations may be found in GSM 01.04 [1].

ACELP                  Algebraic Code Excited Linear Prediction

AGC                      Adaptive Gain Control

CELP                     Code Excited Linear Prediction

FIR                        Finite Impulse Response

ISPP                       Interleaved Single‑Pulse Permutation

LP                          Linear Prediction

LPC                       Linear Predictive Coding

LSF                        Line Spectral Frequency

LSP                        Line Spectral Pair

LTP                        Long Term Predictor (or Long Term Prediction)

MA                        Moving Average

The present document is structured as follows.

Subclause 4.1 contains a functional description of the audio parts including the A/D and D/A functions. Subclause 4.2 describes the conversion between 13‑bit uniform and 8‑bit A‑law or m-law (PCS 1900) samples. Subclauses 4.3 and 4.4 present a simplified description of the principles of the GSM EFR encoding and decoding process respectively. In clause 4.5, the sequence and subjective importance of encoded parameters are given.

Clause 5 presents the functional description of the GSM EFR encoding, whereas clause 6 describes the decoding procedures. Clause 7 describes variables, constants and tables of the C‑code of the GSM EFR codec.

4.1        Functional description of audio parts

The analogue‑to‑digital and digital‑to‑analogue conversion will in principle comprise the following elements:

1)   analogue to uniform digital PCM:

-     microphone;

‑     input level adjustment device;

‑     input anti‑aliasing filter;

‑     sample‑hold device sampling at 8 kHz;

‑     analogue‑to‑uniform digital conversion to 13‑bit representation.

      The uniform format shall be represented in two's complement.

2)   uniform digital PCM to analogue:

‑     conversion from 13‑bit/8 kHz uniform PCM to analogue;

‑     a hold device;

‑     reconstruction filter including x/sin( x ) correction;

‑     output level adjustment device;

‑     earphone or loudspeaker.

      In the terminal equipment, the A/D function may be achieved either:

‑     by direct conversion to 13‑bit uniform PCM format;

‑     or by conversion to 8‑bit/A‑law or m-law (PCS 1900) compounded format, based on a standard A‑law or m-law (PCS 1900) codec/filter according to ITU‑T Recommendations G.711 [8] and G.714, followed by the 8‑bit to 13‑bit conversion as specified in clause 4.2.1.

For the D/A operation, the inverse operations take place.

In the latter case it should be noted that the specifications in ITU‑T G.714 (superseded by G.712) are concerned with PCM equipment located in the central parts of the network. When used in the terminal equipment, the present document does not on its own ensure sufficient out‑of‑band attenuation. The specification of out‑of‑band signals is defined in GSM 03.50 [2] in clause 2.

4.2        Preparation of speech samples

The encoder is fed with data comprising of samples with a resolution of 13 bits left justified in a 16‑bit word. The three least significant bits are set to '0'. The decoder outputs data in the same format. Outside the speech codec further processing must be applied if the traffic data occurs in a different representation.

4.2.1       PCM format conversion

The conversion between 8‑bit A‑Law or m-law (PCS 1900) compressed data and linear data with 13‑bit resolution at the speech encoder input shall be as defined in ITU‑T Rec. G.711 [8].

ITU‑T Recommendation G.711 [8] specifies the A‑Law or m-law (PCS 1900) to linear conversion and vice versa by providing table entries. Examples on how to perform the conversion by fixed‑point arithmetic can be found in ITU‑T Recommendation G.726 [9]. Subclause 4.2.1 of G.726 [9] describes A‑Law and m-law (PCS 1900) to linear expansion and clause 4.2.7 of G.726 [9] provides a solution for linear to A‑Law and m-law (PCS 1900) compression.

4.3        Principles of the GSM enhanced full rate speech encoder

The codec is based on the code‑excited linear predictive (CELP) coding model. A 10th order linear prediction (LP), or short‑term, synthesis filter is used which is given by:

                                                                                                                             (1)

where  are the (quantified) linear prediction (LP) parameters, and  is the predictor order. The long‑term, or pitch, synthesis filter is given by:

                                                                                                                                                           (2)

where  is the pitch delay and  is the pitch gain. The pitch synthesis filter is implemented using the so‑called adaptive codebook approach.

The CELP speech synthesis model is shown in figure 2. In this model, the excitation signal at the input of the short‑term LP synthesis filter is constructed by adding two excitation vectors from adaptive and fixed (innovative) codebooks. The speech is synthesized by feeding the two properly chosen vectors from these codebooks through the short‑term synthesis filter. The optimum excitation sequence in a codebook is chosen using an analysis‑by‑synthesis search procedure in which the error between the original and synthesized speech is minimized according to a perceptually weighted distortion measure.

The perceptual weighting filter used in the analysis‑by‑synthesis search technique is given by:

                                                                                                                                                             (3)

where  is the unquantized LP filter and  are the perceptual weighting factors. The values  and  are used. The weighting filter uses the unquantized LP parameters while the formant synthesis filter uses the quantified ones.

