3 * Copyright (C) 2008 Konstantin Shishkov
5 * This file is part of FFmpeg.
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.
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.
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
27 /***********************************
29 * add sane pulse detection
30 * add temporal noise shaping
31 ***********************************/
33 #include "libavutil/float_dsp.h"
34 #include "libavutil/opt.h"
38 #include "mpeg4audio.h"
45 #include "aacenctab.h"
46 #include "aacenc_utils.h"
51 * Make AAC audio config object.
52 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
54 static void put_audio_specific_config(AVCodecContext *avctx)
57 AACEncContext *s = avctx->priv_data;
59 init_put_bits(&pb, avctx->extradata, avctx->extradata_size);
60 put_bits(&pb, 5, 2); //object type - AAC-LC
61 put_bits(&pb, 4, s->samplerate_index); //sample rate index
62 put_bits(&pb, 4, s->channels);
64 put_bits(&pb, 1, 0); //frame length - 1024 samples
65 put_bits(&pb, 1, 0); //does not depend on core coder
66 put_bits(&pb, 1, 0); //is not extension
68 //Explicitly Mark SBR absent
69 put_bits(&pb, 11, 0x2b7); //sync extension
70 put_bits(&pb, 5, AOT_SBR);
75 #define WINDOW_FUNC(type) \
76 static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
77 SingleChannelElement *sce, \
80 WINDOW_FUNC(only_long)
82 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
83 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
84 float *out = sce->ret_buf;
86 fdsp->vector_fmul (out, audio, lwindow, 1024);
87 fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
90 WINDOW_FUNC(long_start)
92 const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
93 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
94 float *out = sce->ret_buf;
96 fdsp->vector_fmul(out, audio, lwindow, 1024);
97 memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
98 fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
99 memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
102 WINDOW_FUNC(long_stop)
104 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
105 const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
106 float *out = sce->ret_buf;
108 memset(out, 0, sizeof(out[0]) * 448);
109 fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
110 memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
111 fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
114 WINDOW_FUNC(eight_short)
116 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
117 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
118 const float *in = audio + 448;
119 float *out = sce->ret_buf;
122 for (w = 0; w < 8; w++) {
123 fdsp->vector_fmul (out, in, w ? pwindow : swindow, 128);
126 fdsp->vector_fmul_reverse(out, in, swindow, 128);
131 static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
132 SingleChannelElement *sce,
133 const float *audio) = {
134 [ONLY_LONG_SEQUENCE] = apply_only_long_window,
135 [LONG_START_SEQUENCE] = apply_long_start_window,
136 [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
137 [LONG_STOP_SEQUENCE] = apply_long_stop_window
140 static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
144 float *output = sce->ret_buf;
146 apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
148 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
149 s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
151 for (i = 0; i < 1024; i += 128)
152 s->mdct128.mdct_calc(&s->mdct128, sce->coeffs + i, output + i*2);
153 memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
154 memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
158 * Encode ics_info element.
159 * @see Table 4.6 (syntax of ics_info)
161 static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
165 put_bits(&s->pb, 1, 0); // ics_reserved bit
166 put_bits(&s->pb, 2, info->window_sequence[0]);
167 put_bits(&s->pb, 1, info->use_kb_window[0]);
168 if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
169 put_bits(&s->pb, 6, info->max_sfb);
170 put_bits(&s->pb, 1, 0); // no prediction
172 put_bits(&s->pb, 4, info->max_sfb);
173 for (w = 1; w < 8; w++)
174 put_bits(&s->pb, 1, !info->group_len[w]);
180 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
182 static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
186 put_bits(pb, 2, cpe->ms_mode);
187 if (cpe->ms_mode == 1)
188 for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
189 for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
190 put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
194 * Produce integer coefficients from scalefactors provided by the model.
