[Ffmpeg-cvslog] CVS: ffmpeg/libavcodec truespeech.c, NONE, 1.1 truespeech_data.h, NONE, 1.1 Makefile, 1.216, 1.217 allcodecs.c, 1.120, 1.121 avcodec.h, 1.438, 1.439

Diego Biurrun CVS diego
Tue Jan 3 03:30:41 CET 2006


Update of /cvsroot/ffmpeg/ffmpeg/libavcodec
In directory mail:/var2/tmp/cvs-serv25244/libavcodec

Modified Files:
	Makefile allcodecs.c avcodec.h 
Added Files:
	truespeech.c truespeech_data.h 
Log Message:
TrueSpeech compatible audio decoder by Konstantin Shishkov


--- NEW FILE: truespeech.c ---
/*
 * DSP Group TrueSpeech compatible decoder
 * Copyright (c) 2005 Konstantin Shishkov
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */
#include "avcodec.h"

#include "truespeech_data.h"
/**
 * @file truespeech.c
 * TrueSpeech decoder.
 */

/**
 * TrueSpeech decoder context
 */
typedef struct {
    /* input data */
    int16_t vector[8];  //< input vector: 5/5/4/4/4/3/3/3
    int offset1[2];     //< 8-bit value, used in one copying offset
    int offset2[4];     //< 7-bit value, encodes offsets for copying and for two-point filter
    int pulseoff[4];    //< 4-bit offset of pulse values block
    int pulsepos[4];    //< 27-bit variable, encodes 7 pulse positions
    int pulseval[4];    //< 7x2-bit pulse values
    int flag;           //< 1-bit flag, shows how to choose filters
    /* temporary data */
    int filtbuf[146];   // some big vector used for storing filters
    int prevfilt[8];    // filter from previous frame
    int16_t tmp1[8];    // coefficients for adding to out
    int16_t tmp2[8];    // coefficients for adding to out
    int16_t tmp3[8];    // coefficients for adding to out
    int16_t cvector[8]; // correlated input vector
    int filtval;        // gain value for one function
    int16_t newvec[60]; // tmp vector
    int16_t filters[32]; // filters for every subframe
} TSContext;

static int truespeech_decode_init(AVCodecContext * avctx)
{
//    TSContext *c = avctx->priv_data;

    return 0;
}

static void truespeech_read_frame(TSContext *dec, uint8_t *input)
{
    uint32_t t;

    /* first dword */
    t = LE_32(input);
    input += 4;

    dec->flag = t & 1;

    dec->vector[0] = ts_codebook[0][(t >>  1) & 0x1F];
    dec->vector[1] = ts_codebook[1][(t >>  6) & 0x1F];
    dec->vector[2] = ts_codebook[2][(t >> 11) &  0xF];
    dec->vector[3] = ts_codebook[3][(t >> 15) &  0xF];
    dec->vector[4] = ts_codebook[4][(t >> 19) &  0xF];
    dec->vector[5] = ts_codebook[5][(t >> 23) &  0x7];
    dec->vector[6] = ts_codebook[6][(t >> 26) &  0x7];
    dec->vector[7] = ts_codebook[7][(t >> 29) &  0x7];

    /* second dword */
    t = LE_32(input);
    input += 4;

    dec->offset2[0] = (t >>  0) & 0x7F;
    dec->offset2[1] = (t >>  7) & 0x7F;
    dec->offset2[2] = (t >> 14) & 0x7F;
    dec->offset2[3] = (t >> 21) & 0x7F;

    dec->offset1[0] = ((t >> 28) & 0xF) << 4;

    /* third dword */
    t = LE_32(input);
    input += 4;

    dec->pulseval[0] = (t >>  0) & 0x3FFF;
    dec->pulseval[1] = (t >> 14) & 0x3FFF;

    dec->offset1[1] = (t >> 28) & 0x0F;

    /* fourth dword */
    t = LE_32(input);
    input += 4;

    dec->pulseval[2] = (t >>  0) & 0x3FFF;
    dec->pulseval[3] = (t >> 14) & 0x3FFF;

    dec->offset1[1] |= ((t >> 28) & 0x0F) << 4;

    /* fifth dword */
    t = LE_32(input);
    input += 4;

    dec->pulsepos[0] = (t >> 4) & 0x7FFFFFF;

    dec->pulseoff[0] = (t >> 0) & 0xF;

    dec->offset1[0] |= (t >> 31) & 1;

