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/audioconvert.h"
23 #include "libavutil/intreadwrite.h"
28 #include "truespeech_data.h"
35 * TrueSpeech decoder context
41 DECLARE_ALIGNED(16, uint8_t, buffer)[32];
42 int16_t vector[8]; ///< input vector: 5/5/4/4/4/3/3/3
43 int offset1[2]; ///< 8-bit value, used in one copying offset
44 int offset2[4]; ///< 7-bit value, encodes offsets for copying and for two-point filter
45 int pulseoff[4]; ///< 4-bit offset of pulse values block
46 int pulsepos[4]; ///< 27-bit variable, encodes 7 pulse positions
47 int pulseval[4]; ///< 7x2-bit pulse values
48 int flag; ///< 1-bit flag, shows how to choose filters
50 int filtbuf[146]; // some big vector used for storing filters
51 int prevfilt[8]; // filter from previous frame
52 int16_t tmp1[8]; // coefficients for adding to out
53 int16_t tmp2[8]; // coefficients for adding to out
54 int16_t tmp3[8]; // coefficients for adding to out
55 int16_t cvector[8]; // correlated input vector
56 int filtval; // gain value for one function
57 int16_t newvec[60]; // tmp vector
58 int16_t filters[32]; // filters for every subframe
61 static av_cold int truespeech_decode_init(AVCodecContext * avctx)
63 TSContext *c = avctx->priv_data;
65 if (avctx->channels != 1) {
66 av_log_ask_for_sample(avctx, "Unsupported channel count: %d\n", avctx->channels);
67 return AVERROR(EINVAL);
70 avctx->channel_layout = AV_CH_LAYOUT_MONO;
71 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
73 ff_dsputil_init(&c->dsp, avctx);
75 avcodec_get_frame_defaults(&c->frame);
76 avctx->coded_frame = &c->frame;
81 static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
85 dec->dsp.bswap_buf((uint32_t *)dec->buffer, (const uint32_t *)input, 8);
86 init_get_bits(&gb, dec->buffer, 32 * 8);
88 dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)];
89 dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)];
90 dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)];
91 dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)];
92 dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)];
93 dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)];
94 dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)];
95 dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)];
96 dec->flag = get_bits1(&gb);
98 dec->offset1[0] = get_bits(&gb, 4) << 4;
99 dec->offset2[3] = get_bits(&gb, 7);
100 dec->offset2[2] = get_bits(&gb, 7);
101 dec->offset2[1] = get_bits(&gb, 7);
102 dec->offset2[0] = get_bits(&gb, 7);
104 dec->offset1[1] = get_bits(&gb, 4);
105 dec->pulseval[1] = get_bits(&gb, 14);
106 dec->pulseval[0] = get_bits(&gb, 14);
108 dec->offset1[1] |= get_bits(&gb, 4) << 4;
109 dec->pulseval[3] = get_bits(&gb, 14);
110 dec->pulseval[2] = get_bits(&gb, 14);
112 dec->offset1[0] |= get_bits1(&gb);
113 dec->pulsepos[0] = get_bits_long(&gb, 27);
114 dec->pulseoff[0] = get_bits(&gb, 4);
116 dec->offset1[0] |= get_bits1(&gb) << 1;
117 dec->pulsepos[1] = get_bits_long(&gb, 27);
118 dec->pulseoff[1] = get_bits(&gb, 4);
120 dec->offset1[0] |= get_bits1(&gb) << 2;
121 dec->pulsepos[2] = get_bits_long(&gb, 27);
122 dec->pulseoff[2] = get_bits(&gb, 4);
124 dec->offset1[0] |= get_bits1(&gb) << 3;
125 dec->pulsepos[3] = get_bits_long(&gb, 27);
126 dec->pulseoff[3] = get_bits(&gb, 4);
129 static void truespeech_correlate_filter(TSContext *dec)
134 for(i = 0; i < 8; i++){
136 memcpy(tmp, dec->cvector, i * sizeof(*tmp));
137 for(j = 0; j < i; j++)
138 dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) +
139 (dec->cvector[j] << 15) + 0x4000) >> 15;
141 dec->cvector[i] = (8 - dec->vector[i]) >> 3;