196 static void adjust_frame_information(ChannelElement *cpe, int chans)
200 IndividualChannelStream *ics;
202 if (cpe->common_window) {
203 ics = &cpe->ch[0].ics;
204 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
205 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
206 int start = (w+w2) * 128;
207 for (g = 0; g < ics->num_swb; g++) {
208 //apply Intensity stereo coeffs transformation
209 if (cpe->is_mask[w*16 + g]) {
210 int p = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
211 float scale = cpe->ch[0].is_ener[w*16+g];
212 for (i = 0; i < ics->swb_sizes[g]; i++) {
213 cpe->ch[0].coeffs[start+i] = (cpe->ch[0].pcoeffs[start+i] + p*cpe->ch[1].pcoeffs[start+i]) * scale;
214 cpe->ch[1].coeffs[start+i] = 0.0f;
216 } else if (cpe->ms_mask[w*16 + g] &&
217 cpe->ch[0].band_type[w*16 + g] < NOISE_BT &&
218 cpe->ch[1].band_type[w*16 + g] < NOISE_BT) {
219 for (i = 0; i < ics->swb_sizes[g]; i++) {
220 cpe->ch[0].coeffs[start+i] = (cpe->ch[0].pcoeffs[start+i] + cpe->ch[1].pcoeffs[start+i]) * 0.5f;
221 cpe->ch[1].coeffs[start+i] = cpe->ch[0].coeffs[start+i] - cpe->ch[1].pcoeffs[start+i];
224 start += ics->swb_sizes[g];
230 for (ch = 0; ch < chans; ch++) {
231 IndividualChannelStream *ics = &cpe->ch[ch].ics;
233 cpe->ch[ch].pulse.num_pulse = 0;
234 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
235 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
236 for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
238 maxsfb = FFMAX(maxsfb, cmaxsfb);
241 ics->max_sfb = maxsfb;
243 //adjust zero bands for window groups
244 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
245 for (g = 0; g < ics->max_sfb; g++) {
247 for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
248 if (!cpe->ch[ch].zeroes[w2*16 + g]) {
253 cpe->ch[ch].zeroes[w*16 + g] = i;
258 if (chans > 1 && cpe->common_window) {
259 IndividualChannelStream *ics0 = &cpe->ch[0].ics;
260 IndividualChannelStream *ics1 = &cpe->ch[1].ics;
262 ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
263 ics1->max_sfb = ics0->max_sfb;
264 for (w = 0; w < ics0->num_windows*16; w += 16)
265 for (i = 0; i < ics0->max_sfb; i++)
266 if (cpe->ms_mask[w+i])
268 if (msc == 0 || ics0->max_sfb == 0)
271 cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
276 * Encode scalefactor band coding type.
278 static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
282 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
283 s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
287 * Encode scalefactors.
289 static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
290 SingleChannelElement *sce)
292 int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
293 int off_is = 0, noise_flag = 1;
296 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
297 for (i = 0; i < sce->ics.max_sfb; i++) {
298 if (!sce->zeroes[w*16 + i]) {
299 if (sce->band_type[w*16 + i] == NOISE_BT) {
300 diff = sce->sf_idx[w*16 + i] - off_pns;
301 off_pns = sce->sf_idx[w*16 + i];
302 if (noise_flag-- > 0) {
303 put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
306 } else if (sce->band_type[w*16 + i] == INTENSITY_BT ||
307 sce->band_type[w*16 + i] == INTENSITY_BT2) {
308 diff = sce->sf_idx[w*16 + i] - off_is;
309 off_is = sce->sf_idx[w*16 + i];
311 diff = sce->sf_idx[w*16 + i] - off_sf;
312 off_sf = sce->sf_idx[w*16 + i];
314 diff += SCALE_DIFF_ZERO;
315 av_assert0(diff >= 0 && diff <= 120);
316 put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
325 static void encode_pulses(AACEncContext *s, Pulse *pulse)
329 put_bits(&s->pb, 1, !!pulse->num_pulse);
330 if (!pulse->num_pulse)
333 put_bits(&s->pb, 2, pulse->num_pulse - 1);
334 put_bits(&s->pb, 6, pulse->start);
335 for (i = 0; i < pulse->num_pulse; i++) {
336 put_bits(&s->pb, 5, pulse->pos[i]);
337 put_bits(&s->pb, 4, pulse->amp[i]);
342 * Encode spectral coefficients processed by psychoacoustic model.
