|
808
|
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 #include "avcodec.h"
|
|
|
22
|
|
|
23 #include "truespeech_data.h"
|
|
|
24 /**
|
|
|
25 * @file truespeech.c
|
|
|
26 * TrueSpeech decoder.
|
|
|
27 */
|
|
|
28
|
|
|
29 /**
|
|
|
30 * TrueSpeech decoder context
|
|
|
31 */
|
|
|
32 typedef struct {
|
|
|
33 /* input data */
|
|
|
34 int16_t vector[8]; //< input vector: 5/5/4/4/4/3/3/3
|
|
|
35 int offset1[2]; //< 8-bit value, used in one copying offset
|
|
|
36 int offset2[4]; //< 7-bit value, encodes offsets for copying and for two-point filter
|
|
|
37 int pulseoff[4]; //< 4-bit offset of pulse values block
|
|
|
38 int pulsepos[4]; //< 27-bit variable, encodes 7 pulse positions
|
|
|
39 int pulseval[4]; //< 7x2-bit pulse values
|
|
|
40 int flag; //< 1-bit flag, shows how to choose filters
|
|
|
41 /* temporary data */
|
|
|
42 int filtbuf[146]; // some big vector used for storing filters
|
|
|
43 int prevfilt[8]; // filter from previous frame
|
|
|
44 int16_t tmp1[8]; // coefficients for adding to out
|
|
|
45 int16_t tmp2[8]; // coefficients for adding to out
|
|
|
46 int16_t tmp3[8]; // coefficients for adding to out
|
|
|
47 int16_t cvector[8]; // correlated input vector
|
|
|
48 int filtval; // gain value for one function
|
|
|
49 int16_t newvec[60]; // tmp vector
|
|
|
50 int16_t filters[32]; // filters for every subframe
|
|
|
51 } TSContext;
|
|
|
52
|
|
|
53 static int truespeech_decode_init(AVCodecContext * avctx)
|
|
|
54 {
|
|
|
55 // TSContext *c = avctx->priv_data;
|
|
|
56
|
|
|
57 return 0;
|
|
|
58 }
|
|
|
59
|
|
|
60 static void truespeech_read_frame(TSContext *dec, uint8_t *input)
|
|
|
61 {
|
|
|
62 uint32_t t;
|
|
|
63
|
|
|
64 /* first dword */
|
|
|
65 t = LE_32(input);
|
|
|
66 input += 4;
|
|
|
67
|
|
|
68 dec->flag = t & 1;
|
|
|
69
|
|
|
70 dec->vector[0] = ts_codebook[0][(t >> 1) & 0x1F];
|
|
|
71 dec->vector[1] = ts_codebook[1][(t >> 6) & 0x1F];
|
|
|
72 dec->vector[2] = ts_codebook[2][(t >> 11) & 0xF];
|
|
|
73 dec->vector[3] = ts_codebook[3][(t >> 15) & 0xF];
|
|
|
74 dec->vector[4] = ts_codebook[4][(t >> 19) & 0xF];
|
|
|
75 dec->vector[5] = ts_codebook[5][(t >> 23) & 0x7];
|
|
|
76 dec->vector[6] = ts_codebook[6][(t >> 26) & 0x7];
|
|
|
77 dec->vector[7] = ts_codebook[7][(t >> 29) & 0x7];
|
|
|
78
|
|
|
79 /* second dword */
|
|
|
80 t = LE_32(input);
|
|
|
81 input += 4;
|
|
|
82
|
|
|
83 dec->offset2[0] = (t >> 0) & 0x7F;
|
|
|
84 dec->offset2[1] = (t >> 7) & 0x7F;
|
|
|
85 dec->offset2[2] = (t >> 14) & 0x7F;
|
|
|
86 dec->offset2[3] = (t >> 21) & 0x7F;
|
|
|
87
|
|
|
88 dec->offset1[0] = ((t >> 28) & 0xF) << 4;
|
|
|
89
|
|
|
90 /* third dword */
|
|
|
91 t = LE_32(input);
|
|
|
92 input += 4;
|
|
|
93
|
|
|
94 dec->pulseval[0] = (t >> 0) & 0x3FFF;
|
|
|
