rom_enc.h

ffmpeg源码分析

/*
 * ===================================================================
 *  TS 26.104
 *  REL-5 V5.4.0 2004-03
 *  REL-6 V6.1.0 2004-03
 *  3GPP AMR Floating-point Speech Codec
 * ===================================================================
 *
 */

/*
 * rom_enc.h
 *
 *
 * Project:
 *    AMR Floating-Point Codec
 *
 * Contains:
 *    This file contains all the tables needed by AMR encoder functions.
 *
 */
#ifndef _ROM_ENC_H_
#define _ROM_ENC_H_


#include "typedef.h"
#include "sp_enc.h"

#define M               10    /* Order of LP filter */
#define MP1             (M+1) /* Order of LP filter + 1 */
#define L_WINDOW        240   /* Window size in LP analysis */
#define L_NEXT          40    /* Overhead in LP analysis */
#define LTPG_MEM_SIZE   5     /* number of stored past LTP coding gains + 1 */
#define N_FRAME         7     /* old pitch gains in average calculation */
#define DTX_HIST_SIZE   8     /* DTX history size */
#define L_TOTAL         320   /* Total size of speech buffer. */
#define L_FRAME         160   /* Frame size */
#define L_FRAME_BY2     80    /* Frame size divided by 2 */
#define L_SUBFR         40    /* Subframe size */
#define L_CODE          40    /* codevector length */
#define PIT_MAX         143   /* Maximum pitch lag */
#define PIT_MIN         20    /* Minimum pitch lag */
#define PIT_MIN_MR122   18    /* Minimum pitch lag (MR122 mode) */
#define L_INTERPOL      (10+1)/* Length of filter for interpolation */
#define NPRED           4     /* number of prediction taps */
#define SHARPMIN        0     /* Minimum value of pitch sharpening */
#define MAX_PRM_SIZE    57    /* max. num. of params */
#define L_INTER_SRCH    4     /* Length of filter for CL LTP search interpolation */
#define GP_CLIP         0.95F /* Pitch gain clipping */
#define UP_SAMP_MAX     6
#define NB_TRACK        5     /* number of tracks */
#define NB_TRACK_MR102  4     /* number of tracks mode mr102 */
#define STEP            5     /* codebook step size */
#define STEP_MR102      4     /* codebook step size mode mr102 */
#define NC              M/2   /* Order of LP filter divided by 2 */

#define SCALE_LSP_FREQ  (Float32)(4000.0/3.141592654)
#define SCALE_FREQ_LSP  (Float32)(3.141592654/4000.0)
#define SLOPE1_WGHT_LSF (Float32)((3.347-1.8)/(450.0-0.0))
#define SLOPE2_WGHT_LSF (Float32)((1.8-1.0)/(1500.0-450.0))

#define FRAME_LEN 160    /* Length (samples) of the input frame          */
#define COMPLEN 9        /* Number of sub-bands used by VAD              */
#define INV_COMPLEN 3641 /* 1.0/COMPLEN*2^15                             */
#define LOOKAHEAD 40     /* length of the lookahead used by speech coder */

#define UNITY 512        /* Scaling used with SNR calculation            */
#define UNIRSHFT 6       /* = log2(MAX_16/UNITY)                         */

#define TONE_THR 0.65F   /* Threshold for tone detection   */

/* Constants for background spectrum update */
#define ALPHA_UP1   (Float32)(1.0 - 0.95)  /* Normal update, upwards:   */
#define ALPHA_DOWN1 (Float32)(1.0 - 0.936) /* Normal update, downwards  */
#define ALPHA_UP2   (Float32)(1.0 - 0.985) /* Forced update, upwards    */
#define ALPHA_DOWN2 (Float32)(1.0 - 0.943) /* Forced update, downwards  */
#define ALPHA3      (Float32)(1.0 - 0.95)  /* Update downwards          */
#define ALPHA4      (Float32)(1.0 - 0.9)   /* For stationary estimation */
#define ALPHA5      (Float32)(1.0 - 0.5)   /* For stationary estimation */

/* Constants for VAD threshold */
#define VAD_THR_HIGH 1260 /* Highest threshold                 */
#define VAD_THR_LOW  720  /* Lowest threshold                  */
#define VAD_P1 0          /* Noise level for highest threshold */
#define VAD_P2 6300       /* Noise level for lowest threshold  */
#define VAD_SLOPE (float)(VAD_THR_LOW-VAD_THR_HIGH)/(float)(VAD_P2-VAD_P1)