344 static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
348 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
350 for (i = 0; i < sce->ics.max_sfb; i++) {
351 if (sce->zeroes[w*16 + i]) {
352 start += sce->ics.swb_sizes[i];
355 for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++)
356 s->coder->quantize_and_encode_band(s, &s->pb, sce->coeffs + start + w2*128,
357 sce->ics.swb_sizes[i],
358 sce->sf_idx[w*16 + i],
359 sce->band_type[w*16 + i],
360 s->lambda, sce->ics.window_clipping[w]);
361 start += sce->ics.swb_sizes[i];
367 * Downscale spectral coefficients for near-clipping windows to avoid artifacts
369 static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
373 if (sce->ics.clip_avoidance_factor < 1.0f) {
374 for (w = 0; w < sce->ics.num_windows; w++) {
376 for (i = 0; i < sce->ics.max_sfb; i++) {
377 float *swb_coeffs = sce->coeffs + start + w*128;
378 for (j = 0; j < sce->ics.swb_sizes[i]; j++)
379 swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
380 start += sce->ics.swb_sizes[i];
387 * Encode one channel of audio data.
389 static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
390 SingleChannelElement *sce,
393 put_bits(&s->pb, 8, sce->sf_idx[0]);
395 put_ics_info(s, &sce->ics);
396 encode_band_info(s, sce);
397 encode_scale_factors(avctx, s, sce);
398 encode_pulses(s, &sce->pulse);
399 put_bits(&s->pb, 1, 0); //tns
400 put_bits(&s->pb, 1, 0); //ssr
401 encode_spectral_coeffs(s, sce);
406 * Write some auxiliary information about the created AAC file.
408 static void put_bitstream_info(AACEncContext *s, const char *name)
410 int i, namelen, padbits;
412 namelen = strlen(name) + 2;
413 put_bits(&s->pb, 3, TYPE_FIL);
414 put_bits(&s->pb, 4, FFMIN(namelen, 15));
416 put_bits(&s->pb, 8, namelen - 14);
417 put_bits(&s->pb, 4, 0); //extension type - filler
418 padbits = -put_bits_count(&s->pb) & 7;
419 avpriv_align_put_bits(&s->pb);
420 for (i = 0; i < namelen - 2; i++)
421 put_bits(&s->pb, 8, name[i]);
422 put_bits(&s->pb, 12 - padbits, 0);
426 * Copy input samples.
427 * Channels are reordered from libavcodec's default order to AAC order.
429 static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
432 int end = 2048 + (frame ? frame->nb_samples : 0);
433 const uint8_t *channel_map = aac_chan_maps[s->channels - 1];
435 /* copy and remap input samples */
436 for (ch = 0; ch < s->channels; ch++) {
437 /* copy last 1024 samples of previous frame to the start of the current frame */
438 memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
440 /* copy new samples and zero any remaining samples */
442 memcpy(&s->planar_samples[ch][2048],
443 frame->extended_data[channel_map[ch]],
444 frame->nb_samples * sizeof(s->planar_samples[0][0]));
446 memset(&s->planar_samples[ch][end], 0,
447 (3072 - end) * sizeof(s->planar_samples[0][0]));
451 static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
452 const AVFrame *frame, int *got_packet_ptr)
454 AACEncContext *s = avctx->priv_data;
455 float **samples = s->planar_samples, *samples2, *la, *overlap;
457 int i, ch, w, g, chans, tag, start_ch, ret, ms_mode = 0, is_mode = 0;
458 int chan_el_counter[4];
459 FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
461 if (s->last_frame == 2)
464 /* add current frame to queue */
466 if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
470 copy_input_samples(s, frame);
472 ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
474 if (!avctx->frame_number)
478 for (i = 0; i < s->chan_map[0]; i++) {
479 FFPsyWindowInfo* wi = windows + start_ch;
480 tag = s->chan_map[i+1];
481 chans = tag == TYPE_CPE ? 2 : 1;
483 for (ch = 0; ch < chans; ch++) {
484 IndividualChannelStream *ics = &cpe->ch[ch].ics;
485 int cur_channel = start_ch + ch;
486 float clip_avoidance_factor;
487 overlap = &samples[cur_channel][0];
488 samples2 = overlap + 1024;
489 la = samples2 + (448+64);
492 if (tag == TYPE_LFE) {
493 wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;
494 wi[ch].window_shape = 0;
495 wi[ch].num_windows = 1;
496 wi[ch].grouping[0] = 1;
498 /* Only the lowest 12 coefficients are used in a LFE channel.
