e5966052 |
/* |
115329f1 |
* G.726 ADPCM audio codec |
406792e7 |
* Copyright (c) 2004 Roman Shaposhnik |
e5966052 |
*
* This is a very straightforward rendition of the G.726 |
115329f1 |
* Section 4 "Computational Details". |
e5966052 |
* |
b78e7197 |
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or |
e5966052 |
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either |
b78e7197 |
* version 2.1 of the License, or (at your option) any later version. |
e5966052 |
* |
b78e7197 |
* FFmpeg is distributed in the hope that it will be useful, |
e5966052 |
* 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 |
b78e7197 |
* License along with FFmpeg; if not, write to the Free Software |
5509bffa |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
e5966052 |
*/
#include <limits.h>
#include "avcodec.h" |
caa336b4 |
#include "bitstream.h" |
e5966052 |
|
ef9c5d84 |
/**
* G.726 11bit float. |
115329f1 |
* G.726 Standard uses rather odd 11bit floating point arithmentic for |
e5966052 |
* numerous occasions. It's a mistery to me why they did it this way
* instead of simply using 32bit integer arithmetic.
*/
typedef struct Float11 { |
deec52cc |
uint8_t sign; /**< 1bit sign */
uint8_t exp; /**< 4bit exponent */
uint8_t mant; /**< 6bit mantissa */ |
e5966052 |
} Float11;
|
5171fd9d |
static inline Float11* i2f(int i, Float11* f) |
e5966052 |
{ |
e344c1ea |
f->sign = (i < 0);
if (f->sign)
i = -i;
f->exp = av_log2_16bit(i) + !!i;
f->mant = i? (i<<6) >> f->exp : 1<<5;
return f; |
e5966052 |
}
static inline int16_t mult(Float11* f1, Float11* f2)
{ |
bb270c08 |
int res, exp; |
e5966052 |
|
bb270c08 |
exp = f1->exp + f2->exp; |
7d7a9e7f |
res = (((f1->mant * f2->mant) + 0x30) >> 4);
res = exp > 19 ? res << (exp - 19) : res >> (19 - exp); |
bb270c08 |
return (f1->sign ^ f2->sign) ? -res : res; |
e5966052 |
}
static inline int sgn(int value)
{ |
e344c1ea |
return (value < 0) ? -1 : 1; |
e5966052 |
}
typedef struct G726Tables { |
9307db13 |
const int* quant; /**< quantization table */ |
a0e5830a |
const int16_t* iquant; /**< inverse quantization table */ |
e3cafac6 |
const int16_t* W; /**< special table #1 ;-) */
const uint8_t* F; /**< special table #2 */ |
e5966052 |
} G726Tables;
typedef struct G726Context { |
05c9f351 |
G726Tables tbls; /**< static tables needed for computation */ |
e344c1ea |
Float11 sr[2]; /**< prev. reconstructed samples */
Float11 dq[6]; /**< prev. difference */
int a[2]; /**< second order predictor coeffs */
int b[6]; /**< sixth order predictor coeffs */
int pk[2]; /**< signs of prev. 2 sez + dq */
int ap; /**< scale factor control */
int yu; /**< fast scale factor */
int yl; /**< slow scale factor */
int dms; /**< short average magnitude of F[i] */
int dml; /**< long average magnitude of F[i] */
int td; /**< tone detect */
int se; /**< estimated signal for the next iteration */
int sez; /**< estimated second order prediction */
int y; /**< quantizer scaling factor for the next iteration */ |
c79c1a01 |
int code_size; |
e5966052 |
} G726Context;
|
9307db13 |
static const int quant_tbl16[] = /**< 16kbit/s 2bits per sample */ |
115329f1 |
{ 260, INT_MAX }; |
a0e5830a |