/* Parameters for background spectrum recovery function */
#define STAT_COUNT 20         /* threshold of stationary detection counter         */
#define STAT_COUNT_BY_2 10    /* threshold of stationary detection counter         */
#define CAD_MIN_STAT_COUNT 5  /* threshold of stationary detection counter         */

#define STAT_THR_LEVEL 184    /* Threshold level for stationarity detection        */
#define STAT_THR 1000         /* Threshold for stationarity detection              */

/* Limits for background noise estimate */
#define NOISE_MIN 40          /* minimum */
#define NOISE_MAX 16000       /* maximum */
#define NOISE_INIT 150        /* initial */

/* Constants for VAD hangover addition */
#define HANG_NOISE_THR 100
#define BURST_LEN_HIGH_NOISE 4
#define HANG_LEN_HIGH_NOISE 7
#define BURST_LEN_LOW_NOISE 5
#define HANG_LEN_LOW_NOISE 4

/* Thresholds for signal power */
#define VAD_POW_LOW (Word32)15000/2    /* If input power is lower,                    */
                                       /*     VAD is set to 0                         */
#define POW_PITCH_THR (Word32)343040/2 /* If input power is lower, pitch              */
                                       /*     detection is ignored                    */

#define POW_COMPLEX_THR (Word32)15000/2/* If input power is lower, complex            */
                                       /* flags  value for previous frame  is un-set  */
/*
 * VAD Constants
 */

/* Constants for the filter bank */
#define LEVEL_SHIFT 0                     /* scaling                                  */
#define COEFF3   (Float32)13363/32768     /* coefficient for the 3rd order filter     */
#define COEFF5_1 (Float32)21955/32768     /* 1st coefficient the for 5th order filter */
#define COEFF5_2 (Float32)6390/32768      /* 2nd coefficient the for 5th order filter */

/* Constants for pitch detection */
#define LTHRESH 4
#define NTHRESH 4

/* Constants for complex signal VAD  */
#define CVAD_THRESH_ADAPT_HIGH  0.6F            /* threshold for adapt stopping high */
#define CVAD_THRESH_ADAPT_LOW  0.5F             /* threshold for adapt stopping low */
#define CVAD_THRESH_IN_NOISE  0.65F * 32768.0F  /* threshold going into speech on
                                                   a short term basis */
#define CVAD_THRESH_HANG  0.70F           /* threshold */
#define CVAD_HANG_LIMIT  (Word16)(100)    /* 2 second estimation time */
#define CVAD_HANG_LENGTH  (Word16)(250)   /* 5 second hangover */

#define CVAD_LOWPOW_RESET 0.40F  /* init in low power segment */
#define CVAD_MIN_CORR 0.40F      /* lowest adaptation value */

#define CVAD_BURST 20                        /* speech burst length for speech reset */
#define CVAD_ADAPT_SLOW 1.0F - 0.98F         /* threshold for slow adaption */
#define CVAD_ADAPT_FAST 1.0F - 0.92F         /* threshold for fast adaption */
#define CVAD_ADAPT_REALLY_FAST 1.0F - 0.80F  /* threshold for really fast adaption */

/* track table for algebraic code book search (MR475, MR515) */
static Word8 trackTable[4 * 5] =
   {
      /* subframe 1; track to code; -1 do not code this position */ 0,
      1,
      0,
      1,
      - 1,
      /* subframe 2 */ 0,
      - 1,
      1,
      0,
      1,
      /* subframe 3 */ 0,
      1,
      0,
      - 1,
      1,
      /* subframe 4 */ 0,
      1,
      - 1,
      0,
      1
   };
static const Float32 gamma1[M] =
{
   0.9400024414063F,
   0.8836059570313F,
   0.8305969238281F,
   0.78076171875F,
   0.7339172363281F,
   0.6898803710938F,
   0.6484985351563F,
   0.6095886230469F,
   0.5730285644531F,
   0.5386352539063F
};

/*
 *  gamma1 for the 12k2 coder
 */
static const Float32 gamma1_12k2[M] =
{
   0.8999938964844F,
   0.8099975585938F,
   0.72900390625F,
   0.6560974121094F,
   0.5904846191406F,
   0.5314331054688F,
   0.478271484375F,
   0.4304504394531F,
   0.3873901367188F,
   0.3486633300781F
};
static const Float32 gamma2[M] =
{
   0.6000061035156F,
   0.3600158691406F,
   0.2160034179688F,
   0.1296081542969F,
   0.0777587890625F,
   0.04666137695313F,
   0.02798461914063F,
   0.01678466796875F,
   0.01007080078125F,
   0.00604248046875F
};

