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 "get_bits.h" |
29 |
|
|
#include "internal.h" |
30 |
|
|
|
31 |
|
|
#include "truespeech_data.h" |
32 |
|
|
/** |
33 |
|
|
* @file |
34 |
|
|
* TrueSpeech decoder. |
35 |
|
|
*/ |
36 |
|
|
|
37 |
|
|
/** |
38 |
|
|
* TrueSpeech decoder context |
39 |
|
|
*/ |
40 |
|
|
typedef struct TSContext { |
41 |
|
|
BswapDSPContext bdsp; |
42 |
|
|
/* input data */ |
43 |
|
|
DECLARE_ALIGNED(16, uint8_t, buffer)[32]; |
44 |
|
|
int16_t vector[8]; ///< input vector: 5/5/4/4/4/3/3/3 |
45 |
|
|
int offset1[2]; ///< 8-bit value, used in one copying offset |
46 |
|
|
int offset2[4]; ///< 7-bit value, encodes offsets for copying and for two-point filter |
47 |
|
|
int pulseoff[4]; ///< 4-bit offset of pulse values block |
48 |
|
|
int pulsepos[4]; ///< 27-bit variable, encodes 7 pulse positions |
49 |
|
|
int pulseval[4]; ///< 7x2-bit pulse values |
50 |
|
|
int flag; ///< 1-bit flag, shows how to choose filters |
51 |
|
|
/* temporary data */ |
52 |
|
|
int filtbuf[146]; // some big vector used for storing filters |
53 |
|
|
int prevfilt[8]; // filter from previous frame |
54 |
|
|
int16_t tmp1[8]; // coefficients for adding to out |
55 |
|
|
int16_t tmp2[8]; // coefficients for adding to out |
56 |
|
|
int16_t tmp3[8]; // coefficients for adding to out |
57 |
|
|
int16_t cvector[8]; // correlated input vector |
58 |
|
|
int filtval; // gain value for one function |
59 |
|
|
int16_t newvec[60]; // tmp vector |
60 |
|
|
int16_t filters[32]; // filters for every subframe |
61 |
|
|
} TSContext; |
62 |
|
|
|
63 |
|
2 |
static av_cold int truespeech_decode_init(AVCodecContext * avctx) |
64 |
|
|
{ |
65 |
|
2 |
TSContext *c = avctx->priv_data; |
66 |
|
|
|
67 |
✗✓ |
2 |
if (avctx->channels != 1) { |
68 |
|
|
avpriv_request_sample(avctx, "Channel count %d", avctx->channels); |
69 |
|
|
return AVERROR_PATCHWELCOME; |
70 |
|
|
} |
71 |
|
|
|
72 |
|
2 |
avctx->channel_layout = AV_CH_LAYOUT_MONO; |
73 |
|
2 |
avctx->sample_fmt = AV_SAMPLE_FMT_S16; |
74 |
|
|
|
75 |
|
2 |
ff_bswapdsp_init(&c->bdsp); |
76 |
|
|
|
77 |
|
2 |
return 0; |
78 |
|
|
} |
79 |
|
|
|
80 |
|
1442 |
static void truespeech_read_frame(TSContext *dec, const uint8_t *input) |
81 |
|
|
{ |
82 |
|
|
GetBitContext gb; |
83 |
|
|
|
84 |
|
1442 |
dec->bdsp.bswap_buf((uint32_t *) dec->buffer, (const uint32_t *) input, 8); |
85 |
|
1442 |
init_get_bits(&gb, dec->buffer, 32 * 8); |
86 |
|
|
|
87 |
|
1442 |
dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)]; |
88 |
|
1442 |
dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)]; |
89 |
|
1442 |
dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)]; |
90 |
|