The coder operates on speech frames of 20 ms corresponding to 160 samples at the sampling frequency of 8 000 sample/s. At each 160 speech samples, the speech signal is analysed to extract the parameters of the CELP model (LP filter coefficients, adaptive and fixed codebooks' indices and gains). These parameters are encoded and transmitted. At the decoder, these parameters are decoded and speech is synthesized by filtering the reconstructed excitation signal through the LP synthesis filter.

The signal flow at the encoder is shown in figure 3. LP analysis is performed twice per frame. The two sets of LP parameters are converted to line spectrum pairs (LSP) and jointly quantified using split matrix quantization (SMQ) with 38 bits. The speech frame is divided into 4 subframes of 5 ms each (40 samples). The adaptive and fixed codebook parameters are transmitted every subframe. The two sets of quantified and unquantized LP filters are used for the second and fourth subframes while in the first and third subframes interpolated LP filters are used (both quantified and unquantized). An open‑loop pitch lag is estimated twice per frame (every 10 ms) based on the perceptually weighted speech signal.

Then the following operations are repeated for each subframe:

      The target signal  is computed by filtering the LP residual through the weighted synthesis filter  with the initial states of the filters having been updated by filtering the error between LP residual and excitation (this is equivalent to the common approach of subtracting the zero input response of the weighted synthesis filter from the weighted speech signal).

      The impulse response,  of the weighted synthesis filter is computed.

      Closed‑loop pitch analysis is then performed (to find the pitch lag and gain), using the target  and impulse response , by searching around the open‑loop pitch lag. Fractional pitch with 1/6th of a sample resolution is used. The pitch lag is encoded with 9 bits in the first and third subframes and relatively encoded with 6 bits in the second and fourth subframes.

      The target signal  is updated by removing the adaptive codebook contribution (filtered adaptive codevector), and this new target, , is used in the fixed algebraic codebook search (to find the optimum innovation). An algebraic codebook with 35 bits is used for the innovative excitation.

      The gains of the adaptive and fixed codebook are scalar quantified with 4 and 5 bits respectively (with moving average (MA) prediction applied to the fixed codebook gain).

      Finally, the filter memories are updated (using the determined excitation signal) for finding the target signal in the next subframe.

The bit allocation of the codec is shown in table 1. In each 20 ms speech frame, 244 bits are produced, corresponding to a bit rate of 12.2 kbit/s. More detailed bit allocation is available in table 6. Note that the most significant bits (MSB) are always sent first.

Table 1: Bit allocation of the 12.2 kbit/s coding algorithm for 20 ms frame

 

4.4        Principles of the GSM enhanced full rate speech decoder

The signal flow at the decoder is shown in figure 4. At the decoder, the transmitted indices are extracted from the received bitstream. The indices are decoded to obtain the coder parameters at each transmission frame. These parameters are the two LSP vectors, the 4 fractional pitch lags, the 4 innovative codevectors, and the 4 sets of pitch and innovative gains. The LSP vectors are converted to the LP filter coefficients and interpolated to obtain LP filters at each subframe. Then, at each 40‑sample subframe:

‑     the excitation is constructed by adding the adaptive and innovative codevectors scaled by their respective gains;

‑     the speech is reconstructed by filtering the excitation through the LP synthesis filter.

Finally, the reconstructed speech signal is passed through an adaptive postfilter.

4.5        Sequence and subjective importance of encoded parameters

The encoder will produce the output information in a unique sequence and format, and the decoder must receive the same information in the same way. In table 6, the sequence of output bits s1 to s244 and the bit allocation for each parameter is shown.

The different parameters of the encoded speech and their individual bits have unequal importance with respect to subjective quality. Before being submitted to the channel encoding function the bits have to be rearranged in the sequence of importance as given in table 6 in 05.03 [3].

In this clause, the different functions of the encoder represented in figure 3 are described.

5.1        Pre‑processing

Two pre‑processing functions are applied prior to the encoding process: high‑pass filtering and signal down‑scaling.

Down‑scaling consists of dividing the input by a factor of 2 to reduce the possibility of overflows in the fixed‑point implementation.

The high‑pass filter serves as a precaution against undesired low frequency components. A filter with a cut off frequency of 80 Hz is used, and it is given by:

                                                                               (4)

Down‑scaling and high‑pass filtering are combined by dividing the coefficients at the numerator of  by 2.

5.2        Linear prediction analysis and quantization

Short‑term prediction, or linear prediction (LP), analysis is performed twice per speech frame using the auto‑correlation approach with 30 ms asymmetric windows. No lookahead is used in the auto‑correlation computation.

The auto‑correlations of windowed speech are converted to the LP coefficients using the Levinson‑Durbin algorithm. Then the LP coefficients are transformed to the Line Spectral Pair (LSP) domain for quantization and interpolation purposes. The interpolated quantified and unquantized filter coefficients are converted back to the LP filter coefficients (to construct the synthesis and weighting filters at each subframe).

5.2.1       Windowing and auto‑correlation computation

LP analysis is performed twice per frame using two different asymmetric windows. The first window has its weight concentrated at the second subframe and it consists of two halves of Hamming windows with different sizes. The window is given by:

                                      (5)

The values  and  are used. The second window has its weight concentrated at the fourth subframe and it consists of two parts: the first part is half a Hamming window and the second part is a quarter of a cosine function cycle. The window is given by:

                                     (6)

where the values  and  are used.

