diff --git a/Changelog b/Changelog index f612ce7240..613277a084 100644 --- a/Changelog +++ b/Changelog @@ -27,6 +27,7 @@ version : - Electronic Arts Madcow decoder - DivX (XSUB) subtitle encoder - nonfree libamr support for AMR-NB/WB decoding/encoding removed +- Experimental AAC encoder diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 002fe0bc37..62819bc9b5 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -36,6 +36,7 @@ OBJS-$(CONFIG_VDPAU) += vdpau.o # decoders/encoders/hardware accelerators OBJS-$(CONFIG_AAC_DECODER) += aac.o aactab.o mpeg4audio.o aac_parser.o aac_ac3_parser.o +OBJS-$(CONFIG_AAC_ENCODER) += aacenc.o aaccoder.o aacpsy.o aactab.o psymodel.o iirfilter.o mdct.o fft.o mpeg4audio.o OBJS-$(CONFIG_AASC_DECODER) += aasc.o msrledec.o OBJS-$(CONFIG_AC3_DECODER) += eac3dec.o ac3dec.o ac3tab.o ac3dec_data.o ac3.o OBJS-$(CONFIG_AC3_ENCODER) += ac3enc.o ac3tab.o ac3.o diff --git a/libavcodec/aac.h b/libavcodec/aac.h index 95adcb3922..1d24a59667 100644 --- a/libavcodec/aac.h +++ b/libavcodec/aac.h @@ -116,6 +116,12 @@ typedef struct { #define MAX_PREDICTORS 672 +#define SCALE_DIV_512 36 ///< scalefactor difference that corresponds to scale difference in 512 times +#define SCALE_ONE_POS 140 ///< scalefactor index that corresponds to scale=1.0 +#define SCALE_MAX_POS 255 ///< scalefactor index maximum value +#define SCALE_MAX_DIFF 60 ///< maximum scalefactor difference allowed by standard +#define SCALE_DIFF_ZERO 60 ///< codebook index corresponding to zero scalefactor indices difference + /** * Individual Channel Stream */ @@ -126,6 +132,7 @@ typedef struct { int num_window_groups; uint8_t group_len[8]; const uint16_t *swb_offset; ///< table of offsets to the lowest spectral coefficient of a scalefactor band, sfb, for a particular window + const uint8_t *swb_sizes; ///< table of scalefactor band sizes for a particular window int num_swb; ///< number of scalefactor window bands int num_windows; int tns_max_bands; @@ -165,6 +172,7 @@ typedef struct { typedef struct { int num_pulse; + int start; int pos[4]; int amp[4]; } Pulse; @@ -189,11 +197,14 @@ typedef struct { typedef struct { IndividualChannelStream ics; TemporalNoiseShaping tns; - enum BandType band_type[120]; ///< band types + Pulse pulse; + enum BandType band_type[128]; ///< band types int band_type_run_end[120]; ///< band type run end points float sf[120]; ///< scalefactors + int sf_idx[128]; ///< scalefactor indices (used by encoder) + uint8_t zeroes[128]; ///< band is not coded (used by encoder) DECLARE_ALIGNED_16(float, coeffs[1024]); ///< coefficients for IMDCT - DECLARE_ALIGNED_16(float, saved[512]); ///< overlap + DECLARE_ALIGNED_16(float, saved[1024]); ///< overlap DECLARE_ALIGNED_16(float, ret[1024]); ///< PCM output PredictorState predictor_state[MAX_PREDICTORS]; } SingleChannelElement; @@ -203,7 +214,9 @@ typedef struct { */ typedef struct { // CPE specific - uint8_t ms_mask[120]; ///< Set if mid/side stereo is used for each scalefactor window band + int common_window; ///< Set if channels share a common 'IndividualChannelStream' in bitstream. + int ms_mode; ///< Signals mid/side stereo flags coding mode (used by encoder) + uint8_t ms_mask[128]; ///< Set if mid/side stereo is used for each scalefactor window band // shared SingleChannelElement ch[2]; // CCE specific diff --git a/libavcodec/aaccoder.c b/libavcodec/aaccoder.c new file mode 100644 index 0000000000..763d3b90dd --- /dev/null +++ b/libavcodec/aaccoder.c @@ -0,0 +1,1037 @@ +/* + * AAC coefficients encoder + * Copyright (C) 2008-2009 Konstantin Shishkov + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file libavcodec/aaccoder.c + * AAC coefficients encoder + */ + +/*********************************** + * TODOs: + * speedup quantizer selection + * add sane pulse detection + ***********************************/ + +#include "avcodec.h" +#include "put_bits.h" +#include "aac.h" +#include "aacenc.h" +#include "aactab.h" + +/** bits needed to code codebook run value for long windows */ +static const uint8_t run_value_bits_long[64] = { + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 15 +}; + +/** bits needed to code codebook run value for short windows */ +static const uint8_t run_value_bits_short[16] = { + 3, 3, 3, 3, 3, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 9 +}; + +static const uint8_t* run_value_bits[2] = { + run_value_bits_long, run_value_bits_short +}; + + +/** + * Quantize one coefficient. + * @return absolute value of the quantized coefficient + * @see 3GPP TS26.403 5.6.2 "Scalefactor determination" + */ +static av_always_inline int quant(float coef, const float Q) +{ + return pow(coef * Q, 0.75) + 0.4054; +} + +static void quantize_bands(int (*out)[2], const float *in, const float *scaled, int size, float Q34, int is_signed, int maxval) +{ + int i; + double qc; + for (i = 0; i < size; i++) { + qc = scaled[i] * Q34; + out[i][0] = (int)FFMIN((int)qc, maxval); + out[i][1] = (int)FFMIN((int)(qc + 0.4054), maxval); + if (is_signed && in[i] < 0.0f) { + out[i][0] = -out[i][0]; + out[i][1] = -out[i][1]; + } + } +} + +static void abs_pow34_v(float *out, const float* in, const int size) +{ +#ifndef USE_REALLY_FULL_SEARCH + int i; + for (i = 0; i < size; i++) { + out[i] = pow(fabsf(in[i]), 0.75); + } +#endif /* USE_REALLY_FULL_SEARCH */ +} + +static av_always_inline int quant2(float coef, const float Q) +{ + return pow(coef * Q, 0.75); +} + +static const uint8_t aac_cb_range [12] = {0, 3, 3, 3, 3, 9, 9, 8, 8, 13, 13, 17}; +static const uint8_t aac_cb_maxval[12] = {0, 1, 1, 2, 2, 4, 4, 7, 7, 12, 12, 16}; + +/** + * Calculate rate distortion cost for quantizing with given codebook + * + * @return quantization distortion + */ +static float quantize_band_cost(struct AACEncContext *s, const float *in, const float *scaled, int size, int scale_idx, int cb, + const float lambda, const float uplim, int *bits) +{ + const float IQ = ff_aac_pow2sf_tab[200 + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; + const float Q = ff_aac_pow2sf_tab[200 - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; + const float CLIPPED_ESCAPE = 165140.0f*IQ; + int i, j, k; + float cost = 0; + const int dim = cb < FIRST_PAIR_BT ? 