2 * DSP Group TrueSpeech compatible decoder
3 * Copyright (c) 2005 Konstantin Shishkov
5 * This file is part of Libav.
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.
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.
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
22 #include "libavutil/intreadwrite.h"
25 #include "truespeech_data.h"
32 * TrueSpeech decoder context
36 int16_t vector[8]; ///< input vector: 5/5/4/4/4/3/3/3
37 int offset1[2]; ///< 8-bit value, used in one copying offset
38 int offset2[4]; ///< 7-bit value, encodes offsets for copying and for two-point filter
39 int pulseoff[4]; ///< 4-bit offset of pulse values block
40 int pulsepos[4]; ///< 27-bit variable, encodes 7 pulse positions
41 int pulseval[4]; ///< 7x2-bit pulse values
42 int flag; ///< 1-bit flag, shows how to choose filters
44 int filtbuf[146]; // some big vector used for storing filters
45 int prevfilt[8]; // filter from previous frame
46 int16_t tmp1[8]; // coefficients for adding to out
47 int16_t tmp2[8]; // coefficients for adding to out
48 int16_t tmp3[8]; // coefficients for adding to out
49 int16_t cvector[8]; // correlated input vector
50 int filtval; // gain value for one function
51 int16_t newvec[60]; // tmp vector
52 int16_t filters[32]; // filters for every subframe
55 static av_cold int truespeech_decode_init(AVCodecContext * avctx)
57 // TSContext *c = avctx->priv_data;
59 if (avctx->channels != 1) {
60 av_log_ask_for_sample(avctx, "Unsupported channel count: %d\n", avctx->channels);
61 return AVERROR(EINVAL);
64 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
68 static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
78 dec->vector[0] = ts_codebook[0][(t >> 1) & 0x1F];
79 dec->vector[1] = ts_codebook[1][(t >> 6) & 0x1F];
80 dec->vector[2] = ts_codebook[2][(t >> 11) & 0xF];
81 dec->vector[3] = ts_codebook[3][(t >> 15) & 0xF];
82 dec->vector[4] = ts_codebook[4][(t >> 19) & 0xF];
83 dec->vector[5] = ts_codebook[5][(t >> 23) & 0x7];
84 dec->vector[6] = ts_codebook[6][(t >> 26) & 0x7];
85 dec->vector[7] = ts_codebook[7][(t >> 29) & 0x7];
91 dec->offset2[0] = (t >> 0) & 0x7F;
92 dec->offset2[1] = (t >> 7) & 0x7F;
93 dec->offset2[2] = (t >> 14) & 0x7F;
94 dec->offset2[3] = (t >> 21) & 0x7F;
96 dec->offset1[0] = ((t >> 28) & 0xF) << 4;
102 dec->pulseval[0] = (t >> 0) & 0x3FFF;
103 dec->pulseval[1] = (t >> 14) & 0x3FFF;
105 dec->offset1[1] = (t >> 28) & 0x0F;
111 dec->pulseval[2] = (t >> 0) & 0x3FFF;
112 dec->pulseval[3] = (t >> 14) & 0x3FFF;
114 dec->offset1[1] |= ((t >> 28) & 0x0F) << 4;
120 dec->pulsepos[0] = (t >> 4) & 0x7FFFFFF;
122 dec->pulseoff[0] = (t >> 0) & 0xF;
124 dec->offset1[0] |= (t >> 31) & 1;
130 dec->pulsepos[1] = (t >> 4) & 0x7FFFFFF;
132 dec->pulseoff[1] = (t >> 0) & 0xF;
134 dec->offset1[0] |= ((t >> 31) & 1) << 1;
140 dec->pulsepos[2] = (t >> 4) & 0x7FFFFFF;
142 dec->pulseoff[2] = (t >> 0) & 0xF;
144 dec->offset1[0] |= ((t >> 31) & 1) << 2;
150 dec->pulsepos[3] = (t >> 4) & 0x7FFFFFF;
152 dec->pulseoff[3] = (t >> 0) & 0xF;
154 dec->offset1[0] |= ((t >> 31) & 1) << 3;
158 static void truespeech_correlate_filter(TSContext *dec)
163 for(i = 0; i < 8; i++){
165 memcpy(tmp, dec->cvector, i * 2);
166 for(j = 0; j < i; j++)
167 dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) +
168 (dec->cvector[j] << 15) + 0x4000) >> 15;
170 dec->cvector[i] = (8 - dec->vector[i]) >> 3;
172 for(i = 0; i < 8; i++)
173 dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15;
175 dec->filtval = dec->vector[0];
178 static void truespeech_filters_merge(TSContext *dec)
183 for(i = 0; i < 8; i++){
184 dec->filters[i + 0] = dec->prevfilt[i];
185 dec->filters[i + 8] = dec->prevfilt[i];
188 for(i = 0; i < 8; i++){
189 dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
190 dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
193 for(i = 0; i < 8; i++){
194 dec->filters[i + 16] = dec->cvector[i];
195 dec->filters[i + 24] = dec->cvector[i];