Note that both LP analyses are performed on the same set of speech samples. The windows are applied to 80 samples from past speech frame in addition to the 160 samples of the present speech frame. No samples from future frames are used (no lookahead). A diagram of the two LP analysis windows is depicted below.

Figure 1: LP analysis windows

The auto‑correlations of the windowed speech , are computed by:

                                                                                                         (7)

and a 60 Hz bandwidth expansion is used by lag windowing the auto‑correlations using the window:

                                                                                                 (8)

where  Hz is the bandwidth expansion and  Hz is the sampling frequency. Further,  is multiplied by the white noise correction factor 1.0001 which is equivalent to adding a noise floor at ‑40 dB.

5.2.2       Levinson‑Durbin algorithm

The modified auto‑correlations  and  are used to obtain the direct form LP filter coefficients  by solving the set of equations.

                                                                                                           (9)

The set of equations in (9) is solved using the Levinson‑Durbin algorithm. This algorithm uses the following recursion:

The final solution is given as .

The LP filter coefficients are converted to the line spectral pair (LSP) representation for quantization and interpolation purposes. The conversions to the LSP domain and back to the LP filter coefficient domain are described in the next clause.

5.2.3       LP to LSP conversion

The LP filter coefficients , are converted to the line spectral pair (LSP) representation for quantization and interpolation purposes. For a 10th order LP filter, the LSPs are defined as the roots of the sum and difference polynomials:

                                                                                                                                    (10)

and

                                                                     ,                                                              (11)

respectively. The polynomial  and  are symmetric and anti‑symmetric, respectively. It can be proven that all roots of these polynomials are on the unit circle and they alternate each other.  has a root  and  has a root . To eliminate these two roots, we define the new polynomials:

                                                                                                                                          (12)

and

                                                                                                                                        (13)

Each polynomial has 5 conjugate roots on the unit circle , therefore, the polynomials can be written as

                                                                                                                            (14)

and

                                                                 ,                                                          (15)

where  with  being the line spectral frequencies (LSF) and they satisfy the ordering property . We refer to  as the LSPs in the cosine domain.

Since both polynomials  and  are symmetric only the first 5 coefficients of each polynomial need to be computed. The coefficients of these polynomials are found by the recursive relations (for  to 4):

                                                                                                                       (16)

where  is the predictor order.

The LSPs are found by evaluating the polynomials  and  at 60 points equally spaced between 0 and  and checking for sign changes. A sign change signifies the existence of a root and the sign change interval is then divided 4 times to better track the root. The Chebyshev polynomials are used to evaluate  and . In this method the roots are found directly in the cosine domain . The polynomials  or  evaluated at  can be written as:

with:

                 ,          (17)

where  is the th order Chebyshev polynomial, and  are the coefficients of either  or , computed using the equations in (16). The polynomial  is evaluated at a certain value of  using the recursive relation:

with initial values  and  The details of the Chebyshev polynomial evaluation method are found in P. Kabal and R.P. Ramachandran [6].

5.2.4       LSP to LP conversion

Once the LSPs are quantified and interpolated, they are converted back to the LP coefficient domain . The conversion to the LP domain is done as follows. The coefficients of  or  are found by expanding equations (14) and (15) knowing the quantified and interpolated LSPs . The following recursive relation is used to compute :

with initial values  and . The coefficients  are computed similarly by replacing  by .

Once the coefficients  and  are found,  and  are multiplied by  and , respectively, to obtain  and ; that is:

                                                                                                       (18)

Finally the LP coefficients are found by:

                                                                                           (19)

This is directly derived from the relation , and considering the fact that  and  are symmetric and anti‑symmetric polynomials, respectively.

5.2.5       Quantization of the LSP coefficients

The two sets of LP filter coefficients per frame are quantified using the LSP representation in the frequency domain; that is:

                                                                                                                          (20)

where  are the line spectral frequencies (LSF) in Hz [0,4000] and  is the sampling frequency. The LSF vector is given by , with t denoting transpose.

A 1st order MA prediction is applied, and the two residual LSF vectors are jointly quantified using split matrix quantization (SMQ). The prediction and quantization are performed as follows. Let  and  denote the mean‑removed LSF vectors at frame . The prediction residual vectors  and  are given by:

                                                                                                                          (21)

where  is the predicted LSF vector at frame . First order moving‑average (MA) prediction is used where:

                                                                                                                                                   (22)

where  is the quantified second residual vector at the past frame.

The two LSF residual vectors  and  are jointly quantified using split matrix quantization (SMQ). The matrix  is split into 5 submatrices of dimension 2 x 2 (two elements from each vector). For example, the first submatrix consists of the elements , and . The 5 submatrices are quantified with 7, 8, 8+1, 8, and 6 bits, respectively. The third submatrix uses a 256‑entry signed codebook (8‑bit index plus 1‑bit sign).

