bf3027c8 |
/*
* DSP Group TrueSpeech compatible decoder
* Copyright (c) 2005 Konstantin Shishkov
* |
b78e7197 |
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or |
bf3027c8 |
* 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. |
bf3027c8 |
* |
b78e7197 |
* FFmpeg is distributed in the hope that it will be useful, |
bf3027c8 |
* 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 |
bf3027c8 |
*/ |
6a5d31ac |
#include "libavutil/intreadwrite.h" |
bf3027c8 |
#include "avcodec.h" |
b8404893 |
#include "dsputil.h"
#include "get_bits.h" |
bf3027c8 |
#include "truespeech_data.h"
/** |
ba87f080 |
* @file |
bf3027c8 |
* TrueSpeech decoder.
*/
/**
* TrueSpeech decoder context
*/
typedef struct { |
0eea2129 |
AVFrame frame; |
b8404893 |
DSPContext dsp; |
bf3027c8 |
/* input data */ |
d4eeadcb |
DECLARE_ALIGNED(16, uint8_t, buffer)[32]; |
02494787 |
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 |
bf3027c8 |
/* 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;
|
98a6fff9 |
static av_cold int truespeech_decode_init(AVCodecContext * avctx) |
bf3027c8 |
{ |
b8404893 |
TSContext *c = avctx->priv_data; |
bf3027c8 |
|
3e7a1767 |
if (avctx->channels != 1) {
av_log_ask_for_sample(avctx, "Unsupported channel count: %d\n", avctx->channels);
return AVERROR(EINVAL);
} |
bf3027c8 |
|
5d6e4c16 |
avctx->sample_fmt = AV_SAMPLE_FMT_S16; |
b8404893 |
|
9cf0841e |
ff_dsputil_init(&c->dsp, avctx); |
b8404893 |
|
0eea2129 |
avcodec_get_frame_defaults(&c->frame);
avctx->coded_frame = &c->frame;
|
bf3027c8 |
return 0;
}
|
d177b620 |
static void truespeech_read_frame(TSContext *dec, const uint8_t *input) |
bf3027c8 |
{ |
b8404893 |
GetBitContext gb;
dec->dsp.bswap_buf((uint32_t *)dec->buffer, (const uint32_t *)input, 8);
init_get_bits(&gb, dec->buffer, 32 * 8);
dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)];
dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)];
dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)];
dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)];
dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)];
dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)];
dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)];
dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)];
dec->flag = get_bits1(&gb);
dec->offset1[0] = get_bits(&gb, 4) << 4;
dec->offset2[3] = get_bits(&gb, 7);
dec->offset2[2] = get_bits(&gb, 7);
dec->offset2[1] = get_bits(&gb, 7);
dec->offset2[0] = get_bits(&gb, 7);
dec->offset1[1] = get_bits(&gb, 4);
dec->pulseval[1] = get_bits(&gb, 14);
dec->pulseval[0] = get_bits(&gb, 14);
dec->offset1[1] |= get_bits(&gb, 4) << 4;
dec->pulseval[3] = get_bits(&gb, 14);
dec->pulseval[2] = get_bits(&gb, 14);
dec->offset1[0] |= get_bits1(&gb);
dec->pulsepos[0] = get_bits_long(&gb, 27);
dec->pulseoff[0] = get_bits(&gb, 4);
dec->offset1[0] |= get_bits1(&gb) << 1;
dec->pulsepos[1] = get_bits_long(&gb, 27);
dec->pulseoff[1] = get_bits(&gb, 4);
dec->offset1[0] |= get_bits1(&gb) << 2;
dec->pulsepos[2] = get_bits_long(&gb, 27);
dec->pulseoff[2] = get_bits(&gb, 4);
dec->offset1[0] |= get_bits1(&gb) << 3;
dec->pulsepos[3] = get_bits_long(&gb, 27);
dec->pulseoff[3] = get_bits(&gb, 4); |
bf3027c8 |
}
static void truespeech_correlate_filter(TSContext *dec)
{
int16_t tmp[8];
int i, j;
for(i = 0; i < 8; i++){
if(i > 0){ |
c6ac30c8 |
memcpy(tmp, dec->cvector, i * sizeof(*tmp)); |
bf3027c8 |
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++) |
c5cb9c94 |
dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15; |
bf3027c8 |
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)
{ |
7b49ce2e |
int16_t tmp[146 + 60], *ptr0, *ptr1;
const int16_t *filter; |
bf3027c8 |
int i, t, off;
t = dec->offset2[quart];
if(t == 127){ |
c6ac30c8 |
memset(dec->newvec, 0, 60 * sizeof(*dec->newvec)); |