1442 |
dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)]; |
91 |
|
1442 |
dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)]; |
92 |
|
1442 |
dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)]; |
93 |
|
1442 |
dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)]; |
94 |
|
1442 |
dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)]; |
95 |
|
1442 |
dec->flag = get_bits1(&gb); |
96 |
|
|
|
97 |
|
1442 |
dec->offset1[0] = get_bits(&gb, 4) << 4; |
98 |
|
1442 |
dec->offset2[3] = get_bits(&gb, 7); |
99 |
|
1442 |
dec->offset2[2] = get_bits(&gb, 7); |
100 |
|
1442 |
dec->offset2[1] = get_bits(&gb, 7); |
101 |
|
1442 |
dec->offset2[0] = get_bits(&gb, 7); |
102 |
|
|
|
103 |
|
1442 |
dec->offset1[1] = get_bits(&gb, 4); |
104 |
|
1442 |
dec->pulseval[1] = get_bits(&gb, 14); |
105 |
|
1442 |
dec->pulseval[0] = get_bits(&gb, 14); |
106 |
|
|
|
107 |
|
1442 |
dec->offset1[1] |= get_bits(&gb, 4) << 4; |
108 |
|
1442 |
dec->pulseval[3] = get_bits(&gb, 14); |
109 |
|
1442 |
dec->pulseval[2] = get_bits(&gb, 14); |
110 |
|
|
|
111 |
|
1442 |
dec->offset1[0] |= get_bits1(&gb); |
112 |
|
1442 |
dec->pulsepos[0] = get_bits_long(&gb, 27); |
113 |
|
1442 |
dec->pulseoff[0] = get_bits(&gb, 4); |
114 |
|
|
|
115 |
|
1442 |
dec->offset1[0] |= get_bits1(&gb) << 1; |
116 |
|
1442 |
dec->pulsepos[1] = get_bits_long(&gb, 27); |
117 |
|
1442 |
dec->pulseoff[1] = get_bits(&gb, 4); |
118 |
|
|
|
119 |
|
1442 |
dec->offset1[0] |= get_bits1(&gb) << 2; |
120 |
|
1442 |
dec->pulsepos[2] = get_bits_long(&gb, 27); |
121 |
|
1442 |
dec->pulseoff[2] = get_bits(&gb, 4); |
122 |
|
|
|
123 |
|
1442 |
dec->offset1[0] |= get_bits1(&gb) << 3; |
124 |
|
1442 |
dec->pulsepos[3] = get_bits_long(&gb, 27); |
125 |
|
1442 |
dec->pulseoff[3] = get_bits(&gb, 4); |
126 |
|
1442 |
} |
127 |
|
|
|
128 |
|
1442 |
static void truespeech_correlate_filter(TSContext *dec) |
129 |
|
|
{ |
130 |
|
|
int16_t tmp[8]; |
131 |
|
|
int i, j; |
132 |
|
|
|
133 |
✓✓ |
12978 |
for(i = 0; i < 8; i++){ |
134 |
✓✓ |
11536 |
if(i > 0){ |
135 |
|
10094 |
memcpy(tmp, dec->cvector, i * sizeof(*tmp)); |
136 |
✓✓ |
50470 |
for(j = 0; j < i; j++) |
137 |
|
40376 |
dec->cvector[j] += (tmp[i - j - 1] * dec->vector[i] + 0x4000) >> 15; |
138 |
|
|
} |
139 |
|
11536 |
dec->cvector[i] = (8 - dec->vector[i]) >> 3; |
140 |
|
|
} |
141 |
✓✓ |
12978 |
for(i = 0; i < 8; i++) |
142 |
|
11536 |
dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15; |
143 |
|
|
|
144 |
|
1442 |
dec->filtval = dec->vector[0]; |
145 |
|
1442 |
} |
146 |
|
|
|
147 |
|
1442 |
static void truespeech_filters_merge(TSContext *dec) |
148 |
|
|
{ |
149 |
|
|
int i; |
150 |
|
|
|
151 |
✓✓ |
1442 |