4 : 2; + int resbits = 0; +#ifndef USE_REALLY_FULL_SEARCH + const float Q34 = pow(Q, 0.75); + const int range = aac_cb_range[cb]; + const int maxval = aac_cb_maxval[cb]; + int offs[4]; +#endif /* USE_REALLY_FULL_SEARCH */ + + if(!cb){ + for(i = 0; i < size; i++) + cost += in[i]*in[i]*lambda; + return cost; + } +#ifndef USE_REALLY_FULL_SEARCH + offs[0] = 1; + for(i = 1; i < dim; i++) + offs[i] = offs[i-1]*range; + quantize_bands(s->qcoefs, in, scaled, size, Q34, !IS_CODEBOOK_UNSIGNED(cb), maxval); +#endif /* USE_REALLY_FULL_SEARCH */ + for(i = 0; i < size; i += dim){ + float mincost; + int minidx = 0; + int minbits = 0; + const float *vec; +#ifndef USE_REALLY_FULL_SEARCH + int (*quants)[2] = &s->qcoefs[i]; + mincost = 0.0f; + for(j = 0; j < dim; j++){ + mincost += in[i+j]*in[i+j]*lambda; + } + minidx = IS_CODEBOOK_UNSIGNED(cb) ? 0 : 40; + minbits = ff_aac_spectral_bits[cb-1][minidx]; + mincost += minbits; + for(j = 0; j < (1<= CLIPPED_ESCAPE) { + di = t - CLIPPED_ESCAPE; + curbits += 21; + }else{ + int c = av_clip(quant(t, Q), 0, 8191); + di = t - c*cbrt(c)*IQ; + curbits += av_log2(c)*2 - 4 + 1; + } + }else{ + di = t - vec[k]*IQ; + } + if(vec[k] != 0.0f) + curbits++; + rd += di*di*lambda; + } + }else{ + for(k = 0; k < dim; k++){ + float di = in[i+k] - vec[k]*IQ; + rd += di*di*lambda; + } + } + rd += curbits; + if(rd < mincost){ + mincost = rd; + minidx = j; + minbits = curbits; + } + } + cost += mincost; + resbits += minbits; + if(cost >= uplim) + return uplim; + } + + if(bits) + *bits = resbits; + return cost; +} + +static void quantize_and_encode_band(struct AACEncContext *s, PutBitContext *pb, const float *in, int size, + int scale_idx, int cb, const float lambda) +{ + const float IQ = ff_aac_pow2sf_tab[200 + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; + const float Q = ff_aac_pow2sf_tab[200 - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; + const float CLIPPED_ESCAPE = 165140.0f*IQ; + const int dim = (cb < FIRST_PAIR_BT) ? 4 : 2; + int i, j, k; +#ifndef USE_REALLY_FULL_SEARCH + const float Q34 = pow(Q, 0.75); + const int range = aac_cb_range[cb]; + const int maxval = aac_cb_maxval[cb]; + int offs[4]; + float *scaled = s->scoefs; +#endif /* USE_REALLY_FULL_SEARCH */ + +//START_TIMER + if(!cb) + return; + +#ifndef USE_REALLY_FULL_SEARCH + offs[0] = 1; + for(i = 1; i < dim; i++) + offs[i] = offs[i-1]*range; + abs_pow34_v(scaled, in, size); + quantize_bands(s->qcoefs, in, scaled, size, Q34, !IS_CODEBOOK_UNSIGNED(cb), maxval); +#endif /* USE_REALLY_FULL_SEARCH */ + for(i = 0; i < size; i += dim){ + float mincost; + int minidx = 0; + int minbits = 0; + const float *vec; +#ifndef USE_REALLY_FULL_SEARCH + int (*quants)[2] = &s->qcoefs[i]; + mincost = 0.0f; + for(j = 0; j < dim; j++){ + mincost += in[i+j]*in[i+j]*lambda; + } + minidx = IS_CODEBOOK_UNSIGNED(cb) ? 0 : 40; + minbits = ff_aac_spectral_bits[cb-1][minidx]; + mincost += minbits; + for(j = 0; j < (1<= CLIPPED_ESCAPE) { + di = t - CLIPPED_ESCAPE; + curbits += 21; + }else{ + int c = av_clip(quant(t, Q), 0, 8191); + di = t - c*cbrt(c)*IQ; + curbits += av_log2(c)*2 - 4 + 1; + } + }else{ + di = t - vec[k]*IQ; + } + if(vec[k] != 0.0f) + curbits++; + rd += di*di*lambda; + } + }else{ + for(k = 0; k < dim; k++){ + float di = in[i+k] - vec[k]*IQ; + rd += di*di*lambda; + } + } + rd += curbits; + if(rd < mincost){ + mincost = rd; + minidx = curidx; + minbits = curbits; + } + } + put_bits(pb, ff_aac_spectral_bits[cb-1][minidx], ff_aac_spectral_codes[cb-1][minidx]); + if(IS_CODEBOOK_UNSIGNED(cb)) + for(j = 0; j < dim; j++) + if(ff_aac_codebook_vectors[cb-1][minidx*dim+j] != 0.0f) + put_bits(pb, 1, in[i+j] < 0.0f); + if(cb == ESC_BT){ + for(j = 0; j < 2; j++){ + if(ff_aac_codebook_vectors[cb-1][minidx*2+j] == 64.0f){ + int coef = av_clip(quant(fabsf(in[i+j]), Q), 0, 8191); + int len = av_log2(coef); + + put_bits(pb, len - 4 + 1, (1 << (len - 4 + 1)) - 2); + put_bits(pb, len, coef & ((1 << len) - 1)); + } + } + } + } +//STOP_TIMER("quantize_and_encode") +} + +/** + * structure used in optimal codebook search + */ +typedef struct BandCodingPath { + int prev_idx; ///< pointer to the previous path point + int codebook; ///< codebook for coding band run + float cost; ///< path cost + int run; +} BandCodingPath; + +/** + * Encode band info for single window group bands. + */ +static void encode_window_bands_info(AACEncContext *s, SingleChannelElement *sce, + int win, int group_len, const float lambda) +{ + BandCodingPath path[120][12]; + int w, swb, cb, start, start2, size; + int i, j; + const int max_sfb = sce->ics.max_sfb; + const int run_bits = sce->ics.num_windows == 1 ? 