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
22 #include "libavutil/channel_layout.h"
23 #include "libavutil/intreadwrite.h"
24 #include "libavutil/mem_internal.h"
25
26 #include "avcodec.h"
27 #include "bswapdsp.h"
28 #include "codec_internal.h"
29 #include "get_bits.h"
30 #include "internal.h"
31
32 #include "truespeech_data.h"
33 /**
34 * @file
35 * TrueSpeech decoder.
36 */
37
38 /**
39 * TrueSpeech decoder context
40 */
41 typedef struct TSContext {
42 BswapDSPContext bdsp;
43 /* input data */
44 DECLARE_ALIGNED(16, uint8_t, buffer)[32];
45 int16_t vector[8]; ///< input vector: 5/5/4/4/4/3/3/3
46 int offset1[2]; ///< 8-bit value, used in one copying offset
47 int offset2[4]; ///< 7-bit value, encodes offsets for copying and for two-point filter
48 int pulseoff[4]; ///< 4-bit offset of pulse values block
49 int pulsepos[4]; ///< 27-bit variable, encodes 7 pulse positions
50 int pulseval[4]; ///< 7x2-bit pulse values
51 int flag; ///< 1-bit flag, shows how to choose filters
52 /* temporary data */
53 int filtbuf[146]; // some big vector used for storing filters
54 int prevfilt[8]; // filter from previous frame
55 int16_t tmp1[8]; // coefficients for adding to out
56 int16_t tmp2[8]; // coefficients for adding to out
57 int16_t tmp3[8]; // coefficients for adding to out
58 int16_t cvector[8]; // correlated input vector
59 int filtval; // gain value for one function
60 int16_t newvec[60]; // tmp vector
61 int16_t filters[32]; // filters for every subframe
62 } TSContext;
63
truespeech_decode_init(AVCodecContext * avctx)64 static av_cold int truespeech_decode_init(AVCodecContext * avctx)
65 {
66 TSContext *c = avctx->priv_data;
67
68 if (avctx->ch_layout.nb_channels != 1) {
69 avpriv_request_sample(avctx, "Channel count %d", avctx->ch_layout.nb_channels);
70 return AVERROR_PATCHWELCOME;
71 }
72
73 av_channel_layout_uninit(&avctx->ch_layout);
74 avctx->ch_layout = (AVChannelLayout)AV_CHANNEL_LAYOUT_MONO;
75 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
76
77 ff_bswapdsp_init(&c->bdsp);
78
79 return 0;
80 }
81
truespeech_read_frame(TSContext * dec,const uint8_t * input)82 static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
83 {
84 GetBitContext gb;
85
86 dec->bdsp.bswap_buf((uint32_t *) dec->buffer, (const uint32_t *) input, 8);
87 init_get_bits(&gb, dec->buffer, 32 * 8);
88
89 dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)];
90 dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)];
91 dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)];
92 dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)];
93 dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)];
94 dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)];
95 dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)];
96 dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)];
97 dec->flag = get_bits1(&gb);
98
99 dec->offset1[0] = get_bits(&gb, 4) << 4;
100 dec->offset2[3] = get_bits(&gb, 7);
101 dec->offset2[2] = get_bits(&gb, 7);
102 dec->offset2[1] = get_bits(&gb, 7);
103 dec->offset2[0] = get_bits(&gb, 7);
104
105 dec->offset1[1] = get_bits(&gb, 4);
106 dec->pulseval[1] = get_bits(&gb, 14);
107 dec->pulseval[0] = get_bits(&gb, 14);
108
109 dec->offset1[1] |= get_bits(&gb, 4) << 4;
110 dec->pulseval[3] = get_bits(&gb, 14);
111 dec->pulseval[2] = get_bits(&gb, 14);
112
113 dec->offset1[0] |= get_bits1(&gb);