static const int16_t iquant_tbl16[] = |
e5966052 |
{ 116, 365, 365, 116 }; |
e3cafac6 |
static const int16_t W_tbl16[] = |
e5966052 |
{ -22, 439, 439, -22 }; |
e3cafac6 |
static const uint8_t F_tbl16[] = |
e5966052 |
{ 0, 7, 7, 0 }; |
115329f1 |
|
9307db13 |
static const int quant_tbl24[] = /**< 24kbit/s 3bits per sample */ |
e5966052 |
{ 7, 217, 330, INT_MAX }; |
a0e5830a |
static const int16_t iquant_tbl24[] =
{ INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN }; |
e3cafac6 |
static const int16_t W_tbl24[] = |
115329f1 |
{ -4, 30, 137, 582, 582, 137, 30, -4 }; |
e3cafac6 |
static const uint8_t F_tbl24[] = |
e5966052 |
{ 0, 1, 2, 7, 7, 2, 1, 0 }; |
115329f1 |
|
9307db13 |
static const int quant_tbl32[] = /**< 32kbit/s 4bits per sample */ |
e5966052 |
{ -125, 79, 177, 245, 299, 348, 399, INT_MAX }; |
a0e5830a |
static const int16_t iquant_tbl32[] =
{ INT16_MIN, 4, 135, 213, 273, 323, 373, 425,
425, 373, 323, 273, 213, 135, 4, INT16_MIN }; |
e3cafac6 |
static const int16_t W_tbl32[] = |
e5966052 |
{ -12, 18, 41, 64, 112, 198, 355, 1122, |
bb270c08 |
1122, 355, 198, 112, 64, 41, 18, -12}; |
e3cafac6 |
static const uint8_t F_tbl32[] = |
e5966052 |
{ 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 }; |
115329f1 |
|
9307db13 |
static const int quant_tbl40[] = /**< 40kbit/s 5bits per sample */ |
e5966052 |
{ -122, -16, 67, 138, 197, 249, 297, 338, |
bb270c08 |
377, 412, 444, 474, 501, 527, 552, INT_MAX }; |
a0e5830a |
static const int16_t iquant_tbl40[] =
{ INT16_MIN, -66, 28, 104, 169, 224, 274, 318, |
bb270c08 |
358, 395, 429, 459, 488, 514, 539, 566,
566, 539, 514, 488, 459, 429, 395, 358, |
a0e5830a |
318, 274, 224, 169, 104, 28, -66, INT16_MIN }; |
e3cafac6 |
static const int16_t W_tbl40[] = |
e5966052 |
{ 14, 14, 24, 39, 40, 41, 58, 100, |
bb270c08 |
141, 179, 219, 280, 358, 440, 529, 696,
696, 529, 440, 358, 280, 219, 179, 141,
100, 58, 41, 40, 39, 24, 14, 14 }; |
e3cafac6 |
static const uint8_t F_tbl40[] = |
e5966052 |
{ 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6, |
bb270c08 |
6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 }; |
e5966052 |
|
9307db13 |
static const G726Tables G726Tables_pool[] = |
0e0d6cfd |
{{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
{ quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
{ quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
{ quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }}; |
115329f1 |
|
e5966052 |
|
ef9c5d84 |
/** |
115329f1 |
* Para 4.2.2 page 18: Adaptive quantizer. |
e5966052 |
*/
static inline uint8_t quant(G726Context* c, int d)
{ |
e344c1ea |
int sign, exp, i, dln; |
115329f1 |
|
e344c1ea |
sign = i = 0;
if (d < 0) {
sign = 1;
d = -d;
}
exp = av_log2_16bit(d);
dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2); |
115329f1 |
|
05c9f351 |
while (c->tbls.quant[i] < INT_MAX && c->tbls.quant[i] < dln) |
e5966052 |
++i; |
115329f1 |
|
e344c1ea |
if (sign)
i = ~i; |
0e0d6cfd |
if (c->code_size != 2 && i == 0) /* I'm not sure this is a good idea */ |
e344c1ea |
i = 0xff; |
e5966052 |
|
e344c1ea |
return i; |
e5966052 |
}
|
ef9c5d84 |
/** |
e5966052 |
* Para 4.2.3 page 22: Inverse adaptive quantizer.