/* 1/6 resolution interpolation filter  (-3 dB at 3600 Hz)
 * Note: the 1/3 resolution filter is simply a subsampled
 *       version of the 1/6 resolution filter, i.e. it uses
 *       every second coefficient:
 *
 *       inter_6(1/3)[k] = inter_6(1/3)[2*k], 0 <= k <= 3*L_INTER10
 */
static Float32 b60[UP_SAMP_MAX*(L_INTERPOL-1)+1] =
{
   0.898529F,
   0.865051F,
   0.769257F,
   0.624054F,
   0.448639F,
   0.265289F,
   0.0959167F,
   - 0.0412598F,
   - 0.134338F,
   - 0.178986F,
   - 0.178528F,
   - 0.142609F,
   - 0.0849304F,
   - 0.0205078F,
   0.0369568F,
   0.0773926F,
   0.0955200F,
   0.0912781F,
   0.0689392F,
   0.0357056F,
   0.000000F,
   - 0.0305481F,
   - 0.0504150F,
   - 0.0570068F,
   - 0.0508423F,
   - 0.0350037F,
   - 0.0141602F,
   0.00665283F,
   0.0230713F,
   0.0323486F,
   0.0335388F,
   0.0275879F,
   0.0167847F,
   0.00411987F,
   - 0.00747681F,
   - 0.0156860F,
   - 0.0193481F,
   - 0.0183716F,
   - 0.0137634F,
   - 0.00704956F,
   0.000000F,
   0.00582886F,
   0.00939941F,
   0.0103760F,
   0.00903320F,
   0.00604248F,
   0.00238037F,
   - 0.00109863F,
   - 0.00366211F,
   - 0.00497437F,
   - 0.00503540F,
   - 0.00402832F,
   - 0.00241089F,
   - 0.000579834F,
   0.00103760F,
   0.00222778F,
   0.00277710F,
   0.00271606F,
   0.00213623F,
   0.00115967F,
   0.000000F
};

/* same in fixed-point */
static const Word32 inter6[61] =
{
   29443,
   28346,
   25207,
   20449,
   14701,
   8693,
   3143,
   - 1352,
   - 4402,
   - 5865,
   - 5850,
   - 4673,
   - 2783,
   - 672,
   1211,
   2536,
   3130,
   2991,
   2259,
   1170,
   0,
   - 1001,
   - 1652,
   - 1868,
   - 1666,
   - 1147,
   - 464,
   218,
   756,
   1060,
   1099,
   904,
   550,
   135,
   - 245,
   - 514,
   - 634,
   - 602,
   - 451,
   - 231,
   0,
   191,
   308,
   340,
   296,
   198,
   78,
   - 36,
   - 120,
   - 163,
   - 165,
   - 132,
   - 79,
   - 19,
   34,
   73,
   91,
   89,
   70,
   38,
   0
};

static const Word16 startPos1[2] =
   {
      1,
      3
   };

static const Word16 startPos2[4] =
   {
      0,
      1,
      2,
      4
   };

static const Word16 startPos[2 * 4 * 2] =
   {
      0,
      2,
      0,
      3,
      0,
      2,
      0,
      3,
      1,
      3,
      2,
      4,
      1,
      4,
      1,
      4
   };

/* Scalar quantization tables of the pitch gain and the codebook gain. */
#define NB_QUA_PITCH 16

static const Float32 qua_gain_pitch[NB_QUA_PITCH] =
{
0.0F,
0.20001220703125F,
0.400146484375F,
0.5F,
0.5999755859375F,
0.70001220703125F,
0.75F,
0.79998779296875F,
0.8499755859375F,
0.9000244140625F,
0.95001220703125F,
1.0F,
1.04998779296875F,
1.0999755859375F,
1.1500244140625F,
1.20001220703125F
};

static const Float32 qua_gain_pitch_MR122[NB_QUA_PITCH] =
{
0.0F,
0.199951171875F,
0.400146484375F,
0.5F,
0.599853515625F,
0.699951171875F,
0.75F,
0.7998046875F,
0.849853515625F,
0.89990234375F,
0.949951171875F,
1.0F,
1.0498046875F,
1.099853515625F,
1.14990234375F,
1.199951171875F
};

#define NB_QUA_CODE 32
static const Float32 gain_factor[NB_QUA_CODE] =
{
/* gain factor (g_fac) */
   0.0776367F,
   0.100586F,
   0.130859F,
   0.170410F,
   0.204590F,
   0.235352F,
   0.270508F,
   0.311035F,
   0.357910F,
   0.411133F,
   0.473145F,
   0.543945F,
   0.625488F,
   0.719238F,
   0.827148F,
   0.951172F,
   1.09424F,
   1.25830F,
   1.44678F,
   1.66406F,
   1.91357F,
   2.20068F,
   2.53076F,
   2.91016F,
   3.34717F,
   3.84912F,
   4.42627F,
   5.09033F,
   6.10840F,
   7.94092F,
   10.3232F,
   13.4204F
};






static const Word8 gray[8] =
   {
      0,
      1,
      3,
      2,
      6,
      4,
      5,
      7
   };