114 dec->pulsepos[0] = get_bits_long(&gb, 27);
115 dec->pulseoff[0] = get_bits(&gb, 4);
116
117 dec->offset1[0] |= get_bits1(&gb) << 1;
118 dec->pulsepos[1] = get_bits_long(&gb, 27);
119 dec->pulseoff[1] = get_bits(&gb, 4);
120
121 dec->offset1[0] |= get_bits1(&gb) << 2;
122 dec->pulsepos[2] = get_bits_long(&gb, 27);
123 dec->pulseoff[2] = get_bits(&gb, 4);
124
125 dec->offset1[0] |= get_bits1(&gb) << 3;
126 dec->pulsepos[3] = get_bits_long(&gb, 27);
127 dec->pulseoff[3] = get_bits(&gb, 4);
128 }
129
truespeech_correlate_filter(TSContext * dec)130 static void truespeech_correlate_filter(TSContext *dec)
131 {
132 int16_t tmp[8];
133 int i, j;
134
135 for(i = 0; i < 8; i++){
136 if(i > 0){
137 memcpy(tmp, dec->cvector, i * sizeof(*tmp));
138 for(j = 0; j < i; j++)
139 dec->cvector[j] += (tmp[i - j - 1] * dec->vector[i] + 0x4000) >> 15;
140 }
141 dec->cvector[i] = (8 - dec->vector[i]) >> 3;
142 }
143 for(i = 0; i < 8; i++)
144 dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15;
145
146 dec->filtval = dec->vector[0];
147 }
148
truespeech_filters_merge(TSContext * dec)149 static void truespeech_filters_merge(TSContext *dec)
150 {
151 int i;
152
153 if(!dec->flag){
154 for(i = 0; i < 8; i++){
155 dec->filters[i + 0] = dec->prevfilt[i];
156 dec->filters[i + 8] = dec->prevfilt[i];
157 }
158 }else{
159 for(i = 0; i < 8; i++){
160 dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
161 dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
162 }
163 }
164 for(i = 0; i < 8; i++){
165 dec->filters[i + 16] = dec->cvector[i];
166 dec->filters[i + 24] = dec->cvector[i];
167 }
168 }
169
truespeech_apply_twopoint_filter(TSContext * dec,int quart)170 static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
171 {
172 int16_t tmp[146 + 60], *ptr0, *ptr1;
173 const int16_t *filter;
174 int i, t, off;
175
176 t = dec->offset2[quart];
177 if(t == 127){
178 memset(dec->newvec, 0, 60 * sizeof(*dec->newvec));
179 return;
180 }
181 for(i = 0; i < 146; i++)
182 tmp[i] = dec->filtbuf[i];
183 off = (t / 25) + dec->offset1[quart >> 1] + 18;
184 off = av_clip(off, 0, 145);
185 ptr0 = tmp + 145 - off;
186 ptr1 = tmp + 146;
187 filter = ts_order2_coeffs + (t % 25) * 2;
188 for(i = 0; i < 60; i++){
189 t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
190 ptr0++;
191 dec->newvec[i] = t;
192 ptr1[i] = t;
193 }
194 }
195
truespeech_place_pulses(TSContext * dec,int16_t * out,int quart)196 static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
197 {
198 int16_t tmp[7];
199 int i, j, t;
200 const int16_t *ptr1;
201 int16_t *ptr2;
202 int coef;
203
204 memset(out, 0, 60 * sizeof(*out));
205 for(i = 0; i < 7; i++) {
206 t = dec->pulseval[quart] & 3;
207 dec->pulseval[quart] >>= 2;
208 tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t];
209 }
210
211 coef = dec->pulsepos[quart] >> 15;
212 ptr1 = ts_pulse_values + 30;
213 ptr2 = tmp;
214 for(i = 0, j = 3; (i < 30) && (j > 0); i++){
215 t = *ptr1++;
216 if(coef >= t)
217 coef -= t;
218 else{
219 out[i] = *ptr2++;
220 ptr1 += 30;
221 j--;
222 }
223 }
224 coef = dec->pulsepos[quart] & 0x7FFF;
225 ptr1 = ts_pulse_values;
226 for(i = 30, j = 4; (i < 60) && (j > 0); i++){
227 t = *ptr1++;
228 if(coef >= t)
229 coef -= t;
230 else{
231 out[i] = *ptr2++;
232 ptr1 += 30;
233 j--;
234 }
235 }
236
237 }
238
truespeech_update_filters(TSContext * dec,int16_t * out,int quart)239 static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