    /* sixth dword */
    t = LE_32(input);
    input += 4;

    dec->pulsepos[1] = (t >> 4) & 0x7FFFFFF;

    dec->pulseoff[1] = (t >> 0) & 0xF;

    dec->offset1[0] |= ((t >> 31) & 1) << 1;

    /* seventh dword */
    t = LE_32(input);
    input += 4;

    dec->pulsepos[2] = (t >> 4) & 0x7FFFFFF;

    dec->pulseoff[2] = (t >> 0) & 0xF;

    dec->offset1[0] |= ((t >> 31) & 1) << 2;

    /* eighth dword */
    t = LE_32(input);
    input += 4;

    dec->pulsepos[3] = (t >> 4) & 0x7FFFFFF;

    dec->pulseoff[3] = (t >> 0) & 0xF;

    dec->offset1[0] |= ((t >> 31) & 1) << 3;

}

static void truespeech_correlate_filter(TSContext *dec)
{
    int16_t tmp[8];
    int i, j;

    for(i = 0; i < 8; i++){
        if(i > 0){
            memcpy(tmp, dec->cvector, i * 2);
            for(j = 0; j < i; j++)
                dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) +
                                   (dec->cvector[j] << 15) + 0x4000) >> 15;
        }
        dec->cvector[i] = (8 - dec->vector[i]) >> 3;
    }
    for(i = 0; i < 8; i++)
        dec->cvector[i] = (dec->cvector[i] * ts_230[i]) >> 15;

    dec->filtval = dec->vector[0];
}

static void truespeech_filters_merge(TSContext *dec)
{
    int i;

    if(!dec->flag){
        for(i = 0; i < 8; i++){
            dec->filters[i + 0] = dec->prevfilt[i];
            dec->filters[i + 8] = dec->prevfilt[i];
        }
    }else{
        for(i = 0; i < 8; i++){
            dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
            dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
        }
    }
    for(i = 0; i < 8; i++){
        dec->filters[i + 16] = dec->cvector[i];
        dec->filters[i + 24] = dec->cvector[i];
    }
}

static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
{
    int16_t tmp[146 + 60], *ptr0, *ptr1, *filter;
    int i, t, off;

    t = dec->offset2[quart];
    if(t == 127){
        memset(dec->newvec, 0, 60 * 2);
        return;
    }
    for(i = 0; i < 146; i++)
        tmp[i] = dec->filtbuf[i];
    off = (t / 25) + dec->offset1[quart >> 1] + 18;
    ptr0 = tmp + 145 - off;
    ptr1 = tmp + 146;
    filter = ts_240 + (t % 25) * 2;
    for(i = 0; i < 60; i++){
        t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
        ptr0++;
        dec->newvec[i] = t;
        ptr1[i] = t;
    }
}

static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
{
    int16_t tmp[7];
    int i, j, t;
    int16_t *ptr1, *ptr2;
    int coef;

    memset(out, 0, 60 * 2);
    for(i = 0; i < 7; i++) {
        t = dec->pulseval[quart] & 3;
        dec->pulseval[quart] >>= 2;
        tmp[6 - i] = ts_562[dec->pulseoff[quart] * 4 + t];
    }

    coef = dec->pulsepos[quart] >> 15;
    ptr1 = ts_140 + 30;
    ptr2 = tmp;
    for(i = 0, j = 3; (i < 30) && (j > 0); i++){
        t = *ptr1++;
        if(coef >= t)
            coef -= t;
        else{
            out[i] = *ptr2++;
            ptr1 += 30;
            j--;
        }
    }
    coef = dec->pulsepos[quart] & 0x7FFF;
    ptr1 = ts_140;
    for(i = 30, j = 4; (i < 60) && (j > 0); i++){
        t = *ptr1++;
        if(coef >= t)
            coef -= t;
        else{
            out[i] = *ptr2++;
            ptr1 += 30;
            j--;
        }
    }