95 dec->pulseval[1] = (t >> 14) & 0x3FFF;
|
|
|
96
|
|
|
97 dec->offset1[1] = (t >> 28) & 0x0F;
|
|
|
98
|
|
|
99 /* fourth dword */
|
|
|
100 t = LE_32(input);
|
|
|
101 input += 4;
|
|
|
102
|
|
|
103 dec->pulseval[2] = (t >> 0) & 0x3FFF;
|
|
|
104 dec->pulseval[3] = (t >> 14) & 0x3FFF;
|
|
|
105
|
|
|
106 dec->offset1[1] |= ((t >> 28) & 0x0F) << 4;
|
|
|
107
|
|
|
108 /* fifth dword */
|
|
|
109 t = LE_32(input);
|
|
|
110 input += 4;
|
|
|
111
|
|
|
112 dec->pulsepos[0] = (t >> 4) & 0x7FFFFFF;
|
|
|
113
|
|
|
114 dec->pulseoff[0] = (t >> 0) & 0xF;
|
|
|
115
|
|
|
116 dec->offset1[0] |= (t >> 31) & 1;
|
|
|
117
|
|
|
118 /* sixth dword */
|
|
|
119 t = LE_32(input);
|
|
|
120 input += 4;
|
|
|
121
|
|
|
122 dec->pulsepos[1] = (t >> 4) & 0x7FFFFFF;
|
|
|
123
|
|
|
124 dec->pulseoff[1] = (t >> 0) & 0xF;
|
|
|
125
|
|
|
126 dec->offset1[0] |= ((t >> 31) & 1) << 1;
|
|
|
127
|
|
|
128 /* seventh dword */
|
|
|
129 t = LE_32(input);
|
|
|
130 input += 4;
|
|
|
131
|
|
|
132 dec->pulsepos[2] = (t >> 4) & 0x7FFFFFF;
|
|
|
133
|
|
|
134 dec->pulseoff[2] = (t >> 0) & 0xF;
|
|
|
135
|
|
|
136 dec->offset1[0] |= ((t >> 31) & 1) << 2;
|
|
|
137
|
|
|
138 /* eighth dword */
|
|
|
139 t = LE_32(input);
|
|
|
140 input += 4;
|
|
|
141
|
|
|
142 dec->pulsepos[3] = (t >> 4) & 0x7FFFFFF;
|
|
|
143
|
|
|
144 dec->pulseoff[3] = (t >> 0) & 0xF;
|
|
|
145
|
|
|
146 dec->offset1[0] |= ((t >> 31) & 1) << 3;
|
|
|
147
|
|
|
148 }
|
|
|
149
|
|
|
150 static void truespeech_correlate_filter(TSContext *dec)
|
|
|
151 {
|
|
|
152 int16_t tmp[8];
|
|
|
153 int i, j;
|
|
|
154
|
|
|
155 for(i = 0; i < 8; i++){
|
|
|
156 if(i > 0){
|
|
|
157 memcpy(tmp, dec->cvector, i * 2);
|
|
|
158 for(j = 0; j < i; j++)
|
|
|
159 dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) +
|
|
|
160 (dec->cvector[j] << 15) + 0x4000) >> 15;
|
|
|
161 }
|
|
|
162 dec->cvector[i] = (8 - dec->vector[i]) >> 3;
|
|
|
163 }
|
|
|
164 for(i = 0; i < 8; i++)
|
|
|
165 dec->cvector[i] = (dec->cvector[i] * ts_230[i]) >> 15;
|
|
|
166
|
|
|
167 dec->filtval = dec->vector[0];
|
|
|
168 }
|
|
|
169
|
|
|
170 static void truespeech_filters_merge(TSContext *dec)
|
|
|
171 {
|
|
|
172 int i;
|
|
|
173
|
|
|
174 if(!dec->flag){
|
|
|
175 for(i = 0; i < 8; i++){
|
|
|
176 dec->filters[i + 0] = dec->prevfilt[i];
|
|
|
177 dec->filters[i + 8] = dec->prevfilt[i];
|
|
|
178 }
|
|
|
179 }else{
|
|
|
180 for(i = 0; i < 8; i++){
|
|
|
181 dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
|
|
|
182 dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
|
|
|
183 }
|
|
|
184 }
|
|
|
185 for(i = 0; i < 8; i++){
|
|
|
186 dec->filters[i + 16] = dec->cvector[i];
|
|
|
187 dec->filters[i + 24] = dec->cvector[i];
|
|
|
188 }
|
|
|
189 }
|
|
|
190
|
|
|