240 {
241 int i;
242
243 memmove(dec->filtbuf, &dec->filtbuf[60], 86 * sizeof(*dec->filtbuf));
244 for(i = 0; i < 60; i++){
245 dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
246 out[i] += dec->newvec[i];
247 }
248 }
249
truespeech_synth(TSContext * dec,int16_t * out,int quart)250 static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
251 {
252 int i,k;
253 int t[8];
254 int16_t *ptr0, *ptr1;
255
256 ptr0 = dec->tmp1;
257 ptr1 = dec->filters + quart * 8;
258 for(i = 0; i < 60; i++){
259 int sum = 0;
260 for(k = 0; k < 8; k++)
261 sum += ptr0[k] * (unsigned)ptr1[k];
262 sum = out[i] + ((int)(sum + 0x800U) >> 12);
263 out[i] = av_clip(sum, -0x7FFE, 0x7FFE);
264 for(k = 7; k > 0; k--)
265 ptr0[k] = ptr0[k - 1];
266 ptr0[0] = out[i];
267 }
268
269 for(i = 0; i < 8; i++)
270 t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15;
271
272 ptr0 = dec->tmp2;
273 for(i = 0; i < 60; i++){
274 int sum = 0;
275 for(k = 0; k < 8; k++)
276 sum += ptr0[k] * t[k];
277 for(k = 7; k > 0; k--)
278 ptr0[k] = ptr0[k - 1];
279 ptr0[0] = out[i];
280 out[i] += (- sum) >> 12;
281 }
282
283 for(i = 0; i < 8; i++)
284 t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15;
285
286 ptr0 = dec->tmp3;
287 for(i = 0; i < 60; i++){
288 int sum = out[i] * (1 << 12);
289 for(k = 0; k < 8; k++)
290 sum += ptr0[k] * t[k];
291 for(k = 7; k > 0; k--)
292 ptr0[k] = ptr0[k - 1];
293 ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
294
295 sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
296 sum = sum - (sum >> 3);
297 out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
298 }
299 }
300
truespeech_save_prevvec(TSContext * c)301 static void truespeech_save_prevvec(TSContext *c)
302 {
303 int i;
304
305 for(i = 0; i < 8; i++)
306 c->prevfilt[i] = c->cvector[i];
307 }
308
truespeech_decode_frame(AVCodecContext * avctx,AVFrame * frame,int * got_frame_ptr,AVPacket * avpkt)309 static int truespeech_decode_frame(AVCodecContext *avctx, AVFrame *frame,
310 int *got_frame_ptr, AVPacket *avpkt)
311 {
312 const uint8_t *buf = avpkt->data;
313 int buf_size = avpkt->size;
314 TSContext *c = avctx->priv_data;
315
316 int i, j;
317 int16_t *samples;
318 int iterations, ret;
319
320 iterations = buf_size / 32;
321
322 if (!iterations) {
323 av_log(avctx, AV_LOG_ERROR,
324 "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
325 return -1;
326 }
327
328 /* get output buffer */
329 frame->nb_samples = iterations * 240;
330 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
331 return ret;
332 samples = (int16_t *)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
340 truespeech_correlate_filter(c);
341 truespeech_filters_merge(c);
342
343 for(i = 0; i < 4; i++) {
344 truespeech_apply_twopoint_filter(c, 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
351 truespeech_save_prevvec(c);
352 }
353
354 *got_frame_ptr = 1;
355
356 return buf_size;
357 }
358
359 const FFCodec ff_truespeech_decoder = {
360 .p.name = "truespeech",
361 .p.long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),
362 .p.type = AVMEDIA_TYPE_AUDIO,
363 .p.id = AV_CODEC_ID_TRUESPEECH,
364 .priv_data_size = sizeof(TSContext),
365 .init = truespeech_decode_init,
366 FF_CODEC_DECODE_CB(truespeech_decode_frame),
367 .p.capabilities = AV_CODEC_CAP_DR1,
368 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
369 };
370