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 "dsputil.h"
25 #include "get_bits.h"
26 
27 #include "truespeech_data.h"
36 typedef struct {
39  /* input data */
40  uint8_t buffer[32];
41  int16_t vector[8];
42  int offset1[2];
43  int offset2[4];
44  int pulseoff[4];
45  int pulsepos[4];
46  int pulseval[4];
47  int flag;
48  /* temporary data */
49  int filtbuf[146]; // some big vector used for storing filters
50  int prevfilt[8]; // filter from previous frame
51  int16_t tmp1[8]; // coefficients for adding to out
52  int16_t tmp2[8]; // coefficients for adding to out
53  int16_t tmp3[8]; // coefficients for adding to out
54  int16_t cvector[8]; // correlated input vector
55  int filtval; // gain value for one function
56  int16_t newvec[60]; // tmp vector
57  int16_t filters[32]; // filters for every subframe
58 } TSContext;
59 
61 {
62  TSContext *c = avctx->priv_data;
63 
64  if (avctx->channels != 1) {
65  av_log_ask_for_sample(avctx, "Unsupported channel count: %d\n", avctx->channels);
66  return AVERROR(EINVAL);
67  }
68 
70 
71  dsputil_init(&c->dsp, avctx);
72 
74  avctx->coded_frame = &c->frame;
75 
76  return 0;
77 }
78 
79 static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
80 {
81  GetBitContext gb;
82 
83  dec->dsp.bswap_buf((uint32_t *)dec->buffer, (const uint32_t *)input, 8);
84  init_get_bits(&gb, dec->buffer, 32 * 8);
85 
86  dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)];
87  dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)];
88  dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)];
89  dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)];
90  dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)];
91  dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)];
92  dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)];
93  dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)];
94  dec->flag = get_bits1(&gb);
95 
96  dec->offset1[0] = get_bits(&gb, 4) << 4;
97  dec->offset2[3] = get_bits(&gb, 7);
98  dec->offset2[2] = get_bits(&gb, 7);
99  dec->offset2[1] = get_bits(&gb, 7);
100  dec->offset2[0] = get_bits(&gb, 7);
101 
102  dec->offset1[1] = get_bits(&gb, 4);
103  dec->pulseval[1] = get_bits(&gb, 14);
104  dec->pulseval[0] = get_bits(&gb, 14);
105 
106  dec->offset1[1] |= get_bits(&gb, 4) << 4;
107  dec->pulseval[3] = get_bits(&gb, 14);
108  dec->pulseval[2] = get_bits(&gb, 14);
109 
110  dec->offset1[0] |= get_bits1(&gb);
111  dec->pulsepos[0] = get_bits_long(&gb, 27);
112  dec->pulseoff[0] = get_bits(&gb, 4);
113 
114  dec->offset1[0] |= get_bits1(&gb) << 1;
115  dec->pulsepos[1] = get_bits_long(&gb, 27);
116  dec->pulseoff[1] = get_bits(&gb, 4);
117 
118  dec->offset1[0] |= get_bits1(&gb) << 2;
119  dec->pulsepos[2] = get_bits_long(&gb, 27);
120  dec->pulseoff[2] = get_bits(&gb, 4);
121 
122  dec->offset1[0] |= get_bits1(&gb) << 3;
123  dec->pulsepos[3] = get_bits_long(&gb, 27);
124  dec->pulseoff[3] = get_bits(&gb, 4);
125 }
126 
128 {
129  int16_t tmp[8];
130  int i, j;
131 
132  for(i = 0; i < 8; i++){
133  if(i > 0){
134  memcpy(tmp, dec->cvector, i * sizeof(*tmp));
135  for(j = 0; j < i; j++)
136  dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) +
137  (dec->cvector[j] << 15) + 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 = (const int16_t*)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 = (const int16_t*)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 = (const int16_t*)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] * ptr1[k];
260  sum = (sum + (out[i] << 12) + 0x800) >> 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] = ((out[i] << 12) - 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] << 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  const uint8_t *buf = avpkt->data;
311  int buf_size = avpkt->size;
312  TSContext *c = avctx->priv_data;
313 
314  int i, j;
315  int16_t *samples;
316  int iterations, ret;
317 
318  iterations = buf_size / 32;
319 
320  if (!iterations) {
321  av_log(avctx, AV_LOG_ERROR,
322  "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
323  return -1;
324  }
325 
326  /* get output buffer */
327  c->frame.nb_samples = iterations * 240;
328  if ((ret = avctx->get_buffer(avctx, &c->frame)) < 0) {
329  av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
330  return ret;
331  }
332  samples = (int16_t *)c->frame.data[0];
333 
334  memset(samples, 0, iterations * 240 * sizeof(*samples));
335 
336  for(j = 0; j < iterations; j++) {
337  truespeech_read_frame(c, buf);
338  buf += 32;
339 
342 
343  for(i = 0; i < 4; i++) {
345  truespeech_place_pulses (c, samples, i);
346  truespeech_update_filters(c, samples, i);
347  truespeech_synth (c, samples, i);
348  samples += 60;
349  }
350 
352  }
353 
354  *got_frame_ptr = 1;
355  *(AVFrame *)data = c->frame;
356 
357  return buf_size;
358 }
359 
361  .name = "truespeech",
362  .type = AVMEDIA_TYPE_AUDIO,
363  .id = CODEC_ID_TRUESPEECH,
364  .priv_data_size = sizeof(TSContext),
367  .capabilities = CODEC_CAP_DR1,
368  .long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),
369 };