5 : 3; + const int run_esc = (1 << run_bits) - 1; + int idx, ppos, count; + int stackrun[120], stackcb[120], stack_len; + float next_minrd = INFINITY; + int next_mincb = 0; + + abs_pow34_v(s->scoefs, sce->coeffs, 1024); + start = win*128; + for(cb = 0; cb < 12; cb++){ + path[0][cb].cost = 0.0f; + path[0][cb].prev_idx = -1; + path[0][cb].run = 0; + } + for(swb = 0; swb < max_sfb; swb++){ + start2 = start; + size = sce->ics.swb_sizes[swb]; + if(sce->zeroes[win*16 + swb]){ + for(cb = 0; cb < 12; cb++){ + path[swb+1][cb].prev_idx = cb; + path[swb+1][cb].cost = path[swb][cb].cost; + path[swb+1][cb].run = path[swb][cb].run + 1; + } + }else{ + float minrd = next_minrd; + int mincb = next_mincb; + next_minrd = INFINITY; + next_mincb = 0; + for(cb = 0; cb < 12; cb++){ + float cost_stay_here, cost_get_here; + float rd = 0.0f; + for(w = 0; w < group_len; w++){ + FFPsyBand *band = &s->psy.psy_bands[s->cur_channel*PSY_MAX_BANDS+(win+w)*16+swb]; + rd += quantize_band_cost(s, sce->coeffs + start + w*128, + s->scoefs + start + w*128, size, + sce->sf_idx[(win+w)*16+swb], cb, + lambda / band->threshold, INFINITY, NULL); + } + cost_stay_here = path[swb][cb].cost + rd; + cost_get_here = minrd + rd + run_bits + 4; + if( run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run] + != run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run+1]) + cost_stay_here += run_bits; + if (cost_get_here < cost_stay_here) { + path[swb+1][cb].prev_idx = mincb; + path[swb+1][cb].cost = cost_get_here; + path[swb+1][cb].run = 1; + } else { + path[swb+1][cb].prev_idx = cb; + path[swb+1][cb].cost = cost_stay_here; + path[swb+1][cb].run = path[swb][cb].run + 1; + } + if (path[swb+1][cb].cost < next_minrd) { + next_minrd = path[swb+1][cb].cost; + next_mincb = cb; + } + } + } + start += sce->ics.swb_sizes[swb]; + } + + //convert resulting path from backward-linked list + stack_len = 0; + idx = 0; + for(cb = 1; cb < 12; cb++){ + if(path[max_sfb][cb].cost < path[max_sfb][idx].cost) + idx = cb; + } + ppos = max_sfb; + while(ppos > 0){ + cb = idx; + stackrun[stack_len] = path[ppos][cb].run; + stackcb [stack_len] = cb; + idx = path[ppos-path[ppos][cb].run+1][cb].prev_idx; + ppos -= path[ppos][cb].run; + stack_len++; + } + //perform actual band info encoding + start = 0; + for(i = stack_len - 1; i >= 0; i--){ + put_bits(&s->pb, 4, stackcb[i]); + count = stackrun[i]; + memset(sce->zeroes + win*16 + start, !stackcb[i], count); + //XXX: memset when band_type is also uint8_t + for(j = 0; j < count; j++){ + sce->band_type[win*16 + start] = stackcb[i]; + start++; + } + while(count >= run_esc){ + put_bits(&s->pb, run_bits, run_esc); + count -= run_esc; + } + put_bits(&s->pb, run_bits, count); + } +} + +static void encode_window_bands_info_fixed(AACEncContext *s, SingleChannelElement *sce, + int win, int group_len, const float lambda) +{ + encode_window_bands_info(s, sce, win, group_len, 1.0f); +} + + +typedef struct TrellisPath { + float cost; + int prev; + int min_val; + int max_val; +} TrellisPath; + +static void search_for_quantizers_anmr(AVCodecContext *avctx, AACEncContext *s, + SingleChannelElement *sce, const float lambda) +{ + int q, w, w2, g, start = 0; + int i; + int idx; + TrellisPath paths[256*121]; + int bandaddr[121]; + int minq; + float mincost; + + for(i = 0; i < 256; i++){ + paths[i].cost = 0.0f; + paths[i].prev = -1; + paths[i].min_val = i; + paths[i].max_val = i; + } + for(i = 256; i < 256*121; i++){ + paths[i].cost = INFINITY; + paths[i].prev = -2; + paths[i].min_val = INT_MAX; + paths[i].max_val = 0; + } + idx = 256; + abs_pow34_v(s->scoefs, sce->coeffs, 1024); + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ + start = w*128; + for(g = 0; g < sce->ics.num_swb; g++){ + const float *coefs = sce->coeffs + start; + float qmin, qmax; + int nz = 0; + + bandaddr[idx >> 8] = w*16+g; + qmin = INT_MAX; + qmax = 0.0f; + for(w2 = 0; w2 < sce->ics.group_len[w]; w2++){ + FFPsyBand *band = &s->psy.psy_bands[s->cur_channel*PSY_MAX_BANDS+(w+w2)*16+g]; + if(band->energy <= band->threshold || band->threshold == 0.0f){ + sce->zeroes[(w+w2)*16+g] = 1; + continue; + } + sce->zeroes[(w+w2)*16+g] = 0; + nz = 1; + for(i = 0; i < sce->ics.swb_sizes[g]; i++){ + float t = fabsf(coefs[w2*128+i]); + if(t > 0.0f) qmin = fminf(qmin, t); + qmax = fmaxf(qmax, t); + } + } + if(nz){ + int minscale, maxscale; + float minrd = INFINITY; + //minimum scalefactor index is when minimum nonzero coefficient after quantizing is not clipped + minscale = av_clip_uint8(log2(qmin)*4 - 69 + SCALE_ONE_POS - SCALE_DIV_512); + //maximum scalefactor index is when maximum coefficient after quantizing is still not zero + maxscale = av_clip_uint8(log2(qmax)*4 + 6 + SCALE_ONE_POS - SCALE_DIV_512); + for(q = minscale; q < maxscale; q++){ + float dists[12], dist; + memset(dists, 0, sizeof(dists)); + for(w2 = 0; w2 < sce->ics.group_len[w]; w2++){ + FFPsyBand *band = &s->psy.psy_bands[s->cur_channel*PSY_MAX_BANDS+(w+w2)*16+g]; + int cb; + for(cb = 0; cb <= ESC_BT; cb++){ + dists[cb] += quantize_band_cost(s, coefs + w2*128, s->scoefs + start + w2*128, sce->ics.swb_sizes[g], + q, cb, lambda / band->threshold, INFINITY, NULL); + } + } + dist = dists[0]; + for(i = 1; i <= ESC_BT; i++) + dist = fminf(dist, dists[i]); + minrd = fminf(minrd, dist); + + for(i = FFMAX(q - SCALE_MAX_DIFF, 0); i < FFMIN(q + SCALE_MAX_DIFF, 256); i++){ + float cost; + int minv, maxv; + if(isinf(paths[idx - 256 + i].cost)) + continue; + cost = paths[idx - 256 + i].cost + dist + + ff_aac_scalefactor_bits[q - i + SCALE_DIFF_ZERO]; + minv = FFMIN(paths[idx - 256 + i].min_val, q); + maxv = FFMAX(paths[idx - 256 + i].max_val, q); + if(cost < paths[idx + q].cost && maxv-minv < SCALE_MAX_DIFF){ + paths[idx + q].cost = cost; + paths[idx + q].prev = idx - 256 + i; + paths[idx + q].min_val = minv; + paths[idx + q].max_val = maxv; + } + } + } + }else{ + for(q = 0; q < 256; q++){ + if(!isinf(paths[idx - 256 + q].cost)){ + paths[idx + q].cost = paths[idx - 256 + q].cost + 1; + paths[idx + q].prev = idx - 256 + q; + paths[idx + q].min_val = FFMIN(paths[idx - 256 + q].min_val, q); + paths[idx + q].max_val = FFMAX(paths[idx - 256 + q].max_val, q); + continue; + } + for(i = FFMAX(q - SCALE_MAX_DIFF, 0); i < FFMIN(q + SCALE_MAX_DIFF, 256); i++){ + float cost; + int minv, maxv; + if(isinf(paths[idx - 256 + i].cost)) + continue; + cost = paths[idx - 256 + i].cost + ff_aac_scalefactor_bits[q - i + SCALE_DIFF_ZERO]; + minv = FFMIN(paths[idx - 256 + i].min_val, q); + maxv = FFMAX(paths[idx - 256 + i].max_val, q); + if(cost < paths[idx + q].cost && maxv-minv < SCALE_MAX_DIFF){ + paths[idx + q].cost = cost; + paths[idx + q].prev = idx - 256 + i; + paths[idx + q].min_val = minv; + paths[idx + q].max_val = maxv; + } + } + } + } + sce->zeroes[w*16+g] = !nz; + start += sce->ics.swb_sizes[g]; + idx += 256; + } + } + idx -= 256; + mincost = paths[idx].cost; + minq = idx; + for(i = 1; i < 256; i++){ + if(paths[idx + i].cost < mincost){ + mincost = paths[idx + i].cost; + minq = idx + i; + } + } + while(minq >= 256){ + sce->sf_idx[bandaddr[minq>>8]] = minq & 0xFF; + minq = paths[minq].prev; + } + //set the same quantizers inside window groups + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) + for(g = 0; g < sce->ics.num_swb; g++) + for(w2 = 1; w2 < sce->ics.group_len[w]; w2++) + sce->sf_idx[(w+w2)*16+g] = sce->sf_idx[w*16+g]; +} + +/** + * two-loop quantizers search taken from ISO 13818-7 Appendix C + */ +static void search_for_quantizers_twoloop(AVCodecContext *avctx, AACEncContext *s, + SingleChannelElement *sce, const float lambda) +{ + int start = 0, i, w, w2, g; + int destbits = avctx->bit_rate * 1024.0 / avctx->sample_rate / avctx->channels; + float dists[128], uplims[128]; + int fflag, minscaler; + int its = 0; + int allz = 0; + float minthr = INFINITY; + + //XXX: some heuristic to determine initial quantizers will reduce search time + memset(dists, 0, sizeof(dists)); + //determine zero bands and upper limits + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ + for(g = 0; g < sce->ics.num_swb; g++){ + int nz = 0; + float uplim = 0.0f; + for(w2 = 0; w2 < sce->ics.group_len[w]; w2++){ + FFPsyBand *band = &s->psy.psy_bands[s->cur_channel*PSY_MAX_BANDS+(w+w2)*16+g]; + uplim += band->threshold; + if(band->energy <= band->threshold || band->threshold == 0.0f){ + sce->zeroes[(w+w2)*16+g] = 1; + continue; + } + nz = 1; + } + uplims[w*16+g] = uplim *512; + sce->zeroes[w*16+g] = !nz; + if(nz) + minthr = fminf(minthr, uplim); + allz = FFMAX(allz, nz); + } + } + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ + for(g = 0; g < sce->ics.num_swb; g++){ + if(sce->zeroes[w*16+g]){ + sce->sf_idx[w*16+g] = SCALE_ONE_POS; + continue; + } + sce->sf_idx[w*16+g] = SCALE_ONE_POS + fminf(log2(uplims[w*16+g]/minthr)*4,59); + } + } + + if(!allz) + return; + abs_pow34_v(s->scoefs, sce->coeffs, 1024); + //perform two-loop search + //outer loop - improve quality + do{ + int tbits, qstep; + minscaler = sce->sf_idx[0]; + //inner loop - quantize spectrum to fit into given number of bits + qstep = its ? 1 : 32; + do{ + int prev = -1; + tbits = 0; + fflag = 0; + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ + start = w*128; + for(g = 0; g < sce->ics.num_swb; g++){ + const float *coefs = sce->coeffs + start; + const float *scaled = s->scoefs + start; + int bits = 0; + int cb; + float mindist = INFINITY; + int minbits = 0; + + if(sce->zeroes[w*16+g] || sce->sf_idx[w*16+g] >= 218) + continue; + minscaler = FFMIN(minscaler, sce->sf_idx[w*16+g]); + for(cb = 0; cb <= ESC_BT; cb++){ + float dist = 0.0f; + int bb = 0; + for(w2 = 0; w2 < sce->ics.group_len[w]; w2++){ + int b; + dist += quantize_band_cost(s, coefs + w2*128, + scaled + w2*128, + sce->ics.swb_sizes[g], + sce->sf_idx[w*16+g], + ESC_BT, + 1.0, + INFINITY, + &b); + bb += b; + } + if(dist < mindist){ + mindist = dist; + minbits = bb; + } + } + dists[w*16+g] = mindist - minbits; + bits = minbits; + if(prev != -1){ + bits += ff_aac_scalefactor_bits[sce->sf_idx[w*16+g] - prev + SCALE_DIFF_ZERO]; + } + tbits += bits; + start += sce->ics.swb_sizes[g]; + prev = sce->sf_idx[w*16+g]; + } + } + if(tbits > destbits){ + for(i = 0; i < 128; i++){ + if(sce->sf_idx[i] < 218 - qstep){ + sce->sf_idx[i] += qstep; + } + } + }else{ + for(i = 0; i < 128; i++){ + if(sce->sf_idx[i] > 60 - qstep){ + sce->sf_idx[i] -= qstep; + } + } + } + qstep >>= 1; + if(!qstep && tbits > destbits*1.02) + qstep = 1; + if(sce->sf_idx[0] >= 217)break; + }while(qstep); + + fflag = 0; + minscaler = av_clip(minscaler, 60, 255 - SCALE_MAX_DIFF); + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ + start = w*128; + for(g = 0; g < sce->ics.num_swb; g++){ + int prevsc = sce->sf_idx[w*16+g]; + if(dists[w*16+g] > uplims[w*16+g] && sce->sf_idx[w*16+g] > 60) + sce->sf_idx[w*16+g]--; + sce->sf_idx[w*16+g] = av_clip(sce->sf_idx[w*16+g], minscaler, minscaler + SCALE_MAX_DIFF); + sce->sf_idx[w*16+g] = FFMIN(sce->sf_idx[w*16+g], 219); + if(sce->sf_idx[w*16+g] != prevsc) + fflag = 1; + } + } + its++; + }while(fflag && its < 10); +} + +static void search_for_quantizers_faac(AVCodecContext *avctx, AACEncContext *s, + SingleChannelElement *sce, const float lambda) +{ + int start = 0, i, w, w2, g; + float uplim[128], maxq[128]; + int minq, maxsf; + float distfact = ((sce->ics.num_windows > 1) ? 