FFmpeg
truespeech.c
Go to the documentation of this file.
1 /*
2  * DSP Group TrueSpeech compatible decoder
3  * Copyright (c) 2005 Konstantin Shishkov
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * FFmpeg is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
23 #include "libavutil/mem_internal.h"
24 
25 #include "avcodec.h"
26 #include "bswapdsp.h"
27 #include "codec_internal.h"
28 #include "decode.h"
29 #include "get_bits.h"
30 
31 #include "truespeech_data.h"
32 /**
33  * @file
34  * TrueSpeech decoder.
35  */
36 
37 /**
38  * TrueSpeech decoder context
39  */
40 typedef struct TSContext {
42  /* input data */
43  DECLARE_ALIGNED(16, uint8_t, buffer)[32];
44  int16_t vector[8]; ///< input vector: 5/5/4/4/4/3/3/3
45  int offset1[2]; ///< 8-bit value, used in one copying offset
46  int offset2[4]; ///< 7-bit value, encodes offsets for copying and for two-point filter
47  int pulseoff[4]; ///< 4-bit offset of pulse values block
48  int pulsepos[4]; ///< 27-bit variable, encodes 7 pulse positions
49  int pulseval[4]; ///< 7x2-bit pulse values
50  int flag; ///< 1-bit flag, shows how to choose filters
51  /* temporary data */
52  int filtbuf[146]; // some big vector used for storing filters
53  int prevfilt[8]; // filter from previous frame
54  int16_t tmp1[8]; // coefficients for adding to out
55  int16_t tmp2[8]; // coefficients for adding to out
56  int16_t tmp3[8]; // coefficients for adding to out
57  int16_t cvector[8]; // correlated input vector
58  int filtval; // gain value for one function
59  int16_t newvec[60]; // tmp vector
60  int16_t filters[32]; // filters for every subframe
61 } TSContext;
62 
64 {
65  TSContext *c = avctx->priv_data;
66 
67  if (avctx->ch_layout.nb_channels != 1) {
68  avpriv_request_sample(avctx, "Channel count %d", avctx->ch_layout.nb_channels);
69  return AVERROR_PATCHWELCOME;
70  }
71 
75 
76  ff_bswapdsp_init(&c->bdsp);
77 
78  return 0;
79 }
80 
81 static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
82 {
83  GetBitContext gb;
84 
85  dec->bdsp.bswap_buf((uint32_t *) dec->buffer, (const uint32_t *) input, 8);
86  init_get_bits(&gb, dec->buffer, 32 * 8);
87 
88  dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)];
89  dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)];
90  dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)];
91  dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)];
92  dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)];
93  dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)];
94  dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)];
95  dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)];
96  dec->flag = get_bits1(&gb);
97 
98  dec->offset1[0] = get_bits(&gb, 4) << 4;
99  dec->offset2[3] = get_bits(&gb, 7);
100  dec->offset2[2] = get_bits(&gb, 7);
101  dec->offset2[1] = get_bits(&gb, 7);
102  dec->offset2[0] = get_bits(&gb, 7);
103 
104  dec->offset1[1] = get_bits(&gb, 4);
105  dec->pulseval[1] = get_bits(&gb, 14);
106  dec->pulseval[0] = get_bits(&gb, 14);
107 
108  dec->offset1[1] |= get_bits(&gb, 4) << 4;
109  dec->pulseval[3] = get_bits(&gb, 14);
110  dec->pulseval[2] = get_bits(&gb, 14);
111 
112  dec->offset1[0] |= get_bits1(&gb);
113  dec->pulsepos[0] = get_bits_long(&gb, 27);
114  dec->pulseoff[0] = get_bits(&gb, 4);
115 
116  dec->offset1[0] |= get_bits1(&gb) << 1;
117  dec->pulsepos[1] = get_bits_long(&gb, 27);
118  dec->pulseoff[1] = get_bits(&gb, 4);
119 
120  dec->offset1[0] |= get_bits1(&gb) << 2;
121  dec->pulsepos[2] = get_bits_long(&gb, 27);
122  dec->pulseoff[2] = get_bits(&gb, 4);
123 
124  dec->offset1[0] |= get_bits1(&gb) << 3;