499 * The expression below results in only the bottom 8 coefficients
500 * being used for 11.025kHz to 16kHz sample rates.
502 ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
504 wi[ch] = s->psy.model->window(&s->psy, samples2, la, cur_channel,
505 ics->window_sequence[0]);
507 ics->window_sequence[1] = ics->window_sequence[0];
508 ics->window_sequence[0] = wi[ch].window_type[0];
509 ics->use_kb_window[1] = ics->use_kb_window[0];
510 ics->use_kb_window[0] = wi[ch].window_shape;
511 ics->num_windows = wi[ch].num_windows;
512 ics->swb_sizes = s->psy.bands [ics->num_windows == 8];
513 ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
514 clip_avoidance_factor = 0.0f;
515 for (w = 0; w < ics->num_windows; w++)
516 ics->group_len[w] = wi[ch].grouping[w];
517 for (w = 0; w < ics->num_windows; w++) {
518 if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
519 ics->window_clipping[w] = 1;
520 clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
522 ics->window_clipping[w] = 0;
525 if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
526 ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
528 ics->clip_avoidance_factor = 1.0f;
531 apply_window_and_mdct(s, &cpe->ch[ch], overlap);
532 if (isnan(cpe->ch->coeffs[0])) {
533 av_log(avctx, AV_LOG_ERROR, "Input contains NaN\n");
534 return AVERROR(EINVAL);
536 avoid_clipping(s, &cpe->ch[ch]);
540 if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
545 init_put_bits(&s->pb, avpkt->data, avpkt->size);
547 if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
548 put_bitstream_info(s, LIBAVCODEC_IDENT);
550 memset(chan_el_counter, 0, sizeof(chan_el_counter));
551 for (i = 0; i < s->chan_map[0]; i++) {
552 FFPsyWindowInfo* wi = windows + start_ch;
553 const float *coeffs[2];
554 tag = s->chan_map[i+1];
555 chans = tag == TYPE_CPE ? 2 : 1;
557 memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
558 memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
559 put_bits(&s->pb, 3, tag);
560 put_bits(&s->pb, 4, chan_el_counter[tag]++);
561 for (ch = 0; ch < chans; ch++)
562 coeffs[ch] = cpe->ch[ch].coeffs;
563 s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
564 for (ch = 0; ch < chans; ch++) {
565 s->cur_channel = start_ch + ch;
566 s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
568 cpe->common_window = 0;
570 && wi[0].window_type[0] == wi[1].window_type[0]
571 && wi[0].window_shape == wi[1].window_shape) {
573 cpe->common_window = 1;
574 for (w = 0; w < wi[0].num_windows; w++) {
575 if (wi[0].grouping[w] != wi[1].grouping[w]) {
576 cpe->common_window = 0;
581 if (s->options.pns && s->coder->search_for_pns) {
582 for (ch = 0; ch < chans; ch++) {
583 s->cur_channel = start_ch + ch;
584 s->coder->search_for_pns(s, avctx, &cpe->ch[ch]);
587 s->cur_channel = start_ch;
588 if (s->options.stereo_mode && cpe->common_window) {
589 if (s->options.stereo_mode > 0) {
590 IndividualChannelStream *ics = &cpe->ch[0].ics;
591 for (w = 0; w < ics->num_windows; w += ics->group_len[w])
592 for (g = 0; g < ics->num_swb; g++)
593 cpe->ms_mask[w*16+g] = 1;
594 } else if (s->coder->search_for_ms) {
595 s->coder->search_for_ms(s, cpe);
598 if (chans > 1 && s->options.intensity_stereo && s->coder->search_for_is) {
599 s->coder->search_for_is(s, avctx, cpe);
600 if (cpe->is_mode) is_mode = 1;
602 if (s->coder->set_special_band_scalefactors)
603 for (ch = 0; ch < chans; ch++)