85.80 : 147.84) / lambda; + int last = 0, lastband = 0, curband = 0; + float avg_energy = 0.0; + if(sce->ics.num_windows == 1){ + start = 0; + for(i = 0; i < 1024; i++){ + if(i - start >= sce->ics.swb_sizes[curband]){ + start += sce->ics.swb_sizes[curband]; + curband++; + } + if(sce->coeffs[i]){ + avg_energy += sce->coeffs[i] * sce->coeffs[i]; + last = i; + lastband = curband; + } + } + }else{ + for(w = 0; w < 8; w++){ + const float *coeffs = sce->coeffs + w*128; + start = 0; + for(i = 0; i < 128; i++){ + if(i - start >= sce->ics.swb_sizes[curband]){ + start += sce->ics.swb_sizes[curband]; + curband++; + } + if(coeffs[i]){ + avg_energy += coeffs[i] * coeffs[i]; + last = FFMAX(last, i); + lastband = FFMAX(lastband, curband); + } + } + } + } + last++; + avg_energy /= last; + if(avg_energy == 0.0f){ + for(i = 0; i < FF_ARRAY_ELEMS(sce->sf_idx); i++) + sce->sf_idx[i] = SCALE_ONE_POS; + return; + } + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ + start = w*128; + for(g = 0; g < sce->ics.num_swb; g++){ + float *coefs = sce->coeffs + start; + const int size = sce->ics.swb_sizes[g]; + int start2 = start, end2 = start + size, peakpos = start; + float maxval = -1, thr = 0.0f, t; + maxq[w*16+g] = 0.0f; + if(g > lastband){ + maxq[w*16+g] = 0.0f; + start += size; + for(w2 = 0; w2 < sce->ics.group_len[w]; w2++) + memset(coefs + w2*128, 0, sizeof(coefs[0])*size); + continue; + } + for(w2 = 0; w2 < sce->ics.group_len[w]; w2++){ + for(i = 0; i < size; i++){ + float t = coefs[w2*128+i]*coefs[w2*128+i]; + maxq[w*16+g] = fmaxf(maxq[w*16+g], fabsf(coefs[w2*128 + i])); + thr += t; + if(sce->ics.num_windows == 1 && maxval < t){ + maxval = t; + peakpos = start+i; + } + } + } + if(sce->ics.num_windows == 1){ + start2 = FFMAX(peakpos - 2, start2); + end2 = FFMIN(peakpos + 3, end2); + }else{ + start2 -= start; + end2 -= start; + } + start += size; + thr = pow(thr / (avg_energy * (end2 - start2)), 0.3 + 0.1*(lastband - g) / lastband); + t = 1.0 - (1.0 * start2 / last); + uplim[w*16+g] = distfact / (1.4 * thr + t*t*t + 0.075); + } + } + memset(sce->sf_idx, 0, sizeof(sce->sf_idx)); + abs_pow34_v(s->scoefs, sce->coeffs, 1024); + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ + start = w*128; + for(g = 0; g < sce->ics.num_swb; g++){ + const float *coefs = sce->coeffs + start; + const float *scaled = s->scoefs + start; + const int size = sce->ics.swb_sizes[g]; + int scf, prev_scf, step; + int min_scf = 0, max_scf = 255; + float curdiff; + if(maxq[w*16+g] < 21.544){ + sce->zeroes[w*16+g] = 1; + start += size; + continue; + } + sce->zeroes[w*16+g] = 0; + scf = prev_scf = av_clip(SCALE_ONE_POS - SCALE_DIV_512 - log2(1/maxq[w*16+g])*16/3, 60, 218); + step = 16; + for(;;){ + float dist = 0.0f; + int quant_max; + + for(w2 = 0; w2 < sce->ics.group_len[w]; w2++){ + int b; + dist += quantize_band_cost(s, coefs + w2*128, + scaled + w2*128, + sce->ics.swb_sizes[g], + scf, + ESC_BT, + 1.0, + INFINITY, + &b); + dist -= b; + } + dist *= 1.0f/512.0f; + quant_max = quant(maxq[w*16+g], ff_aac_pow2sf_tab[200 - scf + SCALE_ONE_POS - SCALE_DIV_512]); + if(quant_max >= 8191){ // too much, return to the previous quantizer + sce->sf_idx[w*16+g] = prev_scf; + break; + } + prev_scf = scf; + curdiff = fabsf(dist - uplim[w*16+g]); + if(curdiff == 0.0f) + step = 0; + else + step = fabsf(log2(curdiff)); + if(dist > uplim[w*16+g]) + step = -step; + if(FFABS(step) <= 1 || (step > 0 && scf >= max_scf) || (step < 0 && scf <= min_scf)){ + sce->sf_idx[w*16+g] = scf; + break; + } + scf += step; + if(step > 0) + min_scf = scf; + else + max_scf = scf; + } + start += size; + } + } + minq = sce->sf_idx[0] ? sce->sf_idx[0] : INT_MAX; + for(i = 1; i < 128; i++){ + if(!sce->sf_idx[i]) + sce->sf_idx[i] = sce->sf_idx[i-1]; + else + minq = FFMIN(minq, sce->sf_idx[i]); + } + if(minq == INT_MAX) minq = 0; + minq = FFMIN(minq, SCALE_MAX_POS); + maxsf = FFMIN(minq + SCALE_MAX_DIFF, SCALE_MAX_POS); + for(i = 126; i >= 0; i--){ + if(!sce->sf_idx[i]) + sce->sf_idx[i] = sce->sf_idx[i+1]; + sce->sf_idx[i] = av_clip(sce->sf_idx[i], minq, maxsf); + } +} + +static void search_for_quantizers_fast(AVCodecContext *avctx, AACEncContext *s, + SingleChannelElement *sce, const float lambda) +{ + int start = 0, i, w, w2, g; + int minq = 255; + + memset(sce->sf_idx, 0, sizeof(sce->sf_idx)); + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ + start = w*128; + for(g = 0; g < sce->ics.num_swb; g++){ + for(w2 = 0; w2 < sce->ics.group_len[w]; w2++){ + FFPsyBand *band = &s->psy.psy_bands[s->cur_channel*PSY_MAX_BANDS+(w+w2)*16+g]; + if(band->energy <= band->threshold){ + sce->sf_idx[(w+w2)*16+g] = 218; + sce->zeroes[(w+w2)*16+g] = 1; + }else{ + sce->sf_idx[(w+w2)*16+g] = av_clip(SCALE_ONE_POS - SCALE_DIV_512 + log2(band->threshold), 80, 218); + sce->zeroes[(w+w2)*16+g] = 0; + } + minq = FFMIN(minq, sce->sf_idx[(w+w2)*16+g]); + } + } + } + for(i = 0; i < 128; i++){ + sce->sf_idx[i] = 140;//av_clip(sce->sf_idx[i], minq, minq + SCALE_MAX_DIFF - 1); + } + //set the same quantizers inside window groups + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) + for(g = 0; g < sce->ics.num_swb; g++) + for(w2 = 1; w2 < sce->ics.group_len[w]; w2++) + sce->sf_idx[(w+w2)*16+g] = sce->sf_idx[w*16+g]; +} + +static void search_for_ms(AACEncContext *s, ChannelElement *cpe, const float lambda) +{ + int start = 0, i, w, w2, g; + float M[128], S[128]; + float *L34 = s->scoefs, *R34 = s->scoefs + 128, *M34 = s->scoefs + 128*2, *S34 = s->scoefs + 128*3; + SingleChannelElement *sce0 = &cpe->ch[0]; + SingleChannelElement *sce1 = &cpe->ch[1]; + if(!cpe->common_window) + return; + for(w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]){ + for(g = 0; g < sce0->ics.num_swb; g++){ + if(!cpe->ch[0].zeroes[w*16+g] && !cpe->ch[1].zeroes[w*16+g]){ + float dist1 = 0.0f, dist2 = 0.0f; + for(w2 = 0; w2 < sce0->ics.group_len[w]; w2++){ + FFPsyBand *band0 = &s->psy.psy_bands[(s->cur_channel+0)*PSY_MAX_BANDS+(w+w2)*16+g]; + FFPsyBand *band1 = &s->psy.psy_bands[(s->cur_channel+1)*PSY_MAX_BANDS+(w+w2)*16+g]; + float