blob: d4ddfcbf9c147dbe891e308a2517ea11959e4e60
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 "avcodec.h" |
25 | #include "bswapdsp.h" |
26 | #include "get_bits.h" |
27 | #include "internal.h" |
28 | |
29 | #include "truespeech_data.h" |
30 | /** |
31 | * @file |
32 | * TrueSpeech decoder. |
33 | */ |
34 | |
35 | /** |
36 | * TrueSpeech decoder context |
37 | */ |
38 | typedef struct TSContext { |
39 | BswapDSPContext bdsp; |
40 | /* input data */ |
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 |
49 | /* temporary data */ |
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 |
59 | } TSContext; |
60 | |
61 | static av_cold int truespeech_decode_init(AVCodecContext * avctx) |
62 | { |
63 | TSContext *c = avctx->priv_data; |
64 | |
65 | if (avctx->channels != 1) { |
66 | avpriv_request_sample(avctx, "Channel count %d", avctx->channels); |
67 | return AVERROR_PATCHWELCOME; |
68 | } |
69 | |
70 | avctx->channel_layout = AV_CH_LAYOUT_MONO; |
71 | avctx->sample_fmt = AV_SAMPLE_FMT_S16; |
72 | |
73 | ff_bswapdsp_init(&c->bdsp); |
74 | |
75 | return 0; |
76 | } |
77 | |
78 | static void truespeech_read_frame(TSContext *dec, const uint8_t *input) |
79 | { |
80 | GetBitContext gb; |
81 | |
82 | dec->bdsp.bswap_buf((uint32_t *) dec->buffer, (const uint32_t *) input, 8); |
83 | init_get_bits(&gb, dec->buffer, 32 * 8); |
84 | |
85 | dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)]; |
86 | dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)]; |
87 | dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)]; |
88 | dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)]; |
89 | dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)]; |
90 | dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)]; |
91 | dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)]; |
92 | dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)]; |
93 | dec->flag = get_bits1(&gb); |
94 | |
95 | dec->offset1[0] = get_bits(&gb, 4) << 4; |
96 | dec->offset2[3] = get_bits(&gb, 7); |
97 | dec->offset2[2] = get_bits(&gb, 7); |
98 | dec->offset2[1] = get_bits(&gb, 7); |
99 | dec->offset2[0] = get_bits(&gb, 7); |
100 | |
101 | dec->offset1[1] = get_bits(&gb, 4); |
102 | dec->pulseval[1] = get_bits(&gb, 14); |
103 | dec->pulseval[0] = get_bits(&gb, 14); |
104 | |
105 | dec->offset1[1] |= get_bits(&gb, 4) << 4; |
106 | dec->pulseval[3] = get_bits(&gb, 14); |
107 | dec->pulseval[2] = get_bits(&gb, 14); |
108 | |
109 | dec->offset1[0] |= get_bits1(&gb); |
110 | dec->pulsepos[0] = get_bits_long(&gb, 27); |
111 | dec->pulseoff[0] = get_bits(&gb, 4); |
112 | |
113 | dec->offset1[0] |= get_bits1(&gb) << 1; |
114 | dec->pulsepos[1] = get_bits_long(&gb, 27); |
115 | dec->pulseoff[1] = get_bits(&gb, 4); |
116 | |
117 | dec->offset1[0] |= get_bits1(&gb) << 2; |
118 | dec->pulsepos[2] = get_bits_long(&gb, 27); |
119 | dec->pulseoff[2] = get_bits(&gb, 4); |
120 | |
121 | dec->offset1[0] |= get_bits1(&gb) << 3; |
122 | dec->pulsepos[3] = get_bits_long(&gb, 27); |
123 | dec->pulseoff[3] = get_bits(&gb, 4); |
124 | } |
125 | |
126 | static void truespeech_correlate_filter(TSContext *dec) |
127 | { |
128 | int16_t tmp[8]; |
129 | int i, j; |
130 | |
131 | for(i = 0; i < 8; i++){ |
132 | if(i > 0){ |
133 | memcpy(tmp, dec->cvector, i * sizeof(*tmp)); |
134 | for(j = 0; j < i; j++) |
135 | dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) + |