143 for(i = 0; i < 8; i++)
144 dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15;
146 dec->filtval = dec->vector[0];
149 static void truespeech_filters_merge(TSContext *dec)
154 for(i = 0; i < 8; i++){
155 dec->filters[i + 0] = dec->prevfilt[i];
156 dec->filters[i + 8] = dec->prevfilt[i];
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;
164 for(i = 0; i < 8; i++){
165 dec->filters[i + 16] = dec->cvector[i];
166 dec->filters[i + 24] = dec->cvector[i];
170 static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
172 int16_t tmp[146 + 60], *ptr0, *ptr1;
173 const int16_t *filter;
176 t = dec->offset2[quart];
178 memset(dec->newvec, 0, 60 * sizeof(*dec->newvec));
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;
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;
196 static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
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];
211 coef = dec->pulsepos[quart] >> 15;
212 ptr1 = ts_pulse_values + 30;
214 for(i = 0, j = 3; (i < 30) && (j > 0); i++){
224 coef = dec->pulsepos[quart] & 0x7FFF;
225 ptr1 = ts_pulse_values;
226 for(i = 30, j = 4; (i < 60) && (j > 0); i++){
239 static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
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];
250 static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
254 int16_t *ptr0, *ptr1;
257 ptr1 = dec->filters + quart * 8;
258 for(i = 0; i < 60; i++){
260 for(k = 0; k < 8; k++)
261 sum += ptr0[k] * ptr1[k];
262 sum = (sum + (out[i] << 12) + 0x800) >> 12;
263 out[i] = av_clip(sum, -0x7FFE, 0x7FFE);
264 for(k = 7; k > 0; k--)
265 ptr0[k] = ptr0[k - 1];
269 for(i = 0; i < 8; i++)
270 t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15;
273 for(i = 0; i < 60; i++){
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];
280 out[i] = ((out[i] << 12) - sum) >> 12;
283 for(i = 0; i < 8; i++)
284 t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15;
287 for(i = 0; i < 60; i++){
288 int sum = out[i] << 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);
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);
301 static void truespeech_save_prevvec(TSContext *c)
305 for(i = 0; i < 8; i++)
306 c->prevfilt[i] = c->cvector[i];
309 static int truespeech_decode_frame(AVCodecContext *avctx, void *data,
310 int *got_frame_ptr, AVPacket *avpkt)
312 const uint8_t *buf = avpkt->data;
313 int buf_size = avpkt->size;
314 TSContext *c = avctx->priv_data;
320 iterations = buf_size / 32;
323 av_log(avctx, AV_LOG_ERROR,
324 "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
328 /* get output buffer */
329 c->frame.nb_samples = iterations * 240;
330 if ((ret = avctx->get_buffer(avctx, &c->frame)) < 0) {
331 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
334 samples = (int16_t *)c->frame.data[0];
336 memset(samples, 0, iterations * 240 * sizeof(*samples));
338 for(j = 0; j < iterations; j++) {
339 truespeech_read_frame(c, buf);
342 truespeech_correlate_filter(c);
343 truespeech_filters_merge(c);
345 for(i = 0; i < 4; i++) {
346 truespeech_apply_twopoint_filter(c, i);
347 truespeech_place_pulses (c, samples, i);
348 truespeech_update_filters(c, samples, i);
349 truespeech_synth (c, samples, i);
353 truespeech_save_prevvec(c);
357 *(AVFrame *)data = c->frame;
362 AVCodec ff_truespeech_decoder = {
363 .name = "truespeech",
364 .type = AVMEDIA_TYPE_AUDIO,
365 .id = AV_CODEC_ID_TRUESPEECH,
366 .priv_data_size = sizeof(TSContext),
367 .init = truespeech_decode_init,
368 .decode = truespeech_decode_frame,
369 .capabilities = CODEC_CAP_DR1,
370 .long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),