minthr = fminf(band0->threshold, band1->threshold); + float maxthr = fmaxf(band0->threshold, band1->threshold); + for(i = 0; i < sce0->ics.swb_sizes[g]; i++){ + M[i] = (sce0->coeffs[start+w2*128+i] + + sce1->coeffs[start+w2*128+i])*0.5; + S[i] = sce0->coeffs[start+w2*128+i] + - sce1->coeffs[start+w2*128+i]; + } + abs_pow34_v(L34, sce0->coeffs+start+w2*128, sce0->ics.swb_sizes[g]); + abs_pow34_v(R34, sce1->coeffs+start+w2*128, sce0->ics.swb_sizes[g]); + abs_pow34_v(M34, M, sce0->ics.swb_sizes[g]); + abs_pow34_v(S34, S, sce0->ics.swb_sizes[g]); + dist1 += quantize_band_cost(s, sce0->coeffs + start + w2*128, + L34, + sce0->ics.swb_sizes[g], + sce0->sf_idx[(w+w2)*16+g], + sce0->band_type[(w+w2)*16+g], + lambda / band0->threshold, INFINITY, NULL); + dist1 += quantize_band_cost(s, sce1->coeffs + start + w2*128, + R34, + sce1->ics.swb_sizes[g], + sce1->sf_idx[(w+w2)*16+g], + sce1->band_type[(w+w2)*16+g], + lambda / band1->threshold, INFINITY, NULL); + dist2 += quantize_band_cost(s, M, + M34, + sce0->ics.swb_sizes[g], + sce0->sf_idx[(w+w2)*16+g], + sce0->band_type[(w+w2)*16+g], + lambda / maxthr, INFINITY, NULL); + dist2 += quantize_band_cost(s, S, + S34, + sce1->ics.swb_sizes[g], + sce1->sf_idx[(w+w2)*16+g], + sce1->band_type[(w+w2)*16+g], + lambda / minthr, INFINITY, NULL); + } + cpe->ms_mask[w*16+g] = dist2 < dist1; + } + start += sce0->ics.swb_sizes[g]; + } + } +} + +AACCoefficientsEncoder ff_aac_coders[] = { + { + search_for_quantizers_faac, + encode_window_bands_info_fixed, + quantize_and_encode_band, +// search_for_ms, + }, + { + search_for_quantizers_anmr, + encode_window_bands_info, + quantize_and_encode_band, +// search_for_ms, + }, + { + search_for_quantizers_twoloop, + encode_window_bands_info, + quantize_and_encode_band, +// search_for_ms, + }, + { + search_for_quantizers_fast, + encode_window_bands_info, + quantize_and_encode_band, +// search_for_ms, + }, +}; diff --git a/libavcodec/aacenc.c b/libavcodec/aacenc.c index 5537b7eac4..bc18b73a2e 100644 --- a/libavcodec/aacenc.c +++ b/libavcodec/aacenc.c @@ -26,19 +26,20 @@ /*********************************** * TODOs: - * psy model selection with some option * add sane pulse detection * add temporal noise shaping ***********************************/ #include "avcodec.h" -#include "get_bits.h" +#include "put_bits.h" #include "dsputil.h" #include "mpeg4audio.h" -#include "aacpsy.h" #include "aac.h" #include "aactab.h" +#include "aacenc.h" + +#include "psymodel.h" static const uint8_t swb_size_1024_96[] = { 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, @@ -83,7 +84,7 @@ static const uint8_t swb_size_1024_8[] = { 32, 36, 36, 40, 44, 48, 52, 56, 60, 64, 80 }; -static const uint8_t * const swb_size_1024[] = { +static const uint8_t *swb_size_1024[] = { swb_size_1024_96, swb_size_1024_96, swb_size_1024_64, swb_size_1024_48, swb_size_1024_48, swb_size_1024_32, swb_size_1024_24, swb_size_1024_24, swb_size_1024_16, @@ -110,7 +111,7 @@ static const uint8_t swb_size_128_8[] = { 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 12, 16, 20, 20 }; -static const uint8_t * const swb_size_128[] = { +static const uint8_t *swb_size_128[] = { /* the last entry on the following row is swb_size_128_64 but is a duplicate of swb_size_128_96 */ swb_size_128_96, swb_size_128_96, swb_size_128_96, @@ -119,23 +120,6 @@ static const uint8_t * const swb_size_128[] = { swb_size_128_16, swb_size_128_16, swb_size_128_8 }; -/** bits needed to code codebook run value for long windows */ -static const uint8_t run_value_bits_long[64] = { - 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, - 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 10, - 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, - 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 15 -}; - -/** bits needed to code codebook run value for short windows */ -static const uint8_t run_value_bits_short[16] = { - 3, 3, 3, 3, 3, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 9 -}; - -static const uint8_t* const run_value_bits[2] = { - run_value_bits_long, run_value_bits_short -}; - /** default channel configurations */ static const uint8_t aac_chan_configs[6][5] = { {1, TYPE_SCE}, // 1 channel - single channel element @@ -146,33 +130,6 @@ static const uint8_t aac_chan_configs[6][5] = { {4, TYPE_SCE, TYPE_CPE, TYPE_CPE, TYPE_LFE}, // 6 channels - front center + stereo + back stereo + LFE }; -/** - * structure used in optimal codebook search - */ -typedef struct BandCodingPath { - int prev_idx; ///< pointer to the previous path point - int codebook; ///< codebook for coding band run - int bits; ///< number of bit needed to code given number of bands -} BandCodingPath; - -/** - * AAC encoder context - */ -typedef struct { - PutBitContext pb; - MDCTContext mdct1024; ///< long (1024 samples) frame transform context - MDCTContext mdct128; ///< short (128 samples) frame transform context - DSPContext dsp; - DECLARE_ALIGNED_16(FFTSample, output[2048]); ///< temporary buffer for MDCT input coefficients - int16_t* samples; ///< saved preprocessed input - - int samplerate_index; ///< MPEG-4 samplerate index - - ChannelElement *cpe; ///< channel elements - AACPsyContext psy; ///< psychoacoustic model context - int last_frame; -} AACEncContext; - /** * Make AAC audio config object. * @see 1.6.2.1 "Syntax - AudioSpecificConfig" @@ -197,6 +154,8 @@ static av_cold int aac_encode_init(AVCodecContext *avctx) { AACEncContext *s = avctx->priv_data; int i; + const uint8_t *sizes[2]; + int lengths[2]; avctx->frame_size = 1024; @@ -224,25 +183,90 @@ static av_cold int aac_encode_init(AVCodecContext *avctx) s->samples = av_malloc(2 * 1024 * avctx->channels * sizeof(s->samples[0])); s->cpe = av_mallocz(sizeof(ChannelElement) * aac_chan_configs[avctx->channels-1][0]); - if(ff_aac_psy_init(&s->psy, avctx, AAC_PSY_3GPP, - aac_chan_configs[avctx->channels-1][0], 0, - swb_size_1024[i], ff_aac_num_swb_1024[i], swb_size_128[i], ff_aac_num_swb_128[i]) < 0){ - av_log(avctx, AV_LOG_ERROR, "Cannot initialize selected model.