136 | (dec->cvector[j] << 15) + 0x4000) >> 15; |
137 | } |
138 | dec->cvector[i] = (8 - dec->vector[i]) >> 3; |
139 | } |
140 | for(i = 0; i < 8; i++) |
141 | dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15; |
142 | |
143 | dec->filtval = dec->vector[0]; |
144 | } |
145 | |
146 | static void truespeech_filters_merge(TSContext *dec) |
147 | { |
148 | int i; |
149 | |
150 | if(!dec->flag){ |
151 | for(i = 0; i < 8; i++){ |
152 | dec->filters[i + 0] = dec->prevfilt[i]; |
153 | dec->filters[i + 8] = dec->prevfilt[i]; |
154 | } |
155 | }else{ |
156 | for(i = 0; i < 8; i++){ |
157 | dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15; |
158 | dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15; |
159 | } |
160 | } |
161 | for(i = 0; i < 8; i++){ |
162 | dec->filters[i + 16] = dec->cvector[i]; |
163 | dec->filters[i + 24] = dec->cvector[i]; |
164 | } |
165 | } |
166 | |
167 | static void truespeech_apply_twopoint_filter(TSContext *dec, int quart) |
168 | { |
169 | int16_t tmp[146 + 60], *ptr0, *ptr1; |
170 | const int16_t *filter; |
171 | int i, t, off; |
172 | |
173 | t = dec->offset2[quart]; |
174 | if(t == 127){ |
175 | memset(dec->newvec, 0, 60 * sizeof(*dec->newvec)); |
176 | return; |
177 | } |
178 | for(i = 0; i < 146; i++) |
179 | tmp[i] = dec->filtbuf[i]; |
180 | off = (t / 25) + dec->offset1[quart >> 1] + 18; |
181 | off = av_clip(off, 0, 145); |
182 | ptr0 = tmp + 145 - off; |
183 | ptr1 = tmp + 146; |
184 | filter = ts_order2_coeffs + (t % 25) * 2; |
185 | for(i = 0; i < 60; i++){ |
186 | t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14; |
187 | ptr0++; |
188 | dec->newvec[i] = t; |
189 | ptr1[i] = t; |
190 | } |
191 | } |
192 | |
193 | static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart) |
194 | { |
195 | int16_t tmp[7]; |
196 | int i, j, t; |
197 | const int16_t *ptr1; |
198 | int16_t *ptr2; |
199 | int coef; |
200 | |
201 | memset(out, 0, 60 * sizeof(*out)); |
202 | for(i = 0; i < 7; i++) { |
203 | t = dec->pulseval[quart] & 3; |
204 | dec->pulseval[quart] >>= 2; |
205 | tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t]; |
206 | } |
207 | |
208 | coef = dec->pulsepos[quart] >> 15; |
209 | ptr1 = ts_pulse_values + 30; |
210 | ptr2 = tmp; |
211 | for(i = 0, j = 3; (i < 30) && (j > 0); i++){ |
212 | t = *ptr1++; |
213 | if(coef >= t) |
214 | coef -= t; |
215 | else{ |
216 | out[i] = *ptr2++; |
217 | ptr1 += 30; |
218 | j--; |
219 | } |
220 | } |
221 | coef = dec->pulsepos[quart] & 0x7FFF; |
222 | ptr1 = ts_pulse_values; |
223 | for(i = 30, j = 4; (i < 60) && (j > 0); i++){ |
224 | t = *ptr1++; |
225 | if(coef >= t) |
226 | coef -= t; |
227 | else{ |
228 | out[i] = *ptr2++; |
229 | ptr1 += 30; |
230 | j--; |
231 | } |
232 | } |
233 | |
234 | } |
235 | |
236 | static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart) |
237 | { |
238 | int i; |
239 | |
240 | memmove(dec->filtbuf, &dec->filtbuf[60], 86 * sizeof(*dec->filtbuf)); |
241 | for(i = 0; i < 60; i++){ |
242 | dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3); |
243 | out[i] += dec->newvec[i]; |
244 | } |
245 | } |
246 | |
247 | static void truespeech_synth(TSContext *dec, int16_t *out, int quart) |
248 | { |
249 | int i,k; |
250 | int t[8]; |
251 | int16_t *ptr0, *ptr1; |
252 | |
253 | ptr0 = dec->tmp1; |
254 | ptr1 = dec->filters + quart * 8; |