*/
static inline int16_t inverse_quant(G726Context* c, int i)
{
int dql, dex, dqt; |
115329f1 |
|
05c9f351 |
dql = c->tbls.iquant[i] + (c->y >> 2); |
e5966052 |
dex = (dql>>7) & 0xf; /* 4bit exponent */
dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */ |
6aa37bcf |
return (dql < 0) ? 0 : ((dqt<<dex) >> 7); |
e5966052 |
}
|
fc234250 |
static int16_t g726_decode(G726Context* c, int I) |
e5966052 |
{
int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
Float11 f; |
0e0d6cfd |
int I_sig= I >> (c->code_size - 1); |
115329f1 |
|
e5966052 |
dq = inverse_quant(c, I);
/* Transition detect */
ylint = (c->yl >> 15);
ylfrac = (c->yl >> 10) & 0x1f;
thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint; |
73ff4f83 |
tr= (c->td == 1 && dq > ((3*thr2)>>2));
|
cb26c9d6 |
if (I_sig) /* get the sign */ |
73ff4f83 |
dq = -dq;
re_signal = c->se + dq; |
115329f1 |
|
e5966052 |
/* Update second order predictor coefficient A2 and A1 */
pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
dq0 = dq ? sgn(dq) : 0;
if (tr) {
c->a[0] = 0; |
bb270c08 |
c->a[1] = 0; |
e5966052 |
for (i=0; i<6; i++) |
e344c1ea |
c->b[i] = 0; |
e5966052 |
} else { |
bb270c08 |
/* This is a bit crazy, but it really is +255 not +256 */ |
f66e4f5f |
fa1 = av_clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255); |
115329f1 |
|
bb270c08 |
c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7); |
f66e4f5f |
c->a[1] = av_clip(c->a[1], -12288, 12288); |
e5966052 |
c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8); |
f66e4f5f |
c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]); |
e5966052 |
for (i=0; i<6; i++) |
e344c1ea |
c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8); |
e5966052 |
}
/* Update Dq and Sr and Pk */
c->pk[1] = c->pk[0];
c->pk[0] = pk0 ? pk0 : 1;
c->sr[1] = c->sr[0];
i2f(re_signal, &c->sr[0]);
for (i=5; i>0; i--) |
e344c1ea |
c->dq[i] = c->dq[i-1]; |
e5966052 |
i2f(dq, &c->dq[0]); |
cb26c9d6 |
c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */ |
115329f1 |
|
428c82cb |
c->td = c->a[1] < -11776; |
115329f1 |
|
e5966052 |
/* Update Ap */ |
74d94417 |
c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5);
c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7); |
115329f1 |
if (tr) |
e344c1ea |
c->ap = 256; |
50c52d22 |
else { |
e344c1ea |
c->ap += (-c->ap) >> 4; |
50c52d22 |
if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3))
c->ap += 0x20;
} |
e5966052 |
/* Update Yu and Yl */ |
05c9f351 |
c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120); |
e5966052 |
c->yl += c->yu + ((-c->yl)>>6); |
115329f1 |
|
e5966052 |
/* Next iteration for Y */
al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6; |
115329f1 |
|
e5966052 |
/* Next iteration for SE and SEZ */
c->se = 0;
for (i=0; i<6; i++) |
e344c1ea |
c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]); |
e5966052 |
c->sez = c->se >> 1;
for (i=0; i<2; i++) |
e344c1ea |
c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]); |
e5966052 |
c->se >>= 1;
|
f66e4f5f |
return av_clip(re_signal << 2, -0xffff, 0xffff); |
e5966052 |
}
|
8161c28c |
static av_cold int g726_reset(G726Context* c, int index) |
e5966052 |
{
int i;
|
05c9f351 |
c->tbls = G726Tables_pool[index]; |
e5966052 |
for (i=0; i<2; i++) { |
aeaa7c3d |
c->sr[i].mant = 1<<5; |
e344c1ea |
c->pk[i] = 1; |
e5966052 |
}
for (i=0; i<6; i++) { |
aeaa7c3d |
c->dq[i].mant = 1<<5; |
e5966052 |
}
c->yu = 544;
c->yl = 34816;
c->y = 544;
return 0;
}
|
b250f9c6 |
#if CONFIG_ADPCM_G726_ENCODER |