\n"); - return -1; - } avctx->extradata = av_malloc(2); avctx->extradata_size = 2; put_audio_specific_config(avctx); + + sizes[0] = swb_size_1024[i]; + sizes[1] = swb_size_128[i]; + lengths[0] = ff_aac_num_swb_1024[i]; + lengths[1] = ff_aac_num_swb_128[i]; + ff_psy_init(&s->psy, avctx, 2, sizes, lengths); + s->psypp = ff_psy_preprocess_init(avctx); + s->coder = &ff_aac_coders[0]; + + s->lambda = avctx->global_quality ? avctx->global_quality : 120; +#if !CONFIG_HARDCODED_TABLES + for (i = 0; i < 428; i++) + ff_aac_pow2sf_tab[i] = pow(2, (i - 200)/4.); +#endif /* CONFIG_HARDCODED_TABLES */ + + if (avctx->channels > 5) + av_log(avctx, AV_LOG_ERROR, "This encoder does not yet enforce the restrictions on LFEs. " + "The output will most likely be an illegal bitstream.\n"); + return 0; } +static void apply_window_and_mdct(AVCodecContext *avctx, AACEncContext *s, + SingleChannelElement *sce, short *audio, int channel) +{ + int i, j, k; + const float * lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024; + const float * swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128; + const float * pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128; + + if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) { + memcpy(s->output, sce->saved, sizeof(float)*1024); + if(sce->ics.window_sequence[0] == LONG_STOP_SEQUENCE){ + memset(s->output, 0, sizeof(s->output[0]) * 448); + for(i = 448; i < 576; i++) + s->output[i] = sce->saved[i] * pwindow[i - 448]; + for(i = 576; i < 704; i++) + s->output[i] = sce->saved[i]; + } + if(sce->ics.window_sequence[0] != LONG_START_SEQUENCE){ + j = channel; + for (i = 0; i < 1024; i++, j += avctx->channels){ + s->output[i+1024] = audio[j] * lwindow[1024 - i - 1]; + sce->saved[i] = audio[j] * lwindow[i]; + } + }else{ + j = channel; + for(i = 0; i < 448; i++, j += avctx->channels) + s->output[i+1024] = audio[j]; + for(i = 448; i < 576; i++, j += avctx->channels) + s->output[i+1024] = audio[j] * swindow[576 - i - 1]; + memset(s->output+1024+576, 0, sizeof(s->output[0]) * 448); + j = channel; + for(i = 0; i < 1024; i++, j += avctx->channels) + sce->saved[i] = audio[j]; + } + ff_mdct_calc(&s->mdct1024, sce->coeffs, s->output); + }else{ + j = channel; + for (k = 0; k < 1024; k += 128) { + for(i = 448 + k; i < 448 + k + 256; i++) + s->output[i - 448 - k] = (i < 1024) + ? sce->saved[i] + : audio[channel + (i-1024)*avctx->channels]; + s->dsp.vector_fmul (s->output, k ? swindow : pwindow, 128); + s->dsp.vector_fmul_reverse(s->output+128, s->output+128, swindow, 128); + ff_mdct_calc(&s->mdct128, sce->coeffs + k, s->output); + } + j = channel; + for(i = 0; i < 1024; i++, j += avctx->channels) + sce->saved[i] = audio[j]; + } +} + /** * Encode ics_info element. * @see Table 4.6 (syntax of ics_info) */ static void put_ics_info(AACEncContext *s, IndividualChannelStream *info) { - int i; + int w; put_bits(&s->pb, 1, 0); // ics_reserved bit put_bits(&s->pb, 2, info->window_sequence[0]); @@ -252,27 +276,118 @@ static void put_ics_info(AACEncContext *s, IndividualChannelStream *info) put_bits(&s->pb, 1, 0); // no prediction }else{ put_bits(&s->pb, 4, info->max_sfb); - for(i = 1; i < info->num_windows; i++) - put_bits(&s->pb, 1, info->group_len[i]); + for(w = 1; w < 8; w++){ + put_bits(&s->pb, 1, !info->group_len[w]); + } } } /** - * Calculate the number of bits needed to code all coefficient signs in current band. + * Encode MS data. + * @see 4.6.8.1 "Joint Coding - M/S Stereo" */ -static int calculate_band_sign_bits(AACEncContext *s, SingleChannelElement *sce, - int group_len, int start, int size) +static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe) { - int bits = 0; int i, w; - for(w = 0; w < group_len; w++){ - for(i = 0; i < size; i++){ - if(sce->icoefs[start + i]) - bits++; + + put_bits(pb, 2, cpe->ms_mode); + if(cpe->ms_mode == 1){ + for(w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w]){ + for(i = 0; i < cpe->ch[0].ics.max_sfb; i++) + put_bits(pb, 1, cpe->ms_mask[w*16 + i]); + } + } +} + +/** + * Produce integer coefficients from scalefactors provided by the model. + */ +static void adjust_frame_information(AACEncContext *apc, ChannelElement *cpe, int chans) +{ + int i, w, w2, g, ch; + int start, sum, maxsfb, cmaxsfb; + + for(ch = 0; ch < chans; ch++){ + IndividualChannelStream *ics = &cpe->ch[ch].ics; + start = 0; + maxsfb = 0; + cpe->ch[ch].pulse.num_pulse = 0; + for(w = 0; w < ics->num_windows*16; w += 16){ + for(g = 0; g < ics->num_swb; g++){ + sum = 0; + //apply M/S + if(!ch && cpe->ms_mask[w + g]){ + for(i = 0; i < ics->swb_sizes[g]; i++){ + cpe->ch[0].coeffs[start+i] = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) / 2.0; + cpe->ch[1].coeffs[start+i] = cpe->ch[0].coeffs[start+i] - cpe->ch[1].coeffs[start+i]; + } + } + start += ics->swb_sizes[g]; + } + for(cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w+cmaxsfb-1]; cmaxsfb--); + maxsfb = FFMAX(maxsfb, cmaxsfb); + } + ics->max_sfb = maxsfb; + + //adjust zero bands for window groups + for(w = 0; w < ics->num_windows; w += ics->group_len[w]){ + for(g = 0; g < ics->max_sfb; g++){ + i = 1; + for(w2 = w; w2 < w + ics->group_len[w]; w2++){ + if(!cpe->ch[ch].zeroes[w2*16 + g]){ + i = 0; + break; + } + } + cpe->ch[ch].zeroes[w*16 + g] = i; + } + } + } + + if(chans > 1 && cpe->common_window){ + IndividualChannelStream *ics0 = &cpe->ch[0].ics; + IndividualChannelStream *ics1 = &cpe->ch[1].ics; + int msc = 0; + ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb); + ics1->max_sfb = ics0->max_sfb; + for(w = 0; w < ics0->num_windows*16; w += 16) + for(i = 0; i < ics0->max_sfb; i++) + if(cpe->ms_mask[w+i]) msc++; + if(msc == 0 || ics0->max_sfb == 0) cpe->ms_mode = 0; + else cpe->ms_mode = msc < ics0->max_sfb ? 