255 | for(i = 0; i < 60; i++){ |
256 | int sum = 0; |
257 | for(k = 0; k < 8; k++) |
258 | sum += ptr0[k] * ptr1[k]; |
259 | sum = (sum + (out[i] << 12) + 0x800) >> 12; |
260 | out[i] = av_clip(sum, -0x7FFE, 0x7FFE); |
261 | for(k = 7; k > 0; k--) |
262 | ptr0[k] = ptr0[k - 1]; |
263 | ptr0[0] = out[i]; |
264 | } |
265 | |
266 | for(i = 0; i < 8; i++) |
267 | t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15; |
268 | |
269 | ptr0 = dec->tmp2; |
270 | for(i = 0; i < 60; i++){ |
271 | int sum = 0; |
272 | for(k = 0; k < 8; k++) |
273 | sum += ptr0[k] * t[k]; |
274 | for(k = 7; k > 0; k--) |
275 | ptr0[k] = ptr0[k - 1]; |
276 | ptr0[0] = out[i]; |
277 | out[i] = ((out[i] << 12) - sum) >> 12; |
278 | } |
279 | |
280 | for(i = 0; i < 8; i++) |
281 | t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15; |
282 | |
283 | ptr0 = dec->tmp3; |
284 | for(i = 0; i < 60; i++){ |
285 | int sum = out[i] << 12; |
286 | for(k = 0; k < 8; k++) |
287 | sum += ptr0[k] * t[k]; |
288 | for(k = 7; k > 0; k--) |
289 | ptr0[k] = ptr0[k - 1]; |
290 | ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE); |
291 | |
292 | sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum; |
293 | sum = sum - (sum >> 3); |
294 | out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE); |
295 | } |
296 | } |
297 | |
298 | static void truespeech_save_prevvec(TSContext *c) |
299 | { |
300 | int i; |
301 | |
302 | for(i = 0; i < 8; i++) |
303 | c->prevfilt[i] = c->cvector[i]; |
304 | } |
305 | |
306 | static int truespeech_decode_frame(AVCodecContext *avctx, void *data, |
307 | int *got_frame_ptr, AVPacket *avpkt) |
308 | { |
309 | AVFrame *frame = data; |
310 | const uint8_t *buf = avpkt->data; |
311 | int buf_size = avpkt->size; |
312 | TSContext *c = avctx->priv_data; |
313 | |
314 | int i, j; |
315 | int16_t *samples; |
316 | int iterations, ret; |
317 | |
318 | iterations = buf_size / 32; |
319 | |
320 | if (!iterations) { |
321 | av_log(avctx, AV_LOG_ERROR, |
322 | "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size); |
323 | return -1; |
324 | } |
325 | |
326 | /* get output buffer */ |
327 | frame->nb_samples = iterations * 240; |
328 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
329 | return ret; |
330 | samples = (int16_t *)frame->data[0]; |
331 | |
332 | memset(samples, 0, iterations * 240 * sizeof(*samples)); |
333 | |
334 | for(j = 0; j < iterations; j++) { |
335 | truespeech_read_frame(c, buf); |
336 | buf += 32; |
337 | |
338 | truespeech_correlate_filter(c); |
339 | truespeech_filters_merge(c); |
340 | |
341 | for(i = 0; i < 4; i++) { |
342 | truespeech_apply_twopoint_filter(c, i); |
343 | truespeech_place_pulses (c, samples, i); |
344 | truespeech_update_filters(c, samples, i); |
345 | truespeech_synth (c, samples, i); |
346 | samples += 60; |
347 | } |
348 | |
349 | truespeech_save_prevvec(c); |
350 | } |
351 | |
352 | *got_frame_ptr = 1; |
353 | |
354 | return buf_size; |
355 | } |
356 | |
357 | AVCodec ff_truespeech_decoder = { |
358 | .name = "truespeech", |
359 | .long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"), |
360 | .type = AVMEDIA_TYPE_AUDIO, |
361 | .id = AV_CODEC_ID_TRUESPEECH, |
362 | .priv_data_size = sizeof(TSContext), |
363 | .init = truespeech_decode_init, |
364 | .decode = truespeech_decode_frame, |
365 | .capabilities = AV_CODEC_CAP_DR1, |
366 | }; |
367 |