A weighted LSP distortion measure is used in the quantization process. In general, for an input LSP vector  and a quantified vector at index , , the quantization is performed by finding the index  which minimizes:

                                                                                                                                            (23)

The weighting factors , are given by

     

                                                                                                                                                                                                  (24)

where  with  and . Here, two sets of weighting coefficients are computed for the two LSF vectors. In the quantization of each submatrix, two weighting coefficients from each set are used with their corresponding LSFs.

5.2.6       Interpolation of the LSPs

The two sets of quantified (and unquantized) LP parameters are used for the second and fourth subframes whereas the first and third subframes use a linear interpolation of the parameters in the adjacent subframes. The interpolation is performed on the LSPs in the  domain. Let  be the LSP vector at the 4th subframe of the present frame ,  be the LSP vector at the 2nd subframe of the present frame , and  the LSP vector at the 4th subframe of the past frame . The interpolated LSP vectors at the 1st and 3rd subframes are given by:

                                                                                                                                         (25)

The interpolated LSP vectors are used to compute a different LP filter at each subframe (both quantified and unquantized coefficients) using the LSP to LP conversion method described in clause 5.2.4.

5.3        Open‑loop pitch analysis

Open‑loop pitch analysis is performed twice per frame (each 10 ms) to find two estimates of the pitch lag in each frame. This is done in order to simplify the pitch analysis and confine the closed‑loop pitch search to a small number of lags around the open‑loop estimated lags.

Open‑loop pitch estimation is based on the weighted speech signal  which is obtained by filtering the input speech signal through the weighting filter . That is, in a subframe of size , the weighted speech is given by:

                                                      (26)

Open‑loop pitch analysis is performed as follows. In the first step, 3 maxima of the correlation:

                                                                                                                                                 (27)

are found in the three ranges:

The retained maxima , are normalized by dividing by , respectively. The normalized maxima and corresponding delays are denoted by . The winner, , among the three normalized correlations is selected by favouring the delays with the values in the lower range. This is performed by weighting the normalized correlations corresponding to the longer delays. The best open‑loop delay  is determined as follows:

This procedure of dividing the delay range into 3 clauses and favouring the lower clauses is used to avoid choosing pitch multiples.

5.4        Impulse response computation

The impulse response, , of the weighted synthesis filter  is computed each subframe. This impulse response is needed for the search of adaptive and fixed codebooks. The impulse response  is computed by filtering the vector of coefficients of the filter  extended by zeros through the two filters  and .

5.5        Target signal computation

The target signal for adaptive codebook search is usually computed by subtracting the zero input response of the weighted synthesis filter  from the weighted speech signal . This is performed on a subframe basis.

An equivalent procedure for computing the target signal, which is used in the present document, is the filtering of the LP residual signal  through the combination of synthesis filter  and the weighting filter . After determining the excitation for the subframe, the initial states of these filters are updated by filtering the difference between the LP residual and excitation. The memory update of these filters is explained in clause 5.9.

The residual signal  which is needed for finding the target vector is also used in the adaptive codebook search to extend the past excitation buffer. This simplifies the adaptive codebook search procedure for delays less than the subframe size of 40 as will be explained in the next clause. The LP residual is given by:

                                                                                                                                  (28)

5.6        Adaptive codebook search

Adaptive codebook search is performed on a subframe basis. It consists of performing closed‑loop pitch search, and then computing the adaptive codevector by interpolating the past excitation at the selected fractional pitch lag.

The adaptive codebook parameters (or pitch parameters) are the delay and gain of the pitch filter. In the adaptive codebook approach for implementing the pitch filter, the excitation is repeated for delays less than the subframe length. In the search stage, the excitation is extended by the LP residual to simplify the closed‑loop search.

In the first and third subframes, a fractional pitch delay is used with resolutions: 1/6 in the range  and integers only in the range [95, 143]. For the second and fourth subframes, a pitch resolution of 1/6 is always used in the range , where  is nearest integer to the fractional pitch lag of the previous (1st or 3rd) subframe, bounded by 18...143.

Closed‑loop pitch analysis is performed around the open‑loop pitch estimates on a subframe basis. In the first (and third) subframe the range , bounded by 18...143, is searched. For the other subframes, closed‑loop pitch analysis is performed around the integer pitch selected in the previous subframe, as described above. The pitch delay is encoded with 9 bits in the first and third subframes and the relative delay of the other subframes is encoded with 6 bits.

The closed‑loop pitch search is performed by minimizing the mean‑square weighted error between the original and synthesized speech. This is achieved by maximizing the term:

                                                                                                                                      (29)

where  is the target signal and  is the past filtered excitation at delay  (past excitation convolved with ). Note that the search range is limited around the open‑loop pitch as explained earlier.

The convolution  is computed for the first delay t_min in the searched range, and for the other delays in the search range , it is updated using the recursive relation:

                                                                 ,                                                          (30)

where , is the excitation buffer. Note that in search stage, the samples , are not known, and they are needed for pitch delays less than 40. To simplify the search, the LP residual is copied to  in order to make the relation in equation (30) valid for all delays.