1 : 2; + } +} + +/** + * Encode scalefactor band coding type. + */ +static void encode_band_info(AACEncContext *s, SingleChannelElement *sce) +{ + int w; + + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ + s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda); + } +} + +/** + * Encode scalefactors. + */ +static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s, SingleChannelElement *sce) +{ + int off = sce->sf_idx[0], diff; + int i, w; + + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ + for(i = 0; i < sce->ics.max_sfb; i++){ + if(!sce->zeroes[w*16 + i]){ + diff = sce->sf_idx[w*16 + i] - off + SCALE_DIFF_ZERO; + if(diff < 0 || diff > 120) av_log(avctx, AV_LOG_ERROR, "Scalefactor difference is too big to be coded\n"); + off = sce->sf_idx[w*16 + i]; + put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]); + } } - start += 128; } - return bits; } /** @@ -298,27 +413,43 @@ static void encode_pulses(AACEncContext *s, Pulse *pulse) */ static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce) { - int start, i, w, w2, wg; + int start, i, w, w2; - w = 0; - for(wg = 0; wg < sce->ics.num_window_groups; wg++){ + for(w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]){ start = 0; for(i = 0; i < sce->ics.max_sfb; i++){ if(sce->zeroes[w*16 + i]){ start += sce->ics.swb_sizes[i]; continue; } - for(w2 = w; w2 < w + sce->ics.group_len[wg]; w2++){ - encode_band_coeffs(s, sce, start + w2*128, - sce->ics.swb_sizes[i], - sce->band_type[w*16 + i]); + for(w2 = w; w2 < w + sce->ics.group_len[w]; w2++){ + s->coder->quantize_and_encode_band(s, &s->pb, sce->coeffs + start + w2*128, + sce->ics.swb_sizes[i], + sce->sf_idx[w*16 + i], + sce->band_type[w*16 + i], + s->lambda); } start += sce->ics.swb_sizes[i]; } - w += sce->ics.group_len[wg]; } } +/** + * Encode one channel of audio data. + */ +static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s, SingleChannelElement *sce, int common_window) +{ + put_bits(&s->pb, 8, sce->sf_idx[0]); + if(!common_window) put_ics_info(s, &sce->ics); + encode_band_info(s, sce); + encode_scale_factors(avctx, s, sce); + encode_pulses(s, &sce->pulse); + put_bits(&s->pb, 1, 0); //tns + put_bits(&s->pb, 1, 0); //ssr + encode_spectral_coeffs(s, sce); + return 0; +} + /** * Write some auxiliary information about the created AAC file. */ @@ -339,13 +470,130 @@ static void put_bitstream_info(AVCodecContext *avctx, AACEncContext *s, const ch put_bits(&s->pb, 12 - padbits, 0); } +static int aac_encode_frame(AVCodecContext *avctx, + uint8_t *frame, int buf_size, void *data) +{ + AACEncContext *s = avctx->priv_data; + int16_t *samples = s->samples, *samples2, *la; + ChannelElement *cpe; + int i, j, chans, tag, start_ch; + const uint8_t *chan_map = aac_chan_configs[avctx->channels-1]; + int chan_el_counter[4]; + + if(s->last_frame) + return 0; + if(data){ + if(!s->psypp){ + memcpy(s->samples + 1024 * avctx->channels, data, 1024 * avctx->channels * sizeof(s->samples[0])); + }else{ + start_ch = 0; + samples2 = s->samples + 1024 * avctx->channels; + for(i = 0; i < chan_map[0]; i++){ + tag = chan_map[i+1]; + chans = tag == TYPE_CPE ? 2 : 1; + ff_psy_preprocess(s->psypp, (uint16_t*)data + start_ch, samples2 + start_ch, start_ch, chans); + start_ch += chans; + } + } + } + if(!avctx->frame_number){ + memcpy(s->samples, s->samples + 1024 * avctx->channels, 1024 * avctx->channels * sizeof(s->samples[0])); + return 0; + } + + init_put_bits(&s->pb, frame, buf_size*8); + if((avctx->frame_number & 0xFF)==1 && !(avctx->flags & CODEC_FLAG_BITEXACT)){ + put_bitstream_info(avctx, s, LIBAVCODEC_IDENT); + } + start_ch = 0; + memset(chan_el_counter, 0, sizeof(chan_el_counter)); + for(i = 0; i < chan_map[0]; i++){ + FFPsyWindowInfo wi[2]; + tag = chan_map[i+1]; + chans = tag == TYPE_CPE ? 2 : 1; + cpe = &s->cpe[i]; + samples2 = samples + start_ch; + la = samples2 + 1024 * avctx->channels + start_ch; + if(!data) la = NULL; + for(j = 0; j < chans; j++){ + IndividualChannelStream *ics = &cpe->ch[j].ics; + int k; + wi[j] = ff_psy_suggest_window(&s->psy, samples2, la, start_ch + j, ics->window_sequence[0]); + ics->window_sequence[1] = ics->window_sequence[0]; + ics->window_sequence[0] = wi[j].window_type[0]; + ics->use_kb_window[1] = ics->use_kb_window[0]; + ics->use_kb_window[0] = wi[j].window_shape; + ics->num_windows = wi[j].num_windows; + ics->swb_sizes = s->psy.bands [ics->num_windows == 8]; + ics->num_swb = s->psy.num_bands[ics->num_windows == 8]; + for(k = 0; k < ics->num_windows; k++) + ics->group_len[k] = wi[j].grouping[k]; + + s->cur_channel = start_ch + j; + apply_window_and_mdct(avctx, s, &cpe->ch[j], samples2, j); + s->coder->search_for_quantizers(avctx, s, &cpe->ch[j], s->lambda); + } + cpe->common_window = 0; + if(chans > 1 + && wi[0].window_type[0] == wi[1].window_type[0] + && wi[0].window_shape == wi[1].window_shape){ + + cpe->common_window = 1; + for(j = 0; j < wi[0].num_windows; j++){ + if(wi[0].grouping[j] != wi[1].grouping[j]){ + cpe->common_window = 0; + break; + } + } + } + if(cpe->common_window && s->coder->search_for_ms) + s->coder->search_for_ms(s, cpe, s->lambda); + adjust_frame_