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 Libav.
6  *
7  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "libavutil/intreadwrite.h"
23 #include "avcodec.h"
24 #include "internal.h"
25 #include "dsputil.h"
26 #include "get_bits.h"
27 
28 #include "truespeech_data.h"
37 typedef struct {
40  /* input data */
41  uint8_t buffer[32];
42  int16_t vector[8];
43  int offset1[2];
44  int offset2[4];
45  int pulseoff[4];
46  int pulsepos[4];
47  int pulseval[4];
48  int flag;
49  /* temporary data */
50  int filtbuf[146]; // some big vector used for storing filters
51  int prevfilt[8]; // filter from previous frame
52  int16_t tmp1[8]; // coefficients for adding to out
53  int16_t tmp2[8]; // coefficients for adding to out
54  int16_t tmp3[8]; // coefficients for adding to out
55  int16_t cvector[8]; // correlated input vector
56  int filtval; // gain value for one function
57  int16_t newvec[60]; // tmp vector
58  int16_t filters[32]; // filters for every subframe
59 } TSContext;
60 
62 {
63  TSContext *c = avctx->priv_data;
64 
65  if (avctx->channels != 1) {
66  av_log_ask_for_sample(avctx, "Unsupported channel count: %d\n", avctx->channels);
67  return AVERROR(EINVAL);
68  }
69 
71 
72  dsputil_init(&c->dsp, avctx);
73 
75  avctx->coded_frame = &c->frame;
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->dsp.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]) +
138  (dec->cvector[j] << 15) + 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 = (const int16_t*)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 = (const int16_t*)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 = (const int16_t*)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] * ptr1[k];
261  sum = (sum + (out[i] << 12) + 0x800) >> 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] = ((out[i] << 12) - 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] << 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 
308 static int truespeech_decode_frame(AVCodecContext *avctx, void *data,
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  c->frame.nb_samples = iterations * 240;
329  if ((ret = ff_get_buffer(avctx, &c->frame)) < 0) {
330  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
331  return ret;
332  }
333  samples = (int16_t *)c->frame.data[0];
334 
335  memset(samples, 0, iterations * 240 * sizeof(*samples));
336 
337  for(j = 0; j < iterations; j++) {
338  truespeech_read_frame(c, buf);
339  buf += 32;
340 
343 
344  for(i = 0; i < 4; i++) {
346  truespeech_place_pulses (c, samples, i);
347  truespeech_update_filters(c, samples, i);
348  truespeech_synth (c, samples, i);
349  samples += 60;
350  }
351 
353  }
354 
355  *got_frame_ptr = 1;
356  *(AVFrame *)data = c->frame;
357 
358  return buf_size;
359 }
360 
362  .name = "truespeech",
363  .type = AVMEDIA_TYPE_AUDIO,
364  .id = CODEC_ID_TRUESPEECH,
365  .priv_data_size = sizeof(TSContext),
368  .capabilities = CODEC_CAP_DR1,
369  .long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),
370 };