Once the optimum integer pitch delay is determined, the fractions from  to  with a step of  around that integer are tested. The fractional pitch search is performed by interpolating the normalized correlation in equation (29) and searching for its maximum. The interpolation is performed using an FIR filter  based on a Hamming windowed  function truncated at ± 23 and padded with zeros at ± 24 (). The filter has its cut‑off frequency (‑3 dB) at 3 600 Hz in the over‑sampled domain. The interpolated values of  for the fractions  to  are obtained using the interpolation formula:

                                (31)

where corresponds to the fractions 0, , , , , and , respectively. Note that it is necessary to compute the correlation terms in equation (29) using a range  to allow for the proper interpolation.

Once the fractional pitch lag is determined, the adaptive codebook vector  is computed by interpolating the past excitation signal  at the given integer delay  and phase (fraction) :

  (32)

The interpolation filter  is based on a Hamming windowed  function truncated at ± 59 and padded with zeros at ± 60 (). The filter has a cut‑off frequency (‑3 dB) at 3 600 Hz in the over‑sampled domain.

The adaptive codebook gain is then found by:

                                                                                    (33)

where  is the filtered adaptive codebook vector (zero state response of  to ).

The computed adaptive codebook gain is quantified using 4‑bit non‑uniform scalar quantization in the range [0.0,1.2].

5.7        Algebraic codebook structure and search

The algebraic codebook structure is based on interleaved single‑pulse permutation (ISPP) design. In this codebook, the innovation vector contains 10 non‑zero pulses. All pulses can have the amplitudes +1 or ‑1. The 40 positions in a subframe are divided into 5 tracks, where each track contains two pulses, as shown in table 2.

Table 2: Potential positions of individual pulses in the algebraic codebook

 

Each two pulse positions in one track are encoded with 6 bits (total of 30 bits, 3 bits for the position of every pulse), and the sign of the first pulse in the track is encoded with 1 bit (total of 5 bits).

For two pulses located in the same track, only one sign bit is needed. This sign bit indicates the sign of the first pulse. The sign of the second pulse depends on its position relative to the first pulse. If the position of the second pulse is smaller, then it has opposite sign, otherwise it has the same sign than in the first pulse.

All the 3‑bit pulse positions are Gray coded in order to improve robustness against channel errors. This gives a total of 35 bits for the algebraic code.

The algebraic codebook is searched by minimizing the mean square error between the weighted input speech and the weighted synthesized speech. The target signal used in the closed‑loop pitch search is updated by subtracting the adaptive codebook contribution. That is:

                                                                                                                  (34)

where  is the filtered adaptive codebook vector and  is the quantified adaptive codebook gain. If  is the algebraic codevector at index , then the algebraic codebook is searched by maximizing the term:

                                                                                                                                                   (35)

where  is the correlation between the target signal  and the impulse response ,  is a the lower triangular Toepliz convolution matrix with diagonal  and lower diagonals , and  is the matrix of correlations of . The vector  (backward filtered target) and the matrix  are computed prior to the codebook search. The elements of the vector  are computed by

                                                                                                                 (36)

and the elements of the symmetric matrix  are computed by:

                                                                                                                  (37)

The algebraic structure of the codebooks allows for very fast search procedures since the innovation vector  contains only a few nonzero pulses. The correlation in the numerator of Equation (35) is given by:

                                                                                                                                                              (38)

where  is the position of the th pulse,  is its amplitude, and  is the number of pulses (N_p = 10 ). The energy in the denominator of equation (35) is given by:

                                                                                         (39)

To simplify the search procedure, the pulse amplitudes are preset by the mere quantization of an appropriate signal. In this case the signal , which is a sum of the normalized  vector and normalized long‑term prediction residual :

                                                  (40)

is used. This is simply done by setting the amplitude of a pulse at a certain position equal to the sign of  at that position. The simplification proceeds as follows (prior to the codebook search). First, the sign signal  and the signal  are computed. Second, the matrix  is modified by including the sign information; that is, . The correlation in equation (38) is now given by:

                                                                                                                                                                 (41)

and the energy in equation (39) is given by:

                                                                                                           (42)

Having preset the pulse amplitudes, as explained above, the optimal pulse positions are determined using an efficient non‑exhaustive analysis‑by‑synthesis search technique. In this technique, the term in equation (35) is tested for a small percentage of position combinations.

First, for each of the five tracks the pulse positions with maximum absolute values of  are searched. From these the global maximum value for all the pulse positions is selected. The first pulse i0 is always set into the position corresponding to the global maximum value.

Next, four iterations are carried out. During each iteration the position of pulse i1 is set to the local maximum of one track. The rest of the pulses are searched in pairs by sequentially searching each of the pulse pairs {i2,i3}, {i4,i5}, {i6,i7} and {i8,i9} in nested loops. Every pulse has 8 possible positions, i.e., there are four 8x8‑loops, resulting in 256 different combinations of pulse positions for each iteration.

In each iteration all the 9 pulse starting positions are cyclically shifted, so that the pulse pairs are changed and the pulse i1 is placed in a local maximum of a different track. The rest of the pulses are searched also for the other positions in the tracks. At least one pulse is located in a position corresponding to the global maximum and one pulse is located in a position corresponding to one of the 4 local maxima.

