FFmpeg  4.4.6
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/intreadwrite.h"
24 #include "libavutil/mem_internal.h"
25 
26 #include "avcodec.h"
27 #include "bswapdsp.h"
28 #include "get_bits.h"
29 #include "internal.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 */
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->channels != 1) {
68  avpriv_request_sample(avctx, "Channel count %d", avctx->channels);
69  return AVERROR_PATCHWELCOME;
70  }
71 
74 
75  ff_bswapdsp_init(&c->bdsp);
76 
77  return 0;
78 }
79 
80 static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
81 {
82  GetBitContext gb;
83 
84  dec->bdsp.bswap_buf((uint32_t *) dec->buffer, (const uint32_t *) input, 8);
85  init_get_bits(&gb, dec->buffer, 32 * 8);
86 
87  dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)];
88  dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)];
89  dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)];
90  dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)];
91  dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)];
92  dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)];
93  dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)];
94  dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)];
95  dec->flag = get_bits1(&gb);
96 
97  dec->offset1[0] = get_bits(&gb, 4) << 4;
98  dec->offset2[3] = get_bits(&gb, 7);
99  dec->offset2[2] = get_bits(&gb, 7);
100  dec->offset2[1] = get_bits(&gb, 7);
101  dec->offset2[0] = get_bits(&gb, 7);
102 
103  dec->offset1[1] = get_bits(&gb, 4);
104  dec->pulseval[1] = get_bits(&gb, 14);
105  dec->pulseval[0] = get_bits(&gb, 14);
106 
107  dec->offset1[1] |= get_bits(&gb, 4) << 4;
108  dec->pulseval[3] = get_bits(&gb, 14);
109  dec->pulseval[2] = get_bits(&gb, 14);
110 
111  dec->offset1[0] |= get_bits1(&gb);
112  dec->pulsepos[0] = get_bits_long(&gb, 27);
113  dec->pulseoff[0] = get_bits(&gb, 4);
114 
115  dec->offset1[0] |= get_bits1(&gb) << 1;
116  dec->pulsepos[1] = get_bits_long(&gb, 27);
117  dec->pulseoff[1] = get_bits(&gb, 4);
118 
119  dec->offset1[0] |= get_bits1(&gb) << 2;
120  dec->pulsepos[2] = get_bits_long(&gb, 27);
121  dec->pulseoff[2] = get_bits(&gb, 4);
122 
123  dec->offset1[0] |= get_bits1(&gb) << 3;
124  dec->pulsepos[3] = get_bits_long(&gb, 27);
125  dec->pulseoff[3] = get_bits(&gb, 4);
126 }
127 
129 {
130  int16_t tmp[8];
131  int i, j;
132 
133  for(i = 0; i < 8; i++){
134  if(i > 0){
135  memcpy(tmp, dec->cvector, i * sizeof(*tmp));
136  for(j = 0; j < i; j++)
137  dec->cvector[j] += (tmp[i - j - 1] * dec->vector[i] + 0x4000) >> 15;
138  }
139  dec->cvector[i] = (8 - dec->vector[i]) >> 3;
140  }
141  for(i = 0; i < 8; i++)
142  dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15;
143 
144  dec->filtval = dec->vector[0];
145 }
146 
148 {
149  int i;
150 
151  if(!dec->flag){
152  for(i = 0; i < 8; i++){
153  dec->filters[i + 0] = dec->prevfilt[i];
154  dec->filters[i + 8] = dec->prevfilt[i];
155  }
156  }else{
157  for(i = 0; i < 8; i++){
158  dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
159  dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
160  }
161  }
162  for(i = 0; i < 8; i++){
163  dec->filters[i + 16] = dec->cvector[i];
164  dec->filters[i + 24] = dec->cvector[i];
165  }
166 }
167 
168 static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
169 {
170  int16_t tmp[146 + 60], *ptr0, *ptr1;
171  const int16_t *filter;
172  int i, t, off;
173 
174  t = dec->offset2[quart];
175  if(t == 127){
176  memset(dec->newvec, 0, 60 * sizeof(*dec->newvec));
177  return;
178  }
179  for(i = 0; i < 146; i++)
180  tmp[i] = dec->filtbuf[i];
181  off = (t / 25) + dec->offset1[quart >> 1] + 18;
182  off = av_clip(off, 0, 145);
183  ptr0 = tmp + 145 - off;
184  ptr1 = tmp + 146;
185  filter = ts_order2_coeffs + (t % 25) * 2;
186  for(i = 0; i < 60; i++){
187  t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
188  ptr0++;
189  dec->newvec[i] = t;
190  ptr1[i] = t;
191  }
192 }
193 
194 static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
195 {
196  int16_t tmp[7];
197  int i, j, t;