191 static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
|
|
|
192 {
|
|
|
193 int16_t tmp[146 + 60], *ptr0, *ptr1;
|
|
|
194 const int16_t *filter;
|
|
|
195 int i, t, off;
|
|
|
196
|
|
|
197 t = dec->offset2[quart];
|
|
|
198 if(t == 127){
|
|
|
199 memset(dec->newvec, 0, 60 * 2);
|
|
|
200 return;
|
|
|
201 }
|
|
|
202 for(i = 0; i < 146; i++)
|
|
|
203 tmp[i] = dec->filtbuf[i];
|
|
|
204 off = (t / 25) + dec->offset1[quart >> 1] + 18;
|
|
|
205 ptr0 = tmp + 145 - off;
|
|
|
206 ptr1 = tmp + 146;
|
|
|
207 filter = (const int16_t*)ts_240 + (t % 25) * 2;
|
|
|
208 for(i = 0; i < 60; i++){
|
|
|
209 t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
|
|
|
210 ptr0++;
|
|
|
211 dec->newvec[i] = t;
|
|
|
212 ptr1[i] = t;
|
|
|
213 }
|
|
|
214 }
|
|
|
215
|
|
|
216 static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
|
|
|
217 {
|
|
|
218 int16_t tmp[7];
|
|
|
219 int i, j, t;
|
|
|
220 const int16_t *ptr1;
|
|
|
221 int16_t *ptr2;
|
|
|
222 int coef;
|
|
|
223
|
|
|
224 memset(out, 0, 60 * 2);
|
|
|
225 for(i = 0; i < 7; i++) {
|
|
|
226 t = dec->pulseval[quart] & 3;
|
|
|
227 dec->pulseval[quart] >>= 2;
|
|
|
228 tmp[6 - i] = ts_562[dec->pulseoff[quart] * 4 + t];
|
|
|
229 }
|
|
|
230
|
|
|
231 coef = dec->pulsepos[quart] >> 15;
|
|
|
232 ptr1 = (const int16_t*)ts_140 + 30;
|
|
|
233 ptr2 = tmp;
|
|
|
234 for(i = 0, j = 3; (i < 30) && (j > 0); i++){
|
|
|
235 t = *ptr1++;
|
|
|
236 if(coef >= t)
|
|
|
237 coef -= t;
|
|
|
238 else{
|
|
|
239 out[i] = *ptr2++;
|
|
|
240 ptr1 += 30;
|
|
|
241 j--;
|
|
|
242 }
|
|
|
243 }
|
|
|
244 coef = dec->pulsepos[quart] & 0x7FFF;
|
|
|
245 ptr1 = (const int16_t*)ts_140;
|
|
|
246 for(i = 30, j = 4; (i < 60) && (j > 0); i++){
|
|
|
247 t = *ptr1++;
|
|
|
248 if(coef >= t)
|
|
|
249 coef -= t;
|
|
|
250 else{
|
|
|
251 out[i] = *ptr2++;
|
|
|
252 ptr1 += 30;
|
|
|
253 j--;
|
|
|
254 }
|
|
|
255 }
|
|
|
256
|
|
|
257 }
|
|
|
258
|
|
|
259 static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
|
|
|
260 {
|
|
|
261 int i;
|
|
|
262
|
|
|
263 for(i = 0; i < 86; i++)
|
|
|
264 dec->filtbuf[i] = dec->filtbuf[i + 60];
|
|
|
265 for(i = 0; i < 60; i++){
|
|
|
266 dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
|
|
|
267 out[i] += dec->newvec[i];
|
|
|
268 }
|
|
|
269 }
|
|
|
270
|
|
|
271 static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
|
|
|
272 {
|
|
|
273 int i,k;
|
|
|
274 int t[8];
|
|
|
275 int16_t *ptr0, *ptr1;
|
|
|
276
|
|
|
277 ptr0 = dec->tmp1;
|
|
|
278 ptr1 = dec->filters + quart * 8;
|
|
|
279 for(i = 0; i < 60; i++){
|
|
|
280 int sum = 0;
|
|
|
281 for(k = 0; k < 8; k++)
|
|
|
282 sum += ptr0[k] * ptr1[k];
|
|
|
283 sum = (sum + (out[i] << 12) + 0x800) >> 12;
|
|
|
284 out[i] = clip(sum, -0x7FFE, 0x7FFE);
|
|
|
285 for(k = 7; k > 0; k--)
|
|
|