e5966052 |
static int16_t g726_encode(G726Context* c, int16_t sig)
{ |
e344c1ea |
uint8_t i; |
115329f1 |
|
0e0d6cfd |
i = quant(c, sig/4 - c->se) & ((1<<c->code_size) - 1); |
cf409a6f |
g726_decode(c, i); |
e344c1ea |
return i; |
e5966052 |
} |
62bb489b |
#endif |
e5966052 |
/* Interfacing to the libavcodec */
|
98a6fff9 |
static av_cold int g726_init(AVCodecContext * avctx) |
e5966052 |
{ |
c79c1a01 |
G726Context* c = avctx->priv_data; |
bd10f6e1 |
unsigned int index;
if (avctx->sample_rate <= 0) {
av_log(avctx, AV_LOG_ERROR, "Samplerate is invalid\n");
return -1;
}
index = (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate - 2; |
115329f1 |
|
aca49077 |
if (avctx->bit_rate % avctx->sample_rate && avctx->codec->encode) {
av_log(avctx, AV_LOG_ERROR, "Bitrate - Samplerate combination is invalid\n"); |
bb270c08 |
return -1; |
eb5b0422 |
} |
8b470cc5 |
if(avctx->channels != 1){
av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
return -1;
} |
35d0e16b |
if(index>3){
av_log(avctx, AV_LOG_ERROR, "Unsupported number of bits %d\n", index+2);
return -1;
} |
c79c1a01 |
g726_reset(c, index); |
0e0d6cfd |
c->code_size = index+2; |
e5966052 |
|
826ca104 |
avctx->coded_frame = avcodec_alloc_frame();
if (!avctx->coded_frame) |
8fa36ae0 |
return AVERROR(ENOMEM); |
826ca104 |
avctx->coded_frame->key_frame = 1;
|
fd76c37f |
if (avctx->codec->decode)
avctx->sample_fmt = SAMPLE_FMT_S16;
|
826ca104 |
return 0;
}
|
98a6fff9 |
static av_cold int g726_close(AVCodecContext *avctx) |
826ca104 |
{
av_freep(&avctx->coded_frame); |
e5966052 |
return 0;
}
|
b250f9c6 |
#if CONFIG_ADPCM_G726_ENCODER |
e5966052 |
static int g726_encode_frame(AVCodecContext *avctx,
uint8_t *dst, int buf_size, void *data)
{ |
c79c1a01 |
G726Context *c = avctx->priv_data; |
e5966052 |
short *samples = data;
PutBitContext pb;
init_put_bits(&pb, dst, 1024*1024);
for (; buf_size; buf_size--) |
c79c1a01 |
put_bits(&pb, c->code_size, g726_encode(c, *samples++)); |
e5966052 |
flush_put_bits(&pb);
|
115329f1 |
return put_bits_count(&pb)>>3;
} |
62bb489b |
#endif |
e5966052 |
static int g726_decode_frame(AVCodecContext *avctx,
void *data, int *data_size, |
5cbbbf37 |
const uint8_t *buf, int buf_size) |
e5966052 |
{ |
c79c1a01 |
G726Context *c = avctx->priv_data; |
e5966052 |
short *samples = data; |
115329f1 |
GetBitContext gb;
|
e5966052 |
init_get_bits(&gb, buf, buf_size * 8); |
115329f1 |
|
e5966052 |
while (get_bits_count(&gb) + c->code_size <= buf_size*8) |
c79c1a01 |
*samples++ = g726_decode(c, get_bits(&gb, c->code_size)); |
115329f1 |
|
ef4c5c6d |
if(buf_size*8 != get_bits_count(&gb))
av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n"); |
115329f1 |
|
e5966052 |
*data_size = (uint8_t*)samples - (uint8_t*)data;
return buf_size;
}
|
b250f9c6 |
#if CONFIG_ADPCM_G726_ENCODER |
e5966052 |
AVCodec adpcm_g726_encoder = {
"g726",
CODEC_TYPE_AUDIO,
CODEC_ID_ADPCM_G726, |
c79c1a01 |
sizeof(G726Context), |
e5966052 |
g726_init,
g726_encode_frame, |
826ca104 |
g726_close, |
e5966052 |
NULL, |
fd76c37f |
.sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE}, |
fe4bf374 |
.long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"), |
e5966052 |
}; |
f544a5fc |
#endif |
e5966052 |
AVCodec adpcm_g726_decoder = {
"g726",
CODEC_TYPE_AUDIO,
CODEC_ID_ADPCM_G726, |
c79c1a01 |
sizeof(G726Context), |
e5966052 |
g726_init,
NULL, |
826ca104 |
g726_close, |
e5966052 |
g726_decode_frame, |
fe4bf374 |
.long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"), |
e5966052 |
}; |