125  dec->pulsepos[3] = get_bits_long(&gb, 27);
126  dec->pulseoff[3] = get_bits(&gb, 4);
127 }
128 
130 {
131  int16_t tmp[8];
132  int i, j;
133 
134  for(i = 0; i < 8; i++){
135  if(i > 0){
136  memcpy(tmp, dec->cvector, i * sizeof(*tmp));
137  for(j = 0; j < i; j++)
138  dec->cvector[j] += (tmp[i - j - 1] * dec->vector[i] + 0x4000) >> 15;
139  }
140  dec->cvector[i] = (8 - dec->vector[i]) >> 3;
141  }
142  for(i = 0; i < 8; i++)
143  dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15;
144 
145  dec->filtval = dec->vector[0];
146 }
147 
149 {
150  int i;
151 
152  if(!dec->flag){
153  for(i = 0; i < 8; i++){
154  dec->filters[i + 0] = dec->prevfilt[i];
155  dec->filters[i + 8] = dec->prevfilt[i];
156  }
157  }else{
158  for(i = 0; i < 8; i++){
159  dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
160  dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
161  }
162  }
163  for(i = 0; i < 8; i++){
164  dec->filters[i + 16] = dec->cvector[i];
165  dec->filters[i + 24] = dec->cvector[i];
166  }
167 }
168 
169 static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
170 {
171  int16_t tmp[146 + 60], *ptr0, *ptr1;
172  const int16_t *filter;
173  int i, t, off;
174 
175  t = dec->offset2[quart];
176  if(t == 127){
177  memset(dec->newvec, 0, 60 * sizeof(*dec->newvec));
178  return;
179  }
180  for(i = 0; i < 146; i++)
181  tmp[i] = dec->filtbuf[i];
182  off = (t / 25) + dec->offset1[quart >> 1] + 18;
183  off = av_clip(off, 0, 145);
184  ptr0 = tmp + 145 - off;
185  ptr1 = tmp + 146;
186  filter = ts_order2_coeffs + (t % 25) * 2;
187  for(i = 0; i < 60; i++){
188  t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
189  ptr0++;
190  dec->newvec[i] = t;
191  ptr1[i] = t;
192  }
193 }
194 
195 static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
196 {
197  int16_t tmp[7];
198  int i, j, t;
199  const int16_t *ptr1;
200  int16_t *ptr2;
201  int coef;
202 
203  memset(out, 0, 60 * sizeof(*out));
204  for(i = 0; i < 7; i++) {
205  t = dec->pulseval[quart] & 3;
206  dec->pulseval[quart] >>= 2;
207  tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t];
208  }
209 
210  coef = dec->pulsepos[quart] >> 15;
211  ptr1 = ts_pulse_values + 30;
212  ptr2 = tmp;
213  for(i = 0, j = 3; (i < 30) && (j > 0); i++){
214  t = *ptr1++;
215  if(coef >= t)
216  coef -= t;
217  else{
218  out[i] = *ptr2++;
219  ptr1 += 30;
220  j--;
221  }
222  }
223  coef = dec->pulsepos[quart] & 0x7FFF;
224  ptr1 = ts_pulse_values;
225  for(i = 30, j = 4; (i < 60) && (j > 0); i++){
226  t = *ptr1++;
227  if(coef >= t)
228  coef -= t;
229  else{
230  out[i] = *ptr2++;
231  ptr1 += 30;
232  j--;
233  }
234  }
235 
236 }
237 
238 static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
239 {
240  int i;
241 
242  memmove(dec->filtbuf, &dec->filtbuf[60], 86 * sizeof(*dec->filtbuf));
243  for(i = 0; i < 60; i++){
244  dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
245  out[i] += dec->newvec[i];
246  }
247 }
248 
249 static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
250 {
251  int i,k;
252  int t[8];
253  int16_t *ptr0, *ptr1;
254 
255  ptr0 = dec->tmp1;
256  ptr1 = dec->filters + quart * 8;
257  for(i = 0; i < 60; i++){
258  int sum = 0;
259  for(k = 0; k < 8; k++)
260  sum += ptr0[k] * (unsigned)ptr1[k];
261  sum = out[i] + ((int)(sum + 0x800U) >> 12);
262  out[i] = av_clip(sum, -0x7FFE, 0x7FFE);
263  for(k = 7; k > 0; k--)
264  ptr0[k] = ptr0[k - 1];
265  ptr0[0] = out[i];
266  }
267 
268  for(i = 0; i < 8; i++)
269  t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15;
270 
271  ptr0 = dec->tmp2;
272  for(i = 0; i < 60; i++){
273  int sum = 0;
274  for(k = 0; k < 8; k++)