286 ptr0[k] = ptr0[k - 1];
|
|
|
287 ptr0[0] = out[i];
|
|
|
288 }
|
|
|
289
|
|
|
290 for(i = 0; i < 8; i++)
|
|
|
291 t[i] = (ts_5E2[i] * ptr1[i]) >> 15;
|
|
|
292
|
|
|
293 ptr0 = dec->tmp2;
|
|
|
294 for(i = 0; i < 60; i++){
|
|
|
295 int sum = 0;
|
|
|
296 for(k = 0; k < 8; k++)
|
|
|
297 sum += ptr0[k] * t[k];
|
|
|
298 for(k = 7; k > 0; k--)
|
|
|
299 ptr0[k] = ptr0[k - 1];
|
|
|
300 ptr0[0] = out[i];
|
|
|
301 out[i] = ((out[i] << 12) - sum) >> 12;
|
|
|
302 }
|
|
|
303
|
|
|
304 for(i = 0; i < 8; i++)
|
|
|
305 t[i] = (ts_5F2[i] * ptr1[i]) >> 15;
|
|
|
306
|
|
|
307 ptr0 = dec->tmp3;
|
|
|
308 for(i = 0; i < 60; i++){
|
|
|
309 int sum = out[i] << 12;
|
|
|
310 for(k = 0; k < 8; k++)
|
|
|
311 sum += ptr0[k] * t[k];
|
|
|
312 for(k = 7; k > 0; k--)
|
|
|
313 ptr0[k] = ptr0[k - 1];
|
|
|
314 ptr0[0] = clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
|
|
|
315
|
|
|
316 sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
|
|
|
317 sum = sum - (sum >> 3);
|
|
|
318 out[i] = clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
|
|
|
319 }
|
|
|
320 }
|
|
|
321
|
|
|
322 static void truespeech_save_prevvec(TSContext *c)
|
|
|
323 {
|
|
|
324 int i;
|
|
|
325
|
|
|
326 for(i = 0; i < 8; i++)
|
|
|
327 c->prevfilt[i] = c->cvector[i];
|
|
|
328 }
|
|
|
329
|
|
|
330 static int truespeech_decode_frame(AVCodecContext *avctx,
|
|
|
331 void *data, int *data_size,
|
|
|
332 uint8_t *buf, int buf_size)
|
|
|
333 {
|
|
|
334 TSContext *c = avctx->priv_data;
|
|
|
335
|
|
|
336 int i;
|
|
|
337 short *samples = data;
|
|
|
338 int consumed = 0;
|
|
|
339 int16_t out_buf[240];
|
|
|
340
|
|
|
341 if (!buf_size)
|
|
|
342 return 0;
|
|
|
343
|
|
|
344 while (consumed < buf_size) {
|
|
|
345 truespeech_read_frame(c, buf + consumed);
|
|
|
346 consumed += 32;
|
|
|
347
|
|
|
348 truespeech_correlate_filter(c);
|
|
|
349 truespeech_filters_merge(c);
|
|
|
350
|
|
|
351 memset(out_buf, 0, 240 * 2);
|
|
|
352 for(i = 0; i < 4; i++) {
|
|
|
353 truespeech_apply_twopoint_filter(c, i);
|
|
|
354 truespeech_place_pulses(c, out_buf + i * 60, i);
|
|
|
355 truespeech_update_filters(c, out_buf + i * 60, i);
|
|
|
356 truespeech_synth(c, out_buf + i * 60, i);
|
|
|
357 }
|
|
|
358
|
|
|
359 truespeech_save_prevvec(c);
|
|
|
360
|
|
|
361 /* finally output decoded frame */
|
|
|
362 for(i = 0; i < 240; i++)
|
|
|
363 *samples++ = out_buf[i];
|
|
|
364
|
|
|
365 }
|
|
|
366
|
|
|
367 *data_size = consumed * 15;
|
|
|
368
|
|
|
369 return buf_size;
|
|
|
370 }
|
|
|
371
|
|
|
372 AVCodec truespeech_decoder = {
|
|
|
373 "truespeech",
|
|
|
374 CODEC_TYPE_AUDIO,
|
|
|
375 CODEC_ID_TRUESPEECH,
|
|
|
376 sizeof(TSContext),
|
|
|
377 truespeech_decode_init,
|
|
|
378 NULL,
|
|
|
379 NULL,
|
|
|
380 truespeech_decode_frame,
|
|
|
381 };
|