if(!dec->flag){ |
152 |
✓✓ |
99 |
for(i = 0; i < 8; i++){ |
153 |
|
88 |
dec->filters[i + 0] = dec->prevfilt[i]; |
154 |
|
88 |
dec->filters[i + 8] = dec->prevfilt[i]; |
155 |
|
|
} |
156 |
|
|
}else{ |
157 |
✓✓ |
12879 |
for(i = 0; i < 8; i++){ |
158 |
|
11448 |
dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15; |
159 |
|
11448 |
dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15; |
160 |
|
|
} |
161 |
|
|
} |
162 |
✓✓ |
12978 |
for(i = 0; i < 8; i++){ |
163 |
|
11536 |
dec->filters[i + 16] = dec->cvector[i]; |
164 |
|
11536 |
dec->filters[i + 24] = dec->cvector[i]; |
165 |
|
|
} |
166 |
|
1442 |
} |
167 |
|
|
|
168 |
|
5768 |
static void truespeech_apply_twopoint_filter(TSContext *dec, int quart) |
169 |
|
|
{ |
170 |
|
|
int16_t tmp[146 + 60], *ptr0, *ptr1; |
171 |
|
|
const int16_t *filter; |
172 |
|
|
int i, t, off; |
173 |
|
|
|
174 |
|
5768 |
t = dec->offset2[quart]; |
175 |
✓✓ |
5768 |
if(t == 127){ |
176 |
|
358 |
memset(dec->newvec, 0, 60 * sizeof(*dec->newvec)); |
177 |
|
358 |
return; |
178 |
|
|
} |
179 |
✓✓ |
795270 |
for(i = 0; i < 146; i++) |
180 |
|
789860 |
tmp[i] = dec->filtbuf[i]; |
181 |
|
5410 |
off = (t / 25) + dec->offset1[quart >> 1] + 18; |
182 |
|
5410 |
off = av_clip(off, 0, 145); |
183 |
|
5410 |
ptr0 = tmp + 145 - off; |
184 |
|
5410 |
ptr1 = tmp + 146; |
185 |
|
5410 |
filter = ts_order2_coeffs + (t % 25) * 2; |
186 |
✓✓ |
330010 |
for(i = 0; i < 60; i++){ |
187 |
|
324600 |
t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14; |
188 |
|
324600 |
ptr0++; |
189 |
|
324600 |
dec->newvec[i] = t; |
190 |
|
324600 |
ptr1[i] = t; |
191 |
|
|
} |
192 |
|
|
} |
193 |
|
|
|
194 |
|
5768 |
static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart) |
195 |
|
|
{ |
196 |
|
|
int16_t tmp[7]; |
197 |
|
|
int i, j, t; |
198 |
|
|
const int16_t *ptr1; |
199 |
|
|
int16_t *ptr2; |
200 |
|
|
int coef; |
201 |
|
|
|
202 |
|
5768 |
memset(out, 0, 60 * sizeof(*out)); |
203 |
✓✓ |
46144 |
for(i = 0; i < 7; i++) { |
204 |
|
40376 |
t = dec->pulseval[quart] & 3; |
205 |
|
40376 |
dec->pulseval[quart] >>= 2; |
206 |
|
40376 |
tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t]; |
207 |
|
|
} |
208 |
|
|
|
209 |
|
5768 |
coef = dec->pulsepos[quart] >> 15; |
210 |
|
5768 |
ptr1 = ts_pulse_values + 30; |
211 |
|
5768 |
ptr2 = tmp; |
212 |
✓✓✓✓
|
136138 |
for(i = 0, j = 3; (i < 30) && (j > 0); i++){ |
213 |
|
130370 |
t = *ptr1++; |
214 |
✓✓ |
130370 |
if(coef >= t) |
215 |
|
113066 |
coef -= t; |
216 |
|
|
else{ |
217 |
|
17304 |
out[i] = *ptr2++; |
218 |
|
17304 |
ptr1 += 30; |
219 |
|
17304 |
j--; |
220 |
|
|
} |
221 |
|
|
} |
222 |
|
5768 |
coef = dec->pulsepos[quart] & 0x7FFF; |