604 s->coder->set_special_band_scalefactors(s, &cpe->ch[ch]);
605 adjust_frame_information(cpe, chans);
607 put_bits(&s->pb, 1, cpe->common_window);
608 if (cpe->common_window) {
609 put_ics_info(s, &cpe->ch[0].ics);
610 encode_ms_info(&s->pb, cpe);
611 if (cpe->ms_mode) ms_mode = 1;
614 for (ch = 0; ch < chans; ch++) {
615 s->cur_channel = start_ch + ch;
616 encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
621 frame_bits = put_bits_count(&s->pb);
622 if (frame_bits <= 6144 * s->channels - 3) {
623 s->psy.bitres.bits = frame_bits / s->channels;
626 if (is_mode || ms_mode) {
627 for (i = 0; i < s->chan_map[0]; i++) {
628 // Must restore coeffs
629 chans = tag == TYPE_CPE ? 2 : 1;
631 for (ch = 0; ch < chans; ch++)
632 memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
636 s->lambda *= avctx->bit_rate * 1024.0f / avctx->sample_rate / frame_bits;
640 put_bits(&s->pb, 3, TYPE_END);
641 flush_put_bits(&s->pb);
642 avctx->frame_bits = put_bits_count(&s->pb);
644 // rate control stuff
645 if (!(avctx->flags & AV_CODEC_FLAG_QSCALE)) {
646 float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits;
648 s->lambda = FFMIN(s->lambda, 65536.f);
654 ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
657 avpkt->size = put_bits_count(&s->pb) >> 3;
662 static av_cold int aac_encode_end(AVCodecContext *avctx)
664 AACEncContext *s = avctx->priv_data;
666 ff_mdct_end(&s->mdct1024);
667 ff_mdct_end(&s->mdct128);
670 ff_psy_preprocess_end(s->psypp);
671 av_freep(&s->buffer.samples);
674 ff_af_queue_close(&s->afq);
678 static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
682 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
684 return AVERROR(ENOMEM);
687 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
688 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
689 ff_init_ff_sine_windows(10);
690 ff_init_ff_sine_windows(7);
692 if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)
694 if ((ret = ff_mdct_init(&s->mdct128, 8, 0, 32768.0)) < 0)
700 static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
703 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->buffer.samples, s->channels, 3 * 1024 * sizeof(s->buffer.samples[0]), alloc_fail);
704 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->cpe, s->chan_map[0], sizeof(ChannelElement), alloc_fail);
705 FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + AV_INPUT_BUFFER_PADDING_SIZE, alloc_fail);
707 for(ch = 0; ch < s->channels; ch++)
708 s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
712 return AVERROR(ENOMEM);
715 static av_cold int aac_encode_init(AVCodecContext *avctx)
717 AACEncContext *s = avctx->priv_data;
719 const uint8_t *sizes[2];
720 uint8_t grouping[AAC_MAX_CHANNELS];
723 avctx->frame_size = 1024;
725 for (i = 0; i < 16; i++)
726 if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
729 s->channels = avctx->channels;
731 ERROR_IF(i == 16 || i >= swb_size_1024_len || i >= swb_size_128_len,
732 "Unsupported sample rate %d\n", avctx->sample_rate);
733 ERROR_IF(s->channels > AAC_MAX_CHANNELS,
734 "Unsupported number of channels: %d\n", s->channels);
735 ERROR_IF(avctx->profile != FF_PROFILE_UNKNOWN && avctx->profile != FF_PROFILE_AAC_LOW,
736 "Unsupported profile %d\n", avctx->profile);
737 WARN_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
738 "Too many bits per frame requested, clamping to max\n");
740 avctx->bit_rate = (int)FFMIN(
741 6144 * s->channels / 1024.0 * avctx->sample_rate,