}

static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
{
    int i;

    for(i = 0; i < 86; i++)
        dec->filtbuf[i] = dec->filtbuf[i + 60];
    for(i = 0; i < 60; i++){
        dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
        out[i] += dec->newvec[i];
    }
}

static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
{
    int i,k;
    int t[8];
    int16_t *ptr0, *ptr1;

    ptr0 = dec->tmp1;
    ptr1 = dec->filters + quart * 8;
    for(i = 0; i < 60; i++){
        int sum = 0;
        for(k = 0; k < 8; k++)
            sum += ptr0[k] * ptr1[k];
        sum = (sum + (out[i] << 12) + 0x800) >> 12;
        out[i] = clip(sum, -0x7FFE, 0x7FFE);
        for(k = 7; k > 0; k--)
            ptr0[k] = ptr0[k - 1];
        ptr0[0] = out[i];
    }

    for(i = 0; i < 8; i++)
        t[i] = (ts_5E2[i] * ptr1[i]) >> 15;

    ptr0 = dec->tmp2;
    for(i = 0; i < 60; i++){
        int sum = 0;
        for(k = 0; k < 8; k++)
            sum += ptr0[k] * t[k];
        for(k = 7; k > 0; k--)
            ptr0[k] = ptr0[k - 1];
        ptr0[0] = out[i];
        out[i] = ((out[i] << 12) - sum) >> 12;
    }

    for(i = 0; i < 8; i++)
        t[i] = (ts_5F2[i] * ptr1[i]) >> 15;

    ptr0 = dec->tmp3;
    for(i = 0; i < 60; i++){
        int sum = out[i] << 12;
        for(k = 0; k < 8; k++)
            sum += ptr0[k] * t[k];
        for(k = 7; k > 0; k--)
            ptr0[k] = ptr0[k - 1];
        ptr0[0] = clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);

        sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
        sum = sum - (sum >> 3);
        out[i] = clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
    }
}

static void truespeech_save_prevvec(TSContext *c)
{
    int i;

    for(i = 0; i < 8; i++)
        c->prevfilt[i] = c->cvector[i];
}

static int truespeech_decode_frame(AVCodecContext *avctx,
                void *data, int *data_size,
                uint8_t *buf, int buf_size)
{
    TSContext *c = avctx->priv_data;

    int i;
    short *samples = data;
    int consumed = 0;
    int16_t out_buf[240];

    if (!buf_size)
        return 0;

    while (consumed < buf_size) {
        truespeech_read_frame(c, buf + consumed);
        consumed += 32;

        truespeech_correlate_filter(c);
        truespeech_filters_merge(c);

        memset(out_buf, 0, 240 * 2);
        for(i = 0; i < 4; i++) {
            truespeech_apply_twopoint_filter(c, i);
            truespeech_place_pulses(c, out_buf + i * 60, i);
            truespeech_update_filters(c, out_buf + i * 60, i);
            truespeech_synth(c, out_buf + i * 60, i);
        }

        truespeech_save_prevvec(c);

        /* finally output decoded frame */
        for(i = 0; i < 240; i++)
            *samples++ = out_buf[i];

    }

    *data_size = consumed * 15;

    return buf_size;
}

AVCodec truespeech_decoder = {
    "truespeech",
    CODEC_TYPE_AUDIO,
    CODEC_ID_TRUESPEECH,
    sizeof(TSContext),
    truespeech_decode_init,
    NULL,
    NULL,
    truespeech_decode_frame,
};

--- NEW FILE: truespeech_data.h ---
#ifndef __TRUESPEECH_DATA__
#define __TRUESPEECH_DATA__

/* codebooks fo expanding input filter */
const int16_t ts_cb_0[32] = {
    0x8240, 0x8364, 0x84CE, 0x865D, 0x8805, 0x89DE, 0x8BD7, 0x8DF4,
    0x9051, 0x92E2, 0x95DE, 0x990F, 0x9C81, 0xA079, 0xA54C, 0xAAD2,
    0xB18A, 0xB90A, 0xC124, 0xC9CC, 0xD339, 0xDDD3, 0xE9D6, 0xF893,
    0x096F, 0x1ACA, 0x29EC, 0x381F, 0x45F9, 0x546A, 0x63C3, 0x73B5,
};

const int16_t ts_cb_1[32] = {
    0x9F65, 0xB56B, 0xC583, 0xD371, 0xE018, 0xEBB4, 0xF61C, 0xFF59,
    0x085B, 0x1106, 0x1952, 0x214A, 0x28C9, 0x2FF8, 0x36E6, 0x3D92,
    0x43DF, 0x49BB, 0x4F46, 0x5467, 0x5930, 0x5DA3, 0x61EC, 0x65F9,
    0x69D4, 0x6D5A, 0x709E, 0x73AD, 0x766B, 0x78F0, 0x7B5A, 0x7DA5,
};

const int16_t ts_cb_2[16] = {
    0x96F8, 0xA3B4, 0xAF45, 0xBA53, 0xC4B1, 0xCECC, 0xD86F, 0xE21E,
    0xEBF3, 0xF640, 0x00F7, 0x0C20, 0x1881, 0x269A, 0x376B, 0x4D60,
};