198  const int16_t *ptr1;
199  int16_t *ptr2;
200  int coef;
201 
202  memset(out, 0, 60 * sizeof(*out));
203  for(i = 0; i < 7; i++) {
204  t = dec->pulseval[quart] & 3;
205  dec->pulseval[quart] >>= 2;
206  tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t];
207  }
208 
209  coef = dec->pulsepos[quart] >> 15;
210  ptr1 = ts_pulse_values + 30;
211  ptr2 = tmp;
212  for(i = 0, j = 3; (i < 30) && (j > 0); i++){
213  t = *ptr1++;
214  if(coef >= t)
215  coef -= t;
216  else{
217  out[i] = *ptr2++;
218  ptr1 += 30;
219  j--;
220  }
221  }
222  coef = dec->pulsepos[quart] & 0x7FFF;
223  ptr1 = ts_pulse_values;
224  for(i = 30, j = 4; (i < 60) && (j > 0); i++){
225  t = *ptr1++;
226  if(coef >= t)
227  coef -= t;
228  else{
229  out[i] = *ptr2++;
230  ptr1 += 30;
231  j--;
232  }
233  }
234 
235 }
236 
237 static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
238 {
239  int i;
240 
241  memmove(dec->filtbuf, &dec->filtbuf[60], 86 * sizeof(*dec->filtbuf));
242  for(i = 0; i < 60; i++){
243  dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
244  out[i] += dec->newvec[i];
245  }
246 }
247 
248 static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
249 {
250  int i,k;
251  int t[8];
252  int16_t *ptr0, *ptr1;
253 
254  ptr0 = dec->tmp1;
255  ptr1 = dec->filters + quart * 8;
256  for(i = 0; i < 60; i++){
257  int sum = 0;
258  for(k = 0; k < 8; k++)
259  sum += ptr0[k] * (unsigned)ptr1[k];
260  sum = out[i] + ((int)(sum + 0x800U) >> 12);
261  out[i] = av_clip(sum, -0x7FFE, 0x7FFE);
262  for(k = 7; k > 0; k--)
263  ptr0[k] = ptr0[k - 1];
264  ptr0[0] = out[i];
265  }
266 
267  for(i = 0; i < 8; i++)
268  t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15;
269 
270  ptr0 = dec->tmp2;
271  for(i = 0; i < 60; i++){
272  int sum = 0;
273  for(k = 0; k < 8; k++)
274  sum += ptr0[k] * t[k];
275  for(k = 7; k > 0; k--)
276  ptr0[k] = ptr0[k - 1];
277  ptr0[0] = out[i];
278  out[i] += (- sum) >> 12;
279  }
280 
281  for(i = 0; i < 8; i++)
282  t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15;
283 
284  ptr0 = dec->tmp3;
285  for(i = 0; i < 60; i++){
286  int sum = out[i] * (1 << 12);
287  for(k = 0; k < 8; k++)
288  sum += ptr0[k] * t[k];
289  for(k = 7; k > 0; k--)
290  ptr0[k] = ptr0[k - 1];
291  ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
292 
293  sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
294  sum = sum - (sum >> 3);
295  out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
296  }
297 }
298 
300 {
301  int i;
302 
303  for(i = 0; i < 8; i++)
304  c->prevfilt[i] = c->cvector[i];
305 }
306 
307 static int truespeech_decode_frame(AVCodecContext *avctx, void *data,
308  int *got_frame_ptr, AVPacket *avpkt)
309 {
310  AVFrame *frame = data;
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++) {
344  truespeech_place_pulses (c, samples, i);
345  truespeech_update_filters(c, samples, i);
346  truespeech_synth (c, samples, i);
347  samples += 60;
348  }
349 
351  }
352 
353  *got_frame_ptr = 1;
354 
355  return buf_size;
356 }
357 
359  .name = "truespeech",
360  .long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),
361  .type = AVMEDIA_TYPE_AUDIO,
363  .priv_data_size = sizeof(TSContext),
366  .capabilities = AV_CODEC_CAP_DR1,
367 };
#define av_cold
Definition: attributes.h:88
uint8_t
Libavcodec external API header.
static av_cold int init(AVCodecContext *avctx)
Definition: avrndec.c:31
static av_always_inline void filter(int16_t *output, ptrdiff_t out_stride, const int16_t *low, ptrdiff_t low_stride, const int16_t *high, ptrdiff_t high_stride, int len, int clip)
Definition: cfhddsp.c:27
audio channel layout utility functions
#define av_clip
Definition: common.h:122
int ff_get_buffer(AVCodecContext *avctx, AVFrame *frame, int flags)
Get a buffer for a frame.
Definition: decode.c:1900
static void decode(AVCodecContext *dec_ctx, AVPacket *pkt, AVFrame *frame, FILE *outfile)
Definition: decode_audio.c:71
static AVFrame * frame
int
bitstream reader API header.
static unsigned int get_bits_long(GetBitContext *s, int n)
Read 0-32 bits.