223 |
|
5768 |
ptr1 = ts_pulse_values; |
224 |
✓✓✓✓
|
141397 |
for(i = 30, j = 4; (i < 60) && (j > 0); i++){ |
225 |
|
135629 |
t = *ptr1++; |
226 |
✓✓ |
135629 |
if(coef >= t) |
227 |
|
112557 |
coef -= t; |
228 |
|
|
else{ |
229 |
|
23072 |
out[i] = *ptr2++; |
230 |
|
23072 |
ptr1 += 30; |
231 |
|
23072 |
j--; |
232 |
|
|
} |
233 |
|
|
} |
234 |
|
|
|
235 |
|
5768 |
} |
236 |
|
|
|
237 |
|
5768 |
static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart) |
238 |
|
|
{ |
239 |
|
|
int i; |
240 |
|
|
|
241 |
|
5768 |
memmove(dec->filtbuf, &dec->filtbuf[60], 86 * sizeof(*dec->filtbuf)); |
242 |
✓✓ |
351848 |
for(i = 0; i < 60; i++){ |
243 |
|
346080 |
dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3); |
244 |
|
346080 |
out[i] += dec->newvec[i]; |
245 |
|
|
} |
246 |
|
5768 |
} |
247 |
|
|
|
248 |
|
5768 |
static void truespeech_synth(TSContext *dec, int16_t *out, int quart) |
249 |
|
|
{ |
250 |
|
|
int i,k; |
251 |
|
|
int t[8]; |
252 |
|
|
int16_t *ptr0, *ptr1; |
253 |
|
|
|
254 |
|
5768 |
ptr0 = dec->tmp1; |
255 |
|
5768 |
ptr1 = dec->filters + quart * 8; |
256 |
✓✓ |
351848 |
for(i = 0; i < 60; i++){ |
257 |
|
346080 |
int sum = 0; |
258 |
✓✓ |
3114720 |
for(k = 0; k < 8; k++) |
259 |
|
2768640 |
sum += ptr0[k] * (unsigned)ptr1[k]; |
260 |
|
346080 |
sum = out[i] + ((int)(sum + 0x800U) >> 12); |
261 |
|
346080 |
out[i] = av_clip(sum, -0x7FFE, 0x7FFE); |
262 |
✓✓ |
2768640 |
for(k = 7; k > 0; k--) |
263 |
|
2422560 |
ptr0[k] = ptr0[k - 1]; |
264 |
|
346080 |
ptr0[0] = out[i]; |
265 |
|
|
} |
266 |
|
|
|
267 |
✓✓ |
51912 |
for(i = 0; i < 8; i++) |
268 |
|
46144 |
t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15; |
269 |
|
|
|
270 |
|
5768 |
ptr0 = dec->tmp2; |
271 |
✓✓ |
351848 |
for(i = 0; i < 60; i++){ |
272 |
|
346080 |
int sum = 0; |
273 |
✓✓ |
3114720 |
for(k = 0; k < 8; k++) |
274 |
|
2768640 |
sum += ptr0[k] * t[k]; |
275 |
✓✓ |
2768640 |
for(k = 7; k > 0; k--) |
276 |
|
2422560 |
ptr0[k] = ptr0[k - 1]; |
277 |
|
346080 |
ptr0[0] = out[i]; |
278 |
|
346080 |
out[i] += (- sum) >> 12; |
279 |
|
|
} |
280 |
|
|
|
281 |
✓✓ |
51912 |
for(i = 0; i < 8; i++) |
282 |
|
46144 |
t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15; |
283 |
|
|
|
284 |
|
5768 |
ptr0 = dec->tmp3; |
285 |
✓✓ |
351848 |
for(i = 0; i < 60; i++){ |
286 |
|
346080 |
int sum = out[i] * (1 << 12); |
287 |
✓✓ |
3114720 |
for(k = 0; k < 8; k++) |
288 |
|
2768640 |
sum += ptr0[k] * t[k]; |
289 |
✓✓ |
2768640 |
for(k = 7; k > 0; k--) |
290 |
|
2422560 |
ptr0[k] = ptr0[k - 1]; |
291 |
|
346080 |
ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE); |
292 |
|
|
|
293 |
|
346080 |
sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum; |
294 |
|
346080 |
sum = sum - (sum >> 3); |
295 |
|
346080 |
out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE); |
296 |
|
|
} |
297 |
|
5768 |
} |
298 |
|
|
|
299 |
|
1442 |
static void truespeech_save_prevvec(TSContext *c) |
300 |
|
|
{ |
301 |
|
|
int i; |
302 |
|
|
|
303 |
✓✓ |
12978 |
for(i = 0; i < 8; i++) |
304 |
|
11536 |
c->prevfilt[i] = c->cvector[i]; |
305 |
|
1442 |
} |
306 |
|
|
|
307 |
|
12 |
static int truespeech_decode_frame(AVCodecContext *avctx, void *data, |
308 |
|
|
int *got_frame_ptr, AVPacket *avpkt) |
309 |
|
|
{ |
310 |
|
12 |
AVFrame *frame = data; |
311 |
|
12 |
const uint8_t *buf = avpkt->data; |
312 |
|
12 |
int buf_size = avpkt->size; |
313 |
|
12 |
TSContext *c = avctx->priv_data; |
314 |
|
|
|
315 |
|
|
int i, j; |
316 |
|
|
int16_t *samples; |
317 |
|
|
int iterations, ret; |
318 |
|
|
|
319 |
|
12 |
iterations = buf_size / 32; |
320 |
|
|
|
321 |
✗✓ |
12 |
if (!iterations) { |
322 |
|
|
av_log(avctx, AV_LOG_ERROR, |
323 |
|
|
"Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size); |
324 |
|
|
return -1; |
325 |
|
|
} |
326 |
|
|
|
327 |
|
|
/* get output buffer */ |
328 |
|
12 |
frame->nb_samples = iterations * 240; |
329 |
✗✓ |
12 |
if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
330 |
|
|
return ret; |
331 |
|
12 |
samples = (int16_t *)frame->data[0]; |
332 |
|
|
|
333 |
|
12 |
memset(samples, 0, iterations * 240 * sizeof(*samples)); |
334 |
|
|
|
335 |
✓✓ |
1454 |
for(j = 0; j < iterations; j++) { |
336 |
|
1442 |
truespeech_read_frame(c, buf); |
337 |
|
1442 |
buf += 32; |
338 |
|
|
|
339 |
|
1442 |
truespeech_correlate_filter(c); |
340 |
|
1442 |
truespeech_filters_merge(c); |
341 |
|
|
|
342 |
✓✓ |
7210 |
for(i = 0; i < 4; i++) { |
343 |
|
5768 |
truespeech_apply_twopoint_filter(c, i); |
344 |
|
5768 |
truespeech_place_pulses (c, samples, i); |
345 |
|
5768 |
truespeech_update_filters(c, samples, i); |
346 |
|
5768 |
truespeech_synth (c, samples, i); |
347 |
|
5768 |
samples += 60; |
348 |
|
|
} |
349 |
|
|
|
350 |
|
1442 |
truespeech_save_prevvec(c); |
351 |
|
|
} |
352 |
|
|
|
353 |
|
12 |
*got_frame_ptr = 1; |
354 |
|
|
|
355 |
|
12 |
return buf_size; |
356 |
|
|
} |
357 |
|
|
|
358 |
|
|
AVCodec ff_truespeech_decoder = { |
359 |
|
|
.name = "truespeech", |
360 |
|
|
.long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"), |
361 |
|
|
.type = AVMEDIA_TYPE_AUDIO, |
362 |
|
|
.id = AV_CODEC_ID_TRUESPEECH, |
363 |
|
|
.priv_data_size = sizeof(TSContext), |
364 |
|
|
.init = truespeech_decode_init, |
365 |
|
|
.decode = truespeech_decode_frame, |
366 |
|
|
.capabilities = AV_CODEC_CAP_DR1, |
367 |
|
|
}; |