A special feature incorporated in the codebook is that the selected codevector is filtered through an adaptive pre‑filter  which enhances special spectral components in order to improve the synthesized speech quality. Here the filter  is used, where  is the nearest integer pitch lag to the closed‑loop fractional pitch lag of the subframe, and  is a pitch gain. In the present document,  is given by the quantified pitch gain bounded by [0.0,1.0]. Note that prior to the codebook search, the impulse response  must include the pre‑filter . That is, .

The fixed codebook gain is then found by:

                                                                                                                                                                              (43)

where  is the target vector for fixed codebook search and  is the fixed codebook vector convolved with ,

                                                                                                                     (44)

5.8        Quantization of the fixed codebook gain

The fixed codebook gain quantization is performed using MA prediction with fixed coefficients. The 4th order MA prediction is performed on the innovation energy as follows. Let  be the mean‑removed innovation energy (in dB) at subframe , and given by:

                                                                                                                       (45)

where  is the subframe size,  is the fixed codebook excitation, and  dB is the mean of the innovation energy. The predicted energy is given by:

                                                                              ,                                                                      (46)

where  are the MA prediction coefficients, and  is the quantified prediction error at subframe . The predicted energy is used to compute a predicted fixed‑codebook gain  as in equation (45) (by substituting  by  and  by ). This is done as follows. First, the mean innovation energy is found by:

                                                                                                                                          (47)

and then the predicted gain  is found by:

                                                                                                                                                (48)

A correction factor between the gain  and the estimated one  is given by:

                                                                                      .                                                                               (49)

Note that the prediction error is given by:

                                                                                                                          (50)

The correction factor  is quantified using a 5‑bit codebook. The quantization table search is performed by minimizing the error:

                                                                                                                                                       (51)

Once the optimum value  is chosen, the quantified fixed codebook gain is given by .

5.9        Memory update

An update of the states of the synthesis and weighting filters is needed in order to compute the target signal in the next subframe.

After the two gains are quantified, the excitation signal, , in the present subframe is found by:

                                                                                                               (52)

where  and  are the quantified adaptive and fixed codebook gains, respectively,  the adaptive codebook vector (interpolated past excitation), and  is the fixed codebook vector (algebraic code including pitch sharpening). The states of the filters can be updated by filtering the signal  (difference between residual and excitation) through the filters  and  for the 40‑sample subframe and saving the states of the filters. This would require 3 filterings. A simpler approach which requires only one filtering is as follows. The local synthesized speech, , is computed by filtering the excitation signal through . The output of the filter due to the input  is equivalent to . So the states of the synthesis filter  are given by . Updating the states of the filter  can be done by filtering the error signal  through this filter to find the perceptually weighted error . However, the signal  can be equivalently found by:

                                                                                                                           (53)

Since the signals , and  are available, the states of the weighting filter are updated by computing  as in equation (53) for . This saves two filterings.

The function of the decoder consists of decoding the transmitted parameters (LP parameters, adaptive codebook vector, adaptive codebook gain, fixed codebook vector, fixed codebook gain) and performing synthesis to obtain the reconstructed speech. The reconstructed speech is then post‑filtered and upscaled. The signal flow at the decoder is shown in figure 4.

6.1        Decoding and speech synthesis

The decoding process is performed in the following order:

Decoding of LP filter parameters: The received indices of LSP quantization are used to reconstruct the two quantified LSP vectors. The interpolation described in clause 5.2.6 is performed to obtain 4 interpolated LSP vectors (corresponding to 4 subframes). For each subframe, the interpolated LSP vector is converted to LP filter coefficient domain, which is used for synthesizing the reconstructed speech in the subframe.

The following steps are repeated for each subframe:

1)   Decoding of the adaptive codebook vector: The received pitch index (adaptive codebook index) is used to find the integer and fractional parts of the pitch lag. The adaptive codebook vector  is found by interpolating the past excitation  (at the pitch delay) using the FIR filter described in clause 5.6.

2)   Decoding of the adaptive codebook gain: The received index is used to readily find the quantified adaptive codebook gain, from the quantization table.

3)   Decoding of the innovative codebook vector: The received algebraic codebook index is used to extract the positions and amplitudes (signs) of the excitation pulses and to find the algebraic codevector. If the integer part of the pitch lag is less than the subframe size 40, the pitch sharpening procedure is applied which translates into modifying  by, where  is the decoded pitch gain, , bounded by [0.0,1.0].

4)   Decoding of the fixed codebook gain: The received index gives the fixed codebook gain correction factor . The estimated fixed codebook gain  is found as described in clause 5.7. First, the predicted energy is found by:

                                                                                                                                                    (54)

      and then the mean innovation energy is found by:

                                                                                                                                           (55)

      The predicted gain is found by:

                                                                           .                                                                    (56)

      The quantified fixed codebook gain is given by:

                                                                                                                                                                         (57)

5)   Computing the reconstructed speech: The excitation at the input of the synthesis filter is given by:

                                                                                                                                                 (58)

      Before the speech synthesis, a post‑processing of excitation elements is performed. This means that the total excitation is modified by emphasizing the contribution of the adaptive codebook vector:

                                                                                                               (59)

      Adaptive gain control (AGC) is used to compensate for the gain difference between the non‑emphasized excitation  and emphasized excitation  The gain scaling factor h for the emphasized excitation is computed by:

                                                                                                                             (60)

      The gain‑scaled emphasized excitation signal  is given by:

                                                                                                                                                                      (61)

      The reconstructed speech for the subframe of size 40 is given by:

                                                                                                 (62)

      where  are the interpolated LP filter coefficients.

