7747fca6b20fe75b947d02100c14bcf988c0ab21
[libav.git] / libavcodec / truespeech.c
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"
28 /**
29 * @file
30 * TrueSpeech decoder.
31 */
32
33 /**
34 * TrueSpeech decoder context
35 */
36 typedef struct {
37 DSPContext dsp;
38 /* input data */
39 uint8_t buffer[32];
40 int16_t vector[8]; ///< input vector: 5/5/4/4/4/3/3/3
41 int offset1[2]; ///< 8-bit value, used in one copying offset
42 int offset2[4]; ///< 7-bit value, encodes offsets for copying and for two-point filter
43 int pulseoff[4]; ///< 4-bit offset of pulse values block
44 int pulsepos[4]; ///< 27-bit variable, encodes 7 pulse positions
45 int pulseval[4]; ///< 7x2-bit pulse values
46 int flag; ///< 1-bit flag, shows how to choose filters
47 /* temporary data */
48 int filtbuf[146]; // some big vector used for storing filters
49 int prevfilt[8]; // filter from previous frame
50 int16_t tmp1[8]; // coefficients for adding to out
51 int16_t tmp2[8]; // coefficients for adding to out
52 int16_t tmp3[8]; // coefficients for adding to out
53 int16_t cvector[8]; // correlated input vector
54 int filtval; // gain value for one function
55 int16_t newvec[60]; // tmp vector
56 int16_t filters[32]; // filters for every subframe
57 } TSContext;
58
59 static av_cold int truespeech_decode_init(AVCodecContext * avctx)
60 {
61 TSContext *c = avctx->priv_data;
62
63 if (avctx->channels != 1) {
64 av_log_ask_for_sample(avctx, "Unsupported channel count: %d\n", avctx->channels);
65 return AVERROR(EINVAL);
66 }
67
68 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
69
70 dsputil_init(&c->dsp, avctx);
71
72 return 0;
73 }
74
75 static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
76 {
77 GetBitContext gb;
78
79 dec->dsp.bswap_buf((uint32_t *)dec->buffer, (const uint32_t *)input, 8);
80 init_get_bits(&gb, dec->buffer, 32 * 8);
81
82 dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)];
83 dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)];
84 dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)];
85 dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)];
86 dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)];
87 dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)];
88 dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)];
89 dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)];
90 dec->flag = get_bits1(&gb);
91
92 dec->offset1[0] = get_bits(&gb, 4) << 4;
93 dec->offset2[3] = get_bits(&gb, 7);
94 dec->offset2[2] = get_bits(&gb, 7);
95 dec->offset2[1] = get_bits(&gb, 7);
96 dec->offset2[0] = get_bits(&gb, 7);
97
98 dec->offset1[1] = get_bits(&gb, 4);
99 dec->pulseval[1] = get_bits(&gb, 14);
100 dec->pulseval[0] = get_bits(&gb, 14);
101
102 dec->offset1[1] |= get_bits(&gb, 4) << 4;
103 dec->pulseval[3] = get_bits(&gb, 14);
104 dec->pulseval[2] = get_bits(&gb, 14);
105
106 dec->offset1[0] |= get_bits1(&gb);
107 dec->pulsepos[0] = get_bits_long(&gb, 27);
108 dec->pulseoff[0] = get_bits(&gb, 4);
109
110 dec->offset1[0] |= get_bits1(&gb) << 1;
111 dec->pulsepos[1] = get_bits_long(&gb, 27);
112 dec->pulseoff[1] = get_bits(&gb, 4);
113
114 dec->offset1[0] |= get_bits1(&gb) << 2;
115 dec->pulsepos[2] = get_bits_long(&gb, 27);
116 dec->pulseoff[2] = get_bits(&gb, 4);
117
118 dec->offset1[0] |= get_bits1(&gb) << 3;
119 dec->pulsepos[3] = get_bits_long(&gb, 27);
120 dec->pulseoff[3] = get_bits(&gb, 4);
121 }
122
123 static void truespeech_correlate_filter(TSContext *dec)
124 {
125 int16_t tmp[8];
126 int i, j;
127
128 for(i = 0; i < 8; i++){
129 if(i > 0){
130 memcpy(tmp, dec->cvector, i * sizeof(*tmp));
131 for(j = 0; j < i; j++)
132 dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) +
133 (dec->cvector[j] << 15) + 0x4000) >> 15;
134 }
135 dec->cvector[i] = (8 - dec->vector[i]) >> 3;
136 }
137 for(i = 0; i < 8; i++)
138 dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15;
139
140 dec->filtval = dec->vector[0];
141 }
142
143 static void truespeech_filters_merge(TSContext *dec)
144 {
145 int i;
146
147 if(!dec->flag){
148 for(i = 0; i < 8; i++){
149 dec->filters[i + 0] = dec->prevfilt[i];
150 dec->filters[i + 8] = dec->prevfilt[i];
151 }
152 }else{
153 for(i = 0; i < 8; i++){
154 dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
155 dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
156 }
157 }
158 for(i = 0; i < 8; i++){
159 dec->filters[i + 16] = dec->cvector[i];
160 dec->filters[i + 24] = dec->cvector[i];
161 }
162 }
163
164 static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
165 {
166 int16_t tmp[146 + 60], *ptr0, *ptr1;
167 const int16_t *filter;
168 int i, t, off;
169
170 t = dec->offset2[quart];
171 if(t == 127){