const int16_t ts_cb_3[16] = {
    0xC654, 0xDEF2, 0xEFAA, 0xFD94, 0x096A, 0x143F, 0x1E7B, 0x282C,
    0x3176, 0x3A89, 0x439F, 0x4CA2, 0x557F, 0x5E50, 0x6718, 0x6F8D,
};

const int16_t ts_cb_4[16] = {
    0xABE7, 0xBBA8, 0xC81C, 0xD326, 0xDD0E, 0xE5D4, 0xEE22, 0xF618,
    0xFE28, 0x064F, 0x0EB7, 0x17B8, 0x21AA, 0x2D8B, 0x3BA2, 0x4DF9,
};

const int16_t ts_cb_5[8] = {
    0xD51B, 0xF12E, 0x042E, 0x13C7, 0x2260, 0x311B, 0x40DE, 0x5385,
};

const int16_t ts_cb_6[8] = {
    0xB550, 0xC825, 0xD980, 0xE997, 0xF883, 0x0752, 0x1811, 0x2E18,
};

const int16_t ts_cb_7[8] = {
    0xCEF0, 0xE4F9, 0xF6BB, 0x0646, 0x14F5, 0x23FF, 0x356F, 0x4A8D,
};

const int16_t *ts_codebook[8] = {
    ts_cb_0, ts_cb_1, ts_cb_2, ts_cb_3, ts_cb_4, ts_cb_5, ts_cb_6, ts_cb_7
};

/* table used for decoding pulse positions */
const int16_t ts_140[120] = {
    0x0E46, 0x0CCC, 0x0B6D, 0x0A28, 0x08FC, 0x07E8, 0x06EB, 0x0604,
    0x0532, 0x0474, 0x03C9, 0x0330, 0x02A8, 0x0230, 0x01C7, 0x016C,
    0x011E, 0x00DC, 0x00A5, 0x0078, 0x0054, 0x0038, 0x0023, 0x0014,
    0x000A, 0x0004, 0x0001, 0x0000, 0x0000, 0x0000,

    0x0196, 0x017A, 0x015F, 0x0145, 0x012C, 0x0114, 0x00FD, 0x00E7,
    0x00D2, 0x00BE, 0x00AB, 0x0099, 0x0088, 0x0078, 0x0069, 0x005B,
    0x004E, 0x0042, 0x0037, 0x002D, 0x0024, 0x001C, 0x0015, 0x000F,
    0x000A, 0x0006, 0x0003, 0x0001, 0x0000, 0x0000,

    0x001D, 0x001C, 0x001B, 0x001A, 0x0019, 0x0018, 0x0017, 0x0016,
    0x0015, 0x0014, 0x0013, 0x0012, 0x0011, 0x0010, 0x000F, 0x000E,
    0x000D, 0x000C, 0x000B, 0x000A, 0x0009, 0x0008, 0x0007, 0x0006,
    0x0005, 0x0004, 0x0003, 0x0002, 0x0001, 0x0000,

    0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001,
    0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001,
    0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001,
    0x0001, 0x0001, 0x0001, 0x0001, 0x0001, 0x0001
};

/* filter for correlated input filter */
const int16_t ts_230[8] =
    { 0x7F3B, 0x7E78, 0x7DB6, 0x7CF5, 0x7C35, 0x7B76, 0x7AB8, 0x79FC };

/* two-point filters table */
const int16_t ts_240[25 * 2] = {
    0xED2F, 0x5239,
    0x54F1, 0xE4A9,
    0x2620, 0xEE3E,
    0x09D6, 0x2C40,
    0xEFB5, 0x2BE0,

    0x3FE1, 0x3339,
    0x442F, 0xE6FE,
    0x4458, 0xF9DF,
    0xF231, 0x43DB,
    0x3DB0, 0xF705,

    0x4F7B, 0xFEFB,
    0x26AD, 0x0CDC,
    0x33C2, 0x0739,
    0x12BE, 0x43A2,
    0x1BDF, 0x1F3E,

    0x0211, 0x0796,
    0x2AEB, 0x163F,
    0x050D, 0x3A38,
    0x0D1E, 0x0D78,
    0x150F, 0x3346,

    0x38A4, 0x0B7D,
    0x2D5D, 0x1FDF,
    0x19B7, 0x2822,
    0x0D99, 0x1F12,
    0x194C, 0x0CE6
};