The synthesized speech  is then passed through an adaptive postfilter which is described in the following clause.

6.2        Post‑processing

Post‑processing consists of two functions: adaptive post‑filtering and signal up‑scaling.

6.2.1       Adaptive post‑filtering

The adaptive postfilter is the cascade of two filters: a formant postfilter, and a tilt compensation filter. The postfilter is updated every subframe of 5 ms.

The formant postfilter is given by:

                                                                                                                                                          (63)

where  is the received quantified (and interpolated) LP inverse filter (LP analysis is not performed at the decoder), and the factors  and  control the amount of the formant post‑filtering.

Finally, the filter  compensates for the tilt in the formant postfilter  and is given by:

                                                                                                                                                           (64)

where  is a tilt factor, with  being the first reflection coefficient calculated on the truncated impulse response, , of the filter.  is given by:

                                                                                                 (65)

The post‑filtering process is performed as follows. First, the synthesized speech  is inverse filtered through  to produce the residual signal . The signal  is filtered by the synthesis filter . Finally, the signal at the output of the synthesis filter  is passed to the tilt compensation filter  resulting in the post‑filtered speech signal .

Adaptive gain control (AGC) is used to compensate for the gain difference between the synthesized speech signal  and the post‑filtered signal . The gain scaling factor  for the present subframe is computed by:

                                                                                                                                                           (66)

The gain‑scaled post‑filtered signal  is given by:

                                                                                                                                                       (67)

where  is updated in sample‑by‑sample basis and given by:

                                                                                                                           (68)

where  is a AGC factor with value of 0.9.

The adaptive post‑filtering factors are given by:,  and

                                                                              .                                                                      (69)

6.2.2       Up‑scaling

Up‑scaling consists of multiplying the post‑filtered speech by a factor of 2 to compensate for the down‑scaling by 2 which is applied to the input signal.

The various components of the 12,2 kbit/s GSM enhanced full rate codec are described in the form of a fixed‑point bit‑exact ANSI C code, which is found in GSM 06.53 [6]. This C simulation is an integrated software of the speech codec, VAD/DTX, comfort noise and bad frame handler functions. In the fixed‑point ANSI C simulation, all the computations are performed using a predefined set of basic operators.

Two types of variables are used in the fixed‑point implementation. These two types are signed integers in 2's complement representation, defined by:

             Word16   16 bit variables

             Word32   32 bit variables

The variables of the Word16 type are denoted var1, var2,..., varn, and those of type Word32 are denoted L_var1, L_var2,..., L_varn.

7.1        Description of the constants and variables used in the C code

The ANSI C code simulation of the codec is, to a large extent, self‑documented. However, a description of the variables and constants used in the code is given to facilitate the understanding of the code. The fixed‑point precision (in terms of Q format, double precision (DP), or normalized precision) of the vectors and variables is given, along with the vectors dimensions and constant values.

Table 3 gives the coder global constants and table 4 describes the variables and vectors used in the encoder routine with their precision. Table 5 describes the fixed tables in the codec.

Table 3: Codec global constants

 

Table 4: Description of the coder vectors and variables

 

Table 5: Codec fixed tables

 

Table 6: Source Encoder output parameters in order of occurrence
and bit allocation within the speech frame of 244 bits/20 ms

 

8.1        Functional description

The enhanced full rate speech codec is described in a bit‑exact arithmetic to allow for easy type approval as well as general testing purposes of the enhanced full rate speech codec.

The response of the codec to a predefined input sequence can only be foreseen if the internal state variables of the codec are in a predefined state at the beginning of the experiment. Therefore, the codec has to be put in a so called home state before a bit‑exact test can be performed. This is usually done by a reset (a procedure in which the internal state variables of the codec are set to their defined initial values).

To allow a reset of the codec in remote locations, special homing frames have been defined for the encoder and the decoder, thus enabling a codec homing by inband signalling.

The codec homing procedure is defined in such a way, that in either direction (encoder or decoder) the homing functions are called after processing the homing frame that is input. The output corresponding to the first homing frame is therefore dependent on the codec state when receiving that frame and hence usually not known. The response to any further homing frame in one direction is by definition a homing frame of the other direction. This procedure allows homing of both, the encoder and decoder from either side, if a loop back configuration is implemented, taking proper framing into account.

8.2        Definitions

Encoder homing frame: The encoder homing frame consists of 160 identical samples, each 13 bits long, with the least significant bit set to "one" and all other bits set to "zero". When written to 16‑bit words with left justification, the samples have a value of 0008 hex. The speech decoder has to produce this frame as a response to the second and any further decoder homing frame if at least two decoder homing frames were input to the decoder consecutively.

Decoder homing frame: The decoder homing frame has a fixed set of speech parameters as described in table7. It is the natural response of the speech encoder to the second and any further encoder homing frame if at least two encoder homing frames were input to the encoder consecutively.