275  sum += ptr0[k] * t[k];
276  for(k = 7; k > 0; k--)
277  ptr0[k] = ptr0[k - 1];
278  ptr0[0] = out[i];
279  out[i] += (- sum) >> 12;
280  }
281 
282  for(i = 0; i < 8; i++)
283  t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15;
284 
285  ptr0 = dec->tmp3;
286  for(i = 0; i < 60; i++){
287  int sum = out[i] * (1 << 12);
288  for(k = 0; k < 8; k++)
289  sum += ptr0[k] * t[k];
290  for(k = 7; k > 0; k--)
291  ptr0[k] = ptr0[k - 1];
292  ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
293 
294  sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
295  sum = sum - (sum >> 3);
296  out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
297  }
298 }
299 
301 {
302  int i;
303 
304  for(i = 0; i < 8; i++)
305  c->prevfilt[i] = c->cvector[i];
306 }
307 
309  int *got_frame_ptr, AVPacket *avpkt)
310 {
311  const uint8_t *buf = avpkt->data;
312  int buf_size = avpkt->size;
313  TSContext *c = avctx->priv_data;
314 
315  int i, j;
316  int16_t *samples;
317  int iterations, ret;
318 
319  iterations = buf_size / 32;
320 
321  if (!iterations) {
322  av_log(avctx, AV_LOG_ERROR,
323  "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
324  return -1;
325  }
326 
327  /* get output buffer */
328  frame->nb_samples = iterations * 240;
329  if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
330  return ret;
331  samples = (int16_t *)frame->data[0];
332 
333  memset(samples, 0, iterations * 240 * sizeof(*samples));
334 
335  for(j = 0; j < iterations; j++) {
336  truespeech_read_frame(c, buf);
337  buf += 32;
338 
341 
342  for(i = 0; i < 4; i++) {
347  samples += 60;
348  }
349 
351  }
352 
353  *got_frame_ptr = 1;
354 
355  return buf_size;
356 }
357 
359  .p.name = "truespeech",
360  CODEC_LONG_NAME("DSP Group TrueSpeech"),
361  .p.type = AVMEDIA_TYPE_AUDIO,
362  .p.id = AV_CODEC_ID_TRUESPEECH,
363  .priv_data_size = sizeof(TSContext),
366  .p.capabilities = AV_CODEC_CAP_DR1,
367 };
bswapdsp.h
av_clip
#define av_clip
Definition: common.h:95
mem_internal.h
out
FILE * out
Definition: movenc.c:54
truespeech_decode_init
static av_cold int truespeech_decode_init(AVCodecContext *avctx)
Definition: truespeech.c:63
get_bits_long
static unsigned int get_bits_long(GetBitContext *s, int n)
Read 0-32 bits.
Definition: get_bits.h:546
TSContext::flag
int flag
1-bit flag, shows how to choose filters
Definition: truespeech.c:50
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:325
tmp
static uint8_t tmp[11]
Definition: aes_ctr.c:28
AVPacket::data
uint8_t * data
Definition: packet.h:374
TSContext::filters
int16_t filters[32]
Definition: truespeech.c:60
truespeech_decode_frame
static int truespeech_decode_frame(AVCodecContext *avctx, AVFrame *frame, int *got_frame_ptr, AVPacket *avpkt)
Definition: truespeech.c:308
TSContext::pulseoff
int pulseoff[4]
4-bit offset of pulse values block
Definition: truespeech.c:47
FFCodec
Definition: codec_internal.h:119
TSContext::tmp1
int16_t tmp1[8]
Definition: truespeech.c:54
filter
filter_frame For filters that do not use the this method is called when a frame is pushed to the filter s input It can be called at any time except in a reentrant way If the input frame is enough to produce then the filter should push the output frames on the output link immediately As an exception to the previous rule if the input frame is enough to produce several output frames then the filter needs output only at least one per link The additional frames can be left buffered in the filter
Definition: filter_design.txt:228
TSContext::filtval
int filtval
Definition: truespeech.c:58
AVChannelLayout::nb_channels
int nb_channels
Number of channels in this layout.