/* possible pulse values */
const int16_t ts_562[64] = {
    0x0002, 0x0006, 0xFFFE, 0xFFFA,
    0x0004, 0x000C, 0xFFFC, 0xFFF4,
    0x0006, 0x0012, 0xFFFA, 0xFFEE,
    0x000A, 0x001E, 0xFFF6, 0xFFE2,
    0x0010, 0x0030, 0xFFF0, 0xFFD0,
    0x0019, 0x004B, 0xFFE7, 0xFFB5,
    0x0028, 0x0078, 0xFFD8, 0xFF88,
    0x0040, 0x00C0, 0xFFC0, 0xFF40,
    0x0065, 0x012F, 0xFF9B, 0xFED1,
    0x00A1, 0x01E3, 0xFF5F, 0xFE1D,
    0x0100, 0x0300, 0xFF00, 0xFD00,
    0x0196, 0x04C2, 0xFE6A, 0xFB3E,
    0x0285, 0x078F, 0xFD7B, 0xF871,
    0x0400, 0x0C00, 0xFC00, 0xF400,
    0x0659, 0x130B, 0xF9A7, 0xECF5,
    0x0A14, 0x1E3C, 0xF5EC, 0xE1C4
};

/* filters used in final output calculations */
const int16_t ts_5E2[8] =
    { 0x4666, 0x26B8, 0x154C, 0x0BB6, 0x0671, 0x038B, 0x01F3, 0x0112 };
const int16_t ts_5F2[8] =
    { 0x6000, 0x4800, 0x3600, 0x2880, 0x1E60, 0x16C8, 0x1116, 0x0CD1 };

#endif

Index: Makefile
===================================================================
RCS file: /cvsroot/ffmpeg/ffmpeg/libavcodec/Makefile,v
retrieving revision 1.216
retrieving revision 1.217
diff -u -d -r1.216 -r1.217
--- Makefile	17 Dec 2005 18:14:26 -0000	1.216
+++ Makefile	3 Jan 2006 02:30:38 -0000	1.217
@@ -161,6 +161,9 @@
 ifeq ($(CONFIG_TRUEMOTION2_DECODER),yes)
     OBJS+= truemotion2.o
 endif
+ifeq ($(CONFIG_TRUESPEECH_DECODER),yes)
+    OBJS+= truespeech.o
+endif
 ifeq ($(CONFIG_TSCC_DECODER),yes)
     OBJS+= tscc.o
 endif

Index: allcodecs.c
===================================================================
RCS file: /cvsroot/ffmpeg/ffmpeg/libavcodec/allcodecs.c,v
retrieving revision 1.120
retrieving revision 1.121
diff -u -d -r1.120 -r1.121
--- allcodecs.c	22 Dec 2005 01:10:06 -0000	1.120
+++ allcodecs.c	3 Jan 2006 02:30:38 -0000	1.121
@@ -497,6 +497,9 @@
 #ifdef CONFIG_COOK_DECODER
     register_avcodec(&cook_decoder);
 #endif //CONFIG_COOK_DECODER
+#ifdef CONFIG_TRUESPEECH_DECODER
+    register_avcodec(&truespeech_decoder);
+#endif //CONFIG_TRUESPEECH_DECODER
 #ifdef CONFIG_RAWVIDEO_DECODER
     register_avcodec(&rawvideo_decoder);
 #endif //CONFIG_RAWVIDEO_DECODER

Index: avcodec.h
===================================================================
RCS file: /cvsroot/ffmpeg/ffmpeg/libavcodec/avcodec.h,v
retrieving revision 1.438
retrieving revision 1.439
diff -u -d -r1.438 -r1.439
--- avcodec.h	29 Dec 2005 07:24:23 -0000	1.438
+++ avcodec.h	3 Jan 2006 02:30:38 -0000	1.439
@@ -189,6 +189,7 @@
     CODEC_ID_GSM,
     CODEC_ID_QDM2,
     CODEC_ID_COOK,
+    CODEC_ID_TRUESPEECH,
 
     CODEC_ID_OGGTHEORA= 0x16000,
 
@@ -2134,6 +2135,7 @@
 extern AVCodec mp3on4_decoder;
 extern AVCodec qdm2_decoder;
 extern AVCodec cook_decoder;
+extern AVCodec truespeech_decoder;
 extern AVCodec mace3_decoder;
 extern AVCodec mace6_decoder;
 extern AVCodec huffyuv_decoder;





More information about the ffmpeg-cvslog mailing list