Table7: Parameter values for the decoder homing frame

 

8.3        Encoder homing

Whenever the enhanced full rate speech encoder receives at its input an encoder homing frame exactly aligned with its internal speech frame segmentation, the following events take place:

Step 1:                   The speech encoder performs its normal operation including VAD and DTX and produces a speech parameter frame at its output which is in general unknown. But if the speech encoder was in its home state at the beginning of that frame, then the resulting speech parameter frame is identical to the decoder homing frame (this is the way how the decoder homing frame was constructed).

Step 2:                   After successful termination of that operation the speech encoder provokes the homing functions for all sub‑modules including VAD and DTX and sets all state variables into their home state. On the reception of the next input frame, the speech encoder will start from its home state.

NOTE:      Applying a sequence of N encoder homing frames will cause at least N‑1 decoder homing frames at the output of the speech encoder.

8.4        Decoder homing

Whenever the speech decoder receives at its input a decoder homing frame, then the following events take place:

Step 1:                   The speech decoder performs its normal operation and produces a speech frame at its output which is in general unknown. But if the speech decoder was in its home state at the beginning of that frame, then the resulting speech frame is replaced by the encoder homing frame. This would not naturally be the case but is forced by this definition here.

Step 2:                   After successful termination of that operation the speech decoder provokes the homing functions for all sub‑modules including the comfort noise generator and sets all state variables into their home state. On the reception of the next input frame, the speech decoder will start from its home state.

NOTE 1:  Applying a sequence of N decoder homing frames will cause at least N‑1 encoder homing frames at the output of the speech decoder.

NOTE 2:  By definition (!) the first frame of each decoder test sequence must differ from the decoder homing frame at least in one bit position within the parameters for LPC and first subframe. Therefore, if the decoder is in its home state, it is sufficient to check only these parameters to detect a subsequent decoder homing frame. This definition is made to support a delay‑optimized implementation in the TRAU uplink direction.

8.5        Encoder home state

In table 8, a listing of all the encoder state variables with their predefined values when in the home state is given.

Table 8: Initial values of the encoder state variables

 

Initial values for variables used by the VAD algorithm are listed in GSM 06.32 [4].

8.6        Decoder home state

In table 9, a listing of all the decoder state variables with their predefined values when in the home state is given.

Table 9: Initial values of the decoder state variables

Table 9 (concluded): Initial values of the decoder state variables

 

Figure 2: Simplified block diagram of the CELP synthesis model

 

                                                            Figure 3: Simplified block diagram of the GSM enhanced full rate encoder

Figure 4: Simplified block diagram of the GSM enhanced full rate decoder

1)   M.R. Schroeder and B.S. Atal, "Code‑Excited Linear Prediction (CELP): High quality speech at very low bit rates,"' Proc. ICASSP'85, pp. 937‑940, 1985.

2)   Y. Tohkura and F. Itakura, "Spectral smoothing technique in PARCOR speech analysis‑synthesis," IEEE Trans. on ASSP, vol. 26, no. 6, pp. 587‑596, Dec. 1978.

3)   L.R. Rabiner and R.W. Schaefer. Digital processing of speech signals. Prentice‑Hall Int., 1978.

4)   F. Itakura, "Line spectral representation of linear predictive coefficients of speech signals," J. Acoust. Soc. Amer, vol. 57, Supplement no. 1, S35, 1975.

5)   F.K. Soong and B.H. Juang, "Line spectrum pair (LSP) and speech data compression", Proc. ICASSP'84, pp. 1.10.1‑1.10.4, 1984.

6)   P. Kabal and R.P. Ramachandran, "The computation of line spectral frequencies using Chebyshev polynomials", IEEE Trans. on ASSP, vol. 34, no. 6, pp. 1419‑1426, Dec. 1986.

7)   C. Laflamme, J‑P. Adoul, R. Salami, S. Morissette, and P. Mabilleau, "16 kpbs wideband speech coding technique based on algebraic CELP" Proc. ICASSP'91, pp. 13‑16.

 

SMG#	SPEC	CR	PHASE	VERS	NEW_VERS	SUBJECT
s23	06.60	A003	2	4.0.0	4.0.1	Vote 115 comments
s25	06.60	A005	2	4.0.1	4.1.0	Corrections to GSM 06.60
s28	06.60			4.1.0	6.0.0	Release 1997 version
s28	06.60	A007		6.0.0	7.0.0	Addition of mu-Law (PCS 1900)
	06.60			7.0.1	7.0.2	Update to Version 7.0.2 for Publication
s31	06.60			7.0.2	8.0.0	Release 1999 version
	06.60			8.0.0	8.0.1	Update to Version 8.0.1 for Publication

 

 

Change history
Date	TSG #	TSG Doc.	CR	Rev	Subject/Comment	Old	New
03-2001	11				Version for Release 4		4.0.0
06-2002	16				Version for Release 5	4.0.0	5.0.0
12-2004	26				Version for Release 6	5.0.0	6.0.0
06-2007	36				Version for Release 7	6.0.0	7.0.0
12-2008	42				Version for Release 8	7.0.0	8.0.0