Definition: channel_layout.h:311
init_get_bits
static int init_get_bits(GetBitContext *s, const uint8_t *buffer, int bit_size)
Initialize GetBitContext.
Definition: get_bits.h:649
get_bits
static unsigned int get_bits(GetBitContext *s, int n)
Read 1-25 bits.
Definition: get_bits.h:379
FFCodec::p
AVCodec p
The public AVCodec.
Definition: codec_internal.h:123
AVCodecContext::ch_layout
AVChannelLayout ch_layout
Audio channel layout.
Definition: avcodec.h:2059
GetBitContext
Definition: get_bits.h:61
TSContext::offset2
int offset2[4]
7-bit value, encodes offsets for copying and for two-point filter
Definition: truespeech.c:46
AV_LOG_ERROR
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:180
av_cold
#define av_cold
Definition: attributes.h:90
FF_CODEC_DECODE_CB
#define FF_CODEC_DECODE_CB(func)
Definition: codec_internal.h:298
TSContext::pulsepos
int pulsepos[4]
27-bit variable, encodes 7 pulse positions
Definition: truespeech.c:48
AVMEDIA_TYPE_AUDIO
@ AVMEDIA_TYPE_AUDIO
Definition: avutil.h:202
BswapDSPContext::bswap_buf
void(* bswap_buf)(uint32_t *dst, const uint32_t *src, int w)
Definition: bswapdsp.h:25
init
int(* init)(AVBSFContext *ctx)
Definition: dts2pts_bsf.c:363
TSContext::tmp2
int16_t tmp2[8]
Definition: truespeech.c:55
decode.h
get_bits.h
truespeech_read_frame
static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
Definition: truespeech.c:81
CODEC_LONG_NAME
#define CODEC_LONG_NAME(str)
Definition: codec_internal.h:264
truespeech_update_filters
static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
Definition: truespeech.c:238
ff_bswapdsp_init
av_cold void ff_bswapdsp_init(BswapDSPContext *c)
Definition: bswapdsp.c:49
AVERROR_PATCHWELCOME
#define AVERROR_PATCHWELCOME
Not yet implemented in FFmpeg, patches welcome.
Definition: error.h:64
ts_decay_35_64
static const int16_t ts_decay_35_64[8]
Definition: truespeech_data.h:154
get_bits1
static unsigned int get_bits1(GetBitContext *s)
Definition: get_bits.h:498
ts_decay_3_4
static const int16_t ts_decay_3_4[8]
Definition: truespeech_data.h:156
ts_pulse_scales
static const int16_t ts_pulse_scales[64]
Definition: truespeech_data.h:134
truespeech_place_pulses
static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
Definition: truespeech.c:195
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
truespeech_data.h
TSContext::offset1
int offset1[2]
8-bit value, used in one copying offset
Definition: truespeech.c:45
ff_get_buffer
int ff_get_buffer(AVCodecContext *avctx, AVFrame *frame, int flags)
Get a buffer for a frame.
Definition: decode.c:1450
AV_CODEC_CAP_DR1
#define AV_CODEC_CAP_DR1
Codec uses get_buffer() or get_encode_buffer() for allocating buffers and supports custom allocators.
Definition: codec.h:52
AVPacket::size
int size
Definition: packet.h:375
AVChannelLayout
An AVChannelLayout holds information about the channel layout of audio data.