744 s->samplerate_index = i;
746 s->chan_map = aac_chan_configs[s->channels-1];
748 if ((ret = dsp_init(avctx, s)) < 0)
751 if ((ret = alloc_buffers(avctx, s)) < 0)
754 avctx->extradata_size = 5;
755 put_audio_specific_config(avctx);
757 sizes[0] = swb_size_1024[i];
758 sizes[1] = swb_size_128[i];
759 lengths[0] = ff_aac_num_swb_1024[i];
760 lengths[1] = ff_aac_num_swb_128[i];
761 for (i = 0; i < s->chan_map[0]; i++)
762 grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
763 if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
764 s->chan_map[0], grouping)) < 0)
766 s->psypp = ff_psy_preprocess_init(avctx);
767 s->coder = &ff_aac_coders[s->options.aac_coder];
770 ff_aac_coder_init_mips(s);
772 s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
776 avctx->initial_padding = 1024;
777 ff_af_queue_init(avctx, &s->afq);
781 aac_encode_end(avctx);
785 #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
786 static const AVOption aacenc_options[] = {
787 {"stereo_mode", "Stereo coding method", offsetof(AACEncContext, options.stereo_mode), AV_OPT_TYPE_INT, {.i64 = 0}, -1, 1, AACENC_FLAGS, "stereo_mode"},
788 {"auto", "Selected by the Encoder", 0, AV_OPT_TYPE_CONST, {.i64 = -1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
789 {"ms_off", "Disable Mid/Side coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
790 {"ms_force", "Force Mid/Side for the whole frame if possible", 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
791 {"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.aac_coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_TWOLOOP}, 0, AAC_CODER_NB-1, AACENC_FLAGS, "aac_coder"},
792 {"faac", "FAAC-inspired method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAAC}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
793 {"anmr", "ANMR method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
794 {"twoloop", "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
795 {"fast", "Constant quantizer", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
796 {"aac_pns", "Perceptual Noise Substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, AACENC_FLAGS, "aac_pns"},
797 {"disable", "Disable perceptual noise substitution", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pns"},
798 {"enable", "Enable perceptual noise substitution", 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pns"},
799 {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, AACENC_FLAGS, "intensity_stereo"},
800 {"disable", "Disable intensity stereo coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0}, INT_MIN, INT_MAX, AACENC_FLAGS, "intensity_stereo"},
801 {"enable", "Enable intensity stereo coding", 0, AV_OPT_TYPE_CONST, {.i64 = 1}, INT_MIN, INT_MAX, AACENC_FLAGS, "intensity_stereo"},
805 static const AVClass aacenc_class = {
807 av_default_item_name,
809 LIBAVUTIL_VERSION_INT,
812 AVCodec ff_aac_encoder = {
814 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
815 .type = AVMEDIA_TYPE_AUDIO,
816 .id = AV_CODEC_ID_AAC,
817 .priv_data_size = sizeof(AACEncContext),
818 .init = aac_encode_init,
819 .encode2 = aac_encode_frame,
820 .close = aac_encode_end,
821 .supported_samplerates = mpeg4audio_sample_rates,
822 .capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY |
823 AV_CODEC_CAP_EXPERIMENTAL,
824 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
825 AV_SAMPLE_FMT_NONE },
826 .priv_class = &aacenc_class,