Definition: get_bits.h:546
static unsigned int get_bits1(GetBitContext *s)
Definition: get_bits.h:498
static unsigned int get_bits(GetBitContext *s, int n)
Read 1-25 bits.
Definition: get_bits.h:379
static int init_get_bits(GetBitContext *s, const uint8_t *buffer, int bit_size)
Initialize GetBitContext.
Definition: get_bits.h:659
#define AV_CH_LAYOUT_MONO
#define AV_CODEC_CAP_DR1
Codec uses get_buffer() or get_encode_buffer() for allocating buffers and supports custom allocators.
Definition: codec.h:52
@ AV_CODEC_ID_TRUESPEECH
Definition: codec_id.h:445
#define AVERROR_PATCHWELCOME
Not yet implemented in FFmpeg, patches welcome.
Definition: error.h:62
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:194
#define DECLARE_ALIGNED(n, t, v)
Declare a variable that is aligned in memory.
Definition: mem.h:117
@ AVMEDIA_TYPE_AUDIO
Definition: avutil.h:202
@ AV_SAMPLE_FMT_S16
signed 16 bits
Definition: samplefmt.h:61
int i
Definition: input.c:407
av_cold void ff_bswapdsp_init(BswapDSPContext *c)
Definition: bswapdsp.c:49
common internal API header
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification.
Definition: internal.h:117
const char data[16]
Definition: mxf.c:142
main external API structure.
Definition: avcodec.h:536
enum AVSampleFormat sample_fmt
audio sample format
Definition: avcodec.h:1204
int channels
number of audio channels
Definition: avcodec.h:1197
uint64_t channel_layout
Audio channel layout.
Definition: avcodec.h:1247
void * priv_data
Definition: avcodec.h:563
AVCodec.
Definition: codec.h:197
const char * name
Name of the codec implementation.
Definition: codec.h:204
This structure describes decoded (raw) audio or video data.
Definition: frame.h:318
int nb_samples
number of audio samples (per channel) described by this frame
Definition: frame.h:384
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:332
This structure stores compressed data.
Definition: packet.h:346
int size
Definition: packet.h:370
uint8_t * data
Definition: packet.h:369
void(* bswap_buf)(uint32_t *dst, const uint32_t *src, int w)
Definition: bswapdsp.h:25
TrueSpeech decoder context.
Definition: truespeech.c:40
int offset1[2]
8-bit value, used in one copying offset
Definition: truespeech.c:45
int16_t newvec[60]
Definition: truespeech.c:59
int filtbuf[146]
Definition: truespeech.c:52
int16_t tmp3[8]
Definition: truespeech.c:56
int offset2[4]
7-bit value, encodes offsets for copying and for two-point filter
Definition: truespeech.c:46
int flag
1-bit flag, shows how to choose filters
Definition: truespeech.c:50
int16_t tmp2[8]
Definition: truespeech.c:55
int pulseval[4]
7x2-bit pulse values
Definition: truespeech.c:49
int filtval
Definition: truespeech.c:58
int pulseoff[4]
4-bit offset of pulse values block
Definition: truespeech.c:47
int16_t tmp1[8]
Definition: truespeech.c:54
int16_t cvector[8]
Definition: truespeech.c:57
int16_t filters[32]
Definition: truespeech.c:60
int16_t vector[8]
input vector: 5/5/4/4/4/3/3/3
Definition: truespeech.c:44
uint8_t buffer[32]
Definition: truespeech.c:43
int pulsepos[4]
27-bit variable, encodes 7 pulse positions
Definition: truespeech.c:48
int prevfilt[8]
Definition: truespeech.c:53
BswapDSPContext bdsp
Definition: truespeech.c:41
#define avpriv_request_sample(...)
#define av_log(a,...)
static uint8_t tmp[11]
Definition: aes_ctr.c:27
FILE * out
Definition: movenc.c:54
static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
Definition: truespeech.c:168
static av_cold int truespeech_decode_init(AVCodecContext *avctx)
Definition: truespeech.c:63
AVCodec ff_truespeech_decoder
Definition: truespeech.c:358
static void truespeech_save_prevvec(TSContext *c)
Definition: truespeech.c:299
static int truespeech_decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr, AVPacket *avpkt)
Definition: truespeech.c:307
static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
Definition: truespeech.c:194
static void truespeech_correlate_filter(TSContext *dec)
Definition: truespeech.c:128
static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
Definition: truespeech.c:248
static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
Definition: truespeech.c:80
static void truespeech_filters_merge(TSContext *dec)
Definition: truespeech.c:147
static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
Definition: truespeech.c:237
static const int16_t ts_order2_coeffs[25 *2]
static const int16_t *const ts_codebook[8]
static const int16_t ts_decay_35_64[8]
static const int16_t ts_pulse_values[120]
static const int16_t ts_decay_3_4[8]
static const int16_t ts_pulse_scales[64]
static const int16_t ts_decay_994_1000[8]
static double c[64]