Definition: channel_layout.h:301
codec_internal.h
AVCodecContext::sample_fmt
enum AVSampleFormat sample_fmt
audio sample format
Definition: avcodec.h:1023
ff_truespeech_decoder
const FFCodec ff_truespeech_decoder
Definition: truespeech.c:358
TSContext::prevfilt
int prevfilt[8]
Definition: truespeech.c:53
TSContext::tmp3
int16_t tmp3[8]
Definition: truespeech.c:56
TSContext::vector
int16_t vector[8]
input vector: 5/5/4/4/4/3/3/3
Definition: truespeech.c:44
input
and forward the test the status of outputs and forward it to the corresponding return FFERROR_NOT_READY If the filters stores internally one or a few frame for some input
Definition: filter_design.txt:172
DECLARE_ALIGNED
#define DECLARE_ALIGNED(n, t, v)
Definition: mem.h:116
ts_decay_994_1000
static const int16_t ts_decay_994_1000[8]
Definition: truespeech_data.h:97
ts_order2_coeffs
static const int16_t ts_order2_coeffs[25 *2]
Definition: truespeech_data.h:101
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:269
TSContext
TrueSpeech decoder context.
Definition: truespeech.c:40
AV_SAMPLE_FMT_S16
@ AV_SAMPLE_FMT_S16
signed 16 bits
Definition: samplefmt.h:58
AVCodec::name
const char * name
Name of the codec implementation.
Definition: codec.h:211
truespeech_filters_merge
static void truespeech_filters_merge(TSContext *dec)
Definition: truespeech.c:148
ts_codebook
static const int16_t *const ts_codebook[8]
Definition: truespeech_data.h:69
avcodec.h
ret
ret
Definition: filter_design.txt:187
frame
these buffered frames must be flushed immediately if a new input produces new the filter must not call request_frame to get more It must just process the frame or queue it The task of requesting more frames is left to the filter s request_frame method or the application If a filter has several the filter must be ready for frames arriving randomly on any input any filter with several inputs will most likely require some kind of queuing mechanism It is perfectly acceptable to have a limited queue and to drop frames when the inputs are too unbalanced request_frame For filters that do not use the this method is called when a frame is wanted on an output For a it should directly call filter_frame on the corresponding output For a if there are queued frames already one of these frames should be pushed If the filter should request a frame on one of its repeatedly until at least one frame has been pushed Return or at least make progress towards producing a frame
Definition: filter_design.txt:264
TSContext::newvec
int16_t newvec[60]
Definition: truespeech.c:59
AVCodecContext
main external API structure.
Definition: avcodec.h:398
TSContext::cvector
int16_t cvector[8]
Definition: truespeech.c:57
truespeech_save_prevvec
static void truespeech_save_prevvec(TSContext *c)
Definition: truespeech.c:300
channel_layout.h
av_channel_layout_uninit
void av_channel_layout_uninit(AVChannelLayout *channel_layout)
Free any allocated data in the channel layout and reset the channel count to 0.
Definition: channel_layout.c:632
samples
Filter the word “frame” indicates either a video frame or a group of audio samples
Definition: filter_design.txt:8
AV_CODEC_ID_TRUESPEECH
@ AV_CODEC_ID_TRUESPEECH
Definition: codec_id.h:456
TSContext::bdsp
BswapDSPContext bdsp
Definition: truespeech.c:41
avpriv_request_sample
#define avpriv_request_sample(...)
Definition: tableprint_vlc.h:36
AV_CHANNEL_LAYOUT_MONO
#define AV_CHANNEL_LAYOUT_MONO
Definition: channel_layout.h:368
AVPacket
This structure stores compressed data.
Definition: packet.h:351
AVCodecContext::priv_data
void * priv_data
Definition: avcodec.h:425
TSContext::pulseval
int pulseval[4]
7x2-bit pulse values
Definition: truespeech.c:49
TSContext::filtbuf
int filtbuf[146]
Definition: truespeech.c:52
av_log
#define av_log(a,...)
Definition: tableprint_vlc.h:27
BswapDSPContext
Definition: bswapdsp.h:24
truespeech_synth
static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
Definition: truespeech.c:249
ts_pulse_values
static const int16_t ts_pulse_values[120]
Definition: truespeech_data.h:74
TSContext::buffer
uint8_t buffer[32]
Definition: truespeech.c:43
int
int
Definition: ffmpeg_filter.c:156
truespeech_correlate_filter
static void truespeech_correlate_filter(TSContext *dec)
Definition: truespeech.c:129
truespeech_apply_twopoint_filter
static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
Definition: truespeech.c:169