/*
 * grid[0] = 1.0;
 * grid[61] = -1.0;
 * for (i = 1; i < 61; i++)
 *    grid[i] = (Float32)cos((6.283185307*i)/(2.0*60));
 */
static const Float32 grid[61] =
{
   1.00000F,
   0.998630F,
   0.994522F,
   0.987688F,
   0.978148F,
   0.965926F,
   0.951057F,
   0.933580F,
   0.913545F,
   0.891007F,
   0.866025F,
   0.838671F,
   0.809017F,
   0.777146F,
   0.743145F,
   0.707107F,
   0.669131F,
   0.629320F,
   0.587785F,
   0.544639F,
   0.500000F,
   0.453990F,
   0.406737F,
   0.358368F,
   0.309017F,
   0.258819F,
   0.207912F,
   0.156434F,
   0.104528F,
   0.0523360F,
   4.48966e-011F,
   - 0.0523360F,
   - 0.104528F,
   - 0.156434F,
   - 0.207912F,
   - 0.258819F,
   - 0.309017F,
   - 0.358368F,
   - 0.406737F,
   - 0.453990F,
   - 0.500000F,
   - 0.544639F,
   - 0.587785F,
   - 0.629320F,
   - 0.669131F,
   - 0.707107F,
   - 0.743145F,
   - 0.777146F,
   - 0.809017F,
   - 0.838671F,
   - 0.866025F,
   - 0.891007F,
   - 0.913545F,
   - 0.933580F,
   - 0.951057F,
   - 0.965926F,
   - 0.978148F,
   - 0.987688F,
   - 0.994522F,
   - 0.998630F,
   - 1.00000F
};

/*
 * 1/6 resolution interpolation filter  (-3 dB at 3600 Hz)
 * Note: The IS641 (7.4) 1/3 resolution filter is simply a subsampled
 *       version of the 1/6 resolution filter, i.e. it uses
 *       every second coefficient:
 *
 *       b24[k](1/3) = b24[2*k](1/6), 0 <= k <= 3*L_INTER_SRCH
 */
static const Float32 b24[UP_SAMP_MAX*L_INTER_SRCH+1] =
{
   0.900848F,
   0.864136F,
   0.760071F,
   0.605408F,
   0.424072F,
   0.242462F,
   0.0840759F,
   - 0.0343933F,
   - 0.105560F,
   - 0.131348F,
   - 0.121124F,
   - 0.0884705F,
   - 0.0476379F,
   - 0.0102539F,
   0.0162964F,
   0.0296021F,
   0.0312195F,
   0.0251160F,
   0.0157471F,
   0.00671387F,
   0.000000F,
   - 0.00399780F,
   - 0.00592041F,
   - 0.00656128F,
   0.000000F
};

/*
 * Exponential Window coefficients used to weight the autocorrelation
 * coefficients for 60 Hz bandwidth expansion of high pitched voice
 * before Levinson-Durbin recursion to compute the LPC coefficients.
 *
 * lagwindow[i] =  exp( -0.5*(2*pi*F0*(i+1)/Fs)^2 ); i = 0,...,9
 * F0 = 60 Hz, Fs = 8000 Hz
 */
static Float32 lag_wind[M] =
{
   0.99889028F,
   0.99556851F,
   0.99005681F,
   0.98239160F,
   0.97262347F,
   0.96081644F,
   0.94704735F,
   0.93140495F,
   0.91398895F,
   0.89490914F
};

/* initialization table for lsp history in DTX */
static const Float32 lsp_init_data[M] =
   {
      0.9595F,
      0.8413F,
      0.6549F,
      0.4154F,
      0.1423F,
      - 0.1423F,
      - 0.4154F,
      - 0.6549F,
      - 0.8413F,
      - 0.9595F
   };