199 static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
201 int16_t tmp[146 + 60], *ptr0, *ptr1;
202 const int16_t *filter;
205 t = dec->offset2[quart];
207 memset(dec->newvec, 0, 60 * 2);
210 for(i = 0; i < 146; i++)
211 tmp[i] = dec->filtbuf[i];
212 off = (t / 25) + dec->offset1[quart >> 1] + 18;
213 ptr0 = tmp + 145 - off;
215 filter = (const int16_t*)ts_order2_coeffs + (t % 25) * 2;
216 for(i = 0; i < 60; i++){
217 t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
224 static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
232 memset(out, 0, 60 * 2);
233 for(i = 0; i < 7; i++) {
234 t = dec->pulseval[quart] & 3;
235 dec->pulseval[quart] >>= 2;
236 tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t];
239 coef = dec->pulsepos[quart] >> 15;
240 ptr1 = (const int16_t*)ts_pulse_values + 30;
242 for(i = 0, j = 3; (i < 30) && (j > 0); i++){
252 coef = dec->pulsepos[quart] & 0x7FFF;
253 ptr1 = (const int16_t*)ts_pulse_values;
254 for(i = 30, j = 4; (i < 60) && (j > 0); i++){
267 static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
271 for(i = 0; i < 86; i++)
272 dec->filtbuf[i] = dec->filtbuf[i + 60];
273 for(i = 0; i < 60; i++){
274 dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
275 out[i] += dec->newvec[i];
279 static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
283 int16_t *ptr0, *ptr1;
286 ptr1 = dec->filters + quart * 8;
287 for(i = 0; i < 60; i++){
289 for(k = 0; k < 8; k++)
290 sum += ptr0[k] * ptr1[k];
291 sum = (sum + (out[i] << 12) + 0x800) >> 12;
292 out[i] = av_clip(sum, -0x7FFE, 0x7FFE);
293 for(k = 7; k > 0; k--)
294 ptr0[k] = ptr0[k - 1];
298 for(i = 0; i < 8; i++)
299 t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15;
302 for(i = 0; i < 60; i++){
304 for(k = 0; k < 8; k++)
305 sum += ptr0[k] * t[k];
306 for(k = 7; k > 0; k--)
307 ptr0[k] = ptr0[k - 1];
309 out[i] = ((out[i] << 12) - sum) >> 12;
312 for(i = 0; i < 8; i++)
313 t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15;
316 for(i = 0; i < 60; i++){
317 int sum = out[i] << 12;
318 for(k = 0; k < 8; k++)
319 sum += ptr0[k] * t[k];
320 for(k = 7; k > 0; k--)
321 ptr0[k] = ptr0[k - 1];
322 ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
324 sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
325 sum = sum - (sum >> 3);
326 out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
330 static void truespeech_save_prevvec(TSContext *c)
334 for(i = 0; i < 8; i++)
335 c->prevfilt[i] = c->cvector[i];
338 static int truespeech_decode_frame(AVCodecContext *avctx,
339 void *data, int *data_size,
342 const uint8_t *buf = avpkt->data;
343 int buf_size = avpkt->size;
344 TSContext *c = avctx->priv_data;
347 short *samples = data;
349 int16_t out_buf[240];
350 int iterations, out_size;
352 iterations = buf_size / 32;
355 av_log(avctx, AV_LOG_ERROR,
356 "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
360 out_size = iterations * 240 * av_get_bytes_per_sample(avctx->sample_fmt);
361 if (*data_size < out_size) {
362 av_log(avctx, AV_LOG_ERROR, "Output buffer is too small\n");
363 return AVERROR(EINVAL);
366 for(j = 0; j < iterations; j++) {
367 truespeech_read_frame(c, buf + consumed);
370 truespeech_correlate_filter(c);
371 truespeech_filters_merge(c);
373 memset(out_buf, 0, 240 * 2);
374 for(i = 0; i < 4; i++) {
375 truespeech_apply_twopoint_filter(c, i);
376 truespeech_place_pulses(c, out_buf + i * 60, i);
377 truespeech_update_filters(c, out_buf + i * 60, i);
378 truespeech_synth(c, out_buf + i * 60, i);
381 truespeech_save_prevvec(c);
383 /* finally output decoded frame */
384 for(i = 0; i < 240; i++)
385 *samples++ = out_buf[i];
389 *data_size = out_size;
394 AVCodec ff_truespeech_decoder = {
395 .name = "truespeech",
396 .type = AVMEDIA_TYPE_AUDIO,
397 .id = CODEC_ID_TRUESPEECH,
398 .priv_data_size = sizeof(TSContext),
399 .init = truespeech_decode_init,
400 .decode = truespeech_decode_frame,
401 .long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),