172 memset(dec->newvec, 0, 60 * sizeof(*dec->newvec));
173 return;
174 }
175 for(i = 0; i < 146; i++)
176 tmp[i] = dec->filtbuf[i];
177 off = (t / 25) + dec->offset1[quart >> 1] + 18;
178 ptr0 = tmp + 145 - off;
179 ptr1 = tmp + 146;
180 filter = (const int16_t*)ts_order2_coeffs + (t % 25) * 2;
181 for(i = 0; i < 60; i++){
182 t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
183 ptr0++;
184 dec->newvec[i] = t;
185 ptr1[i] = t;
186 }
187 }
188
189 static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
190 {
191 int16_t tmp[7];
192 int i, j, t;
193 const int16_t *ptr1;
194 int16_t *ptr2;
195 int coef;
196
197 memset(out, 0, 60 * sizeof(*out));
198 for(i = 0; i < 7; i++) {
199 t = dec->pulseval[quart] & 3;
200 dec->pulseval[quart] >>= 2;
201 tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t];
202 }
203
204 coef = dec->pulsepos[quart] >> 15;
205 ptr1 = (const int16_t*)ts_pulse_values + 30;
206 ptr2 = tmp;
207 for(i = 0, j = 3; (i < 30) && (j > 0); i++){
208 t = *ptr1++;
209 if(coef >= t)
210 coef -= t;
211 else{
212 out[i] = *ptr2++;
213 ptr1 += 30;
214 j--;
215 }
216 }
217 coef = dec->pulsepos[quart] & 0x7FFF;
218 ptr1 = (const int16_t*)ts_pulse_values;
219 for(i = 30, j = 4; (i < 60) && (j > 0); i++){
220 t = *ptr1++;
221 if(coef >= t)
222 coef -= t;
223 else{
224 out[i] = *ptr2++;
225 ptr1 += 30;
226 j--;
227 }
228 }
229
230 }
231
232 static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
233 {
234 int i;
235
236 for(i = 0; i < 86; i++)
237 dec->filtbuf[i] = dec->filtbuf[i + 60];
238 for(i = 0; i < 60; i++){
239 dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
240 out[i] += dec->newvec[i];
241 }
242 }
243
244 static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
245 {
246 int i,k;
247 int t[8];
248 int16_t *ptr0, *ptr1;
249
250 ptr0 = dec->tmp1;
251 ptr1 = dec->filters + quart * 8;
252 for(i = 0; i < 60; i++){
253 int sum = 0;
254 for(k = 0; k < 8; k++)
255 sum += ptr0[k] * ptr1[k];
256 sum = (sum + (out[i] << 12) + 0x800) >> 12;
257 out[i] = av_clip(sum, -0x7FFE, 0x7FFE);
258 for(k = 7; k > 0; k--)
259 ptr0[k] = ptr0[k - 1];
260 ptr0[0] = out[i];
261 }
262
263 for(i = 0; i < 8; i++)
264 t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15;
265
266 ptr0 = dec->tmp2;
267 for(i = 0; i < 60; i++){
268 int sum = 0;
269 for(k = 0; k < 8; k++)
270 sum += ptr0[k] * t[k];
271 for(k = 7; k > 0; k--)
272 ptr0[k] = ptr0[k - 1];
273 ptr0[0] = out[i];
274 out[i] = ((out[i] << 12) - sum) >> 12;
275 }
276
277 for(i = 0; i < 8; i++)
278 t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15;
279
280 ptr0 = dec->tmp3;
281 for(i = 0; i < 60; i++){
282 int sum = out[i] << 12;
283 for(k = 0; k < 8; k++)
284 sum += ptr0[k] * t[k];
285 for(k = 7; k > 0; k--)
286 ptr0[k] = ptr0[k - 1];
287 ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
288
289 sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
290 sum = sum - (sum >> 3);
291 out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
292 }
293 }
294
295 static void truespeech_save_prevvec(TSContext *c)
296 {
297 int i;
298
299 for(i = 0; i < 8; i++)
300 c->prevfilt[i] = c->cvector[i];
301 }
302
303 static int truespeech_decode_frame(AVCodecContext *avctx,
304 void *data, int *data_size,
305 AVPacket *avpkt)
306 {
307 const uint8_t *buf = avpkt->data;
308 int buf_size = avpkt->size;
309 TSContext *c = avctx->priv_data;
310
311 int i, j;
312 short *samples = data;
313 int iterations, out_size;
314
315 iterations = buf_size / 32;
316
317 if (!iterations) {
318 av_log(avctx, AV_LOG_ERROR,
319 "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
320 return -1;
321 }
322
323 out_size = iterations * 240 * av_get_bytes_per_sample(avctx->sample_fmt);
324 if (*data_size < out_size) {
325 av_log(avctx, AV_LOG_ERROR, "Output buffer is too small\n");
326 return AVERROR(EINVAL);
327 }
328
329 memset(samples, 0, out_size);
330
331 for(j = 0; j < iterations; j++) {
332 truespeech_read_frame(c, buf);
333 buf += 32;
334
335 truespeech_correlate_filter(c);
336 truespeech_filters_merge(c);
337
338 for(i = 0; i < 4; i++) {
339 truespeech_apply_twopoint_filter(c, i);
340 truespeech_place_pulses (c, samples, i);
341 truespeech_update_filters(c, samples, i);
342 truespeech_synth (c, samples, i);
343 samples += 60;
344 }
345
346 truespeech_save_prevvec(c);
347 }
348
349 *data_size = out_size;
350
351 return buf_size;
352 }
353
354 AVCodec ff_truespeech_decoder = {
355 .name = "truespeech",
356 .type = AVMEDIA_TYPE_AUDIO,
357 .id = CODEC_ID_TRUESPEECH,
358 .priv_data_size = sizeof(TSContext),
359 .init = truespeech_decode_init,
360 .decode = truespeech_decode_frame,
361 .long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),
362 };