#define PAST_RQ_INIT_SIZE 8

/* initalization table for MA predictor in dtx mode */
static const Float32 past_rq_init[80] =
{
-62.9883F, -77.6367F, -107.178F, -154.785F, -160.156F, -188.721F, -173.584F, -122.559F,
-65.4297F, -47.1191F, -0.488281F, 30.5176F, 29.7852F, -9.52148F, -2.19727F, 25.6348F,
31.4941F, 69.0918F, 90.8203F, 140.381F, -67.627F, -79.1016F, -48.0957F, -118.896F,
-108.643F, -88.3789F, -71.2891F, -6.5918F, 43.2129F, 132.568F, 83.4961F, 126.221F,
125.977F, 31.7383F, 6.5918F, -25.3906F, -29.2969F, -34.1797F, -18.0664F, -13.6719F,
-137.695F, -230.225F, -371.094F, -235.596F, -198.73F, -128.418F, -78.6133F, -0.488281F,
38.8184F, 160.4F, -76.1719F, -69.3359F, -94.2383F, -145.752F, -120.361F, -128.418F,
-102.051F, -55.9082F, 25.6348F, 109.619F, -135.986F, -212.402F, -262.451F, -224.365F,
-231.934F, -183.594F, -173.096F, -77.1484F, 15.1367F, 118.652F, -76.6602F, -46.6309F,
-49.5605F, -80.5664F, -39.0625F, -25.1465F, -12.4512F, 31.9824F, 82.5195F, 125.732F
};

/* LSF means ->normalize frequency domain */

static const Float32 mean_lsf_3[10] =
{
   377.441F,
	554.688F,
	922.363F,
	1339.84F,
	1702.15F,
	2046.39F,
	2452.88F,
	2741.46F,
	3116.70F,
	3348.14F,
};

static const Float32 mean_lsf_5[10] =
{
   337.891F,
	507.080F,
	834.961F,
	1247.07F,
	1646.00F,
	1982.91F,
	2407.96F,
	2708.01F,
	3104.00F,
	3344.97F,
};

/* LSF prediction factors (not in MR122) */
static const Float32 pred_fac[10] =
{
   0.291626F,
   0.328644F,
   0.383636F,
   0.405640F,
   0.438873F,
   0.355560F,
   0.323120F,
   0.298065F,
   0.262238F,
   0.197876F
};

/* codebooks from IS641 */
#define DICO1_SIZE_3  256
#define DICO2_SIZE_3  512
#define DICO3_SIZE_3  512
static const Float32 dico1_lsf_3[] =
{
   	1.46484F,
		20.0195F,
		-31.9824F,
		37.5977F,
		-13.6719F,
		-179.443F,
		44.6777F,
		-15.8691F,
		-64.6973F,
		2.19727F,
		-51.2695F,
		-88.1348F,
		27.5879F,
		175.293F,
		443.604F,
		246.582F,
		296.387F,
		384.033F,
		209.229F,
		325.439F,
		555.664F,
		201.904F,
		382.813F,
		471.924F,
		175.049F,
		485.596F,
		538.574F,
		204.590F,
		286.133F,
		445.068F,
		176.025F,
		244.141F,
		525.879F,
		69.8242F,
		116.211F,
		368.408F,
		-60.3027F,
		-129.639F,
		56.1523F,
		35.8887F,
		-20.0195F,
		138.916F,
		6.34766F,
		-43.2129F,
		-230.469F,
		-6.59180F,
		-66.6504F,
		168.945F,
		-40.0391F,
		-64.4531F,
		-44.6777F,
		54.6875F,
		192.871F,
		253.662F,
		219.482F,
		230.957F,
		146.729F,
		118.408F,
		188.232F,
		280.762F,
		127.930F,
		165.283F,
		220.459F,
		-34.1797F,
		91.5527F,
		189.941F,
		100.098F,
		165.039F,
		104.736F,
		73.4863F,
		129.395F,
		246.338F,
		175.537F,
		157.715F,
		9.27734F,
		55.1758F,
		89.5996F,
		9.76563F,
		35.4004F,
		-10.9863F,
		-123.291F,
		70.8008F,
		29.5410F,
		-29.5410F,
		73.7305F,
		31.0059F,
		40.5273F,
		-30.2734F,
		-93.5059F,
		-233.398F,
		-87.4023F,
		-111.084F,
		-238.525F,
		174.561F,
		214.355F,
		218.262F,
		238.770F,
		225.342F,
		51.5137F,
		116.455F,
		66.4063F,
		15.6250F,
		45.8984F,
		-19.0430F,
		4.15039F,
		-34.9121F,
		-15.8691F,
		9.27734F,
		156.982F,
		143.066F,
		151.611F,
		-32.7148F,
		-104.004F,
		-158.936F,
		84.7168F,
		133.057F,
		688.477F,
		290.039F,
		665.527F,
		596.191F,
		34.6680F,
		-19.5313F,
		423.584F,
		69.0918F,
		31.7383F,
		112.549F,
		-63.9648F,
		-97.4121F,
		-279.541F,
		-100.342F,
		37.8418F,