bf3027c8 |
return;
}
for(i = 0; i < 146; i++)
tmp[i] = dec->filtbuf[i];
off = (t / 25) + dec->offset1[quart >> 1] + 18; |
f264d336 |
off = av_clip(off, 0, 145); |
bf3027c8 |
ptr0 = tmp + 145 - off;
ptr1 = tmp + 146; |
c5cb9c94 |
filter = (const int16_t*)ts_order2_coeffs + (t % 25) * 2; |
bf3027c8 |
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; |
7b49ce2e |
const int16_t *ptr1;
int16_t *ptr2; |
bf3027c8 |
int coef;
|
c6ac30c8 |
memset(out, 0, 60 * sizeof(*out)); |
bf3027c8 |
for(i = 0; i < 7; i++) {
t = dec->pulseval[quart] & 3;
dec->pulseval[quart] >>= 2; |
c5cb9c94 |
tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t]; |
bf3027c8 |
}
coef = dec->pulsepos[quart] >> 15; |
c5cb9c94 |
ptr1 = (const int16_t*)ts_pulse_values + 30; |
bf3027c8 |
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; |
c5cb9c94 |
ptr1 = (const int16_t*)ts_pulse_values; |
bf3027c8 |
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;
|
59f4d1b8 |
memmove(dec->filtbuf, &dec->filtbuf[60], 86 * sizeof(*dec->filtbuf)); |
bf3027c8 |
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; |
f66e4f5f |
out[i] = av_clip(sum, -0x7FFE, 0x7FFE); |
bf3027c8 |
for(k = 7; k > 0; k--)
ptr0[k] = ptr0[k - 1];
ptr0[0] = out[i];
}
for(i = 0; i < 8; i++) |
c5cb9c94 |
t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15; |
bf3027c8 |
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++) |
c5cb9c94 |
t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15; |
bf3027c8 |
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]; |
f66e4f5f |
ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE); |
bf3027c8 |
sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
sum = sum - (sum >> 3); |
f66e4f5f |
out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE); |
bf3027c8 |
}
}
static void truespeech_save_prevvec(TSContext *c)
{
int i;
for(i = 0; i < 8; i++)
c->prevfilt[i] = c->cvector[i];
}
|
0eea2129 |
static int truespeech_decode_frame(AVCodecContext *avctx, void *data,
int *got_frame_ptr, AVPacket *avpkt) |
bf3027c8 |
{ |
7a00bbad |
const uint8_t *buf = avpkt->data;
int buf_size = avpkt->size; |
bf3027c8 |
TSContext *c = avctx->priv_data;
|
e938637b |
int i, j; |
0eea2129 |
int16_t *samples;
int iterations, ret; |
bf3027c8 |
|
a8f8db26 |
iterations = buf_size / 32; |
bf3027c8 |
|
a8f8db26 |
if (!iterations) { |
23174578 |
av_log(avctx, AV_LOG_ERROR,
"Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
return -1;
} |
a8f8db26 |
|
0eea2129 |
/* get output buffer */
c->frame.nb_samples = iterations * 240;
if ((ret = avctx->get_buffer(avctx, &c->frame)) < 0) {
av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
return ret; |
a8f8db26 |
} |
0eea2129 |
samples = (int16_t *)c->frame.data[0]; |
a8f8db26 |
|
0eea2129 |
memset(samples, 0, iterations * 240 * sizeof(*samples)); |
5e5ce70f |
|
e938637b |
for(j = 0; j < iterations; j++) { |
6d55506c |
truespeech_read_frame(c, buf);
buf += 32; |
bf3027c8 |
truespeech_correlate_filter(c);
truespeech_filters_merge(c);
for(i = 0; i < 4; i++) {
truespeech_apply_twopoint_filter(c, i); |
5e5ce70f |
truespeech_place_pulses (c, samples, i);
truespeech_update_filters(c, samples, i);
truespeech_synth (c, samples, i);
samples += 60; |
bf3027c8 |
}
truespeech_save_prevvec(c);
}
|
0eea2129 |
*got_frame_ptr = 1;
*(AVFrame *)data = c->frame; |
bf3027c8 |
|
6d55506c |
return buf_size; |
bf3027c8 |
}
|
e7e2df27 |
AVCodec ff_truespeech_decoder = { |
ec6402b7 |
.name = "truespeech",
.type = AVMEDIA_TYPE_AUDIO,
.id = CODEC_ID_TRUESPEECH,
.priv_data_size = sizeof(TSContext),
.init = truespeech_decode_init,
.decode = truespeech_decode_frame, |
0eea2129 |
.capabilities = CODEC_CAP_DR1, |
00c3b67b |
.long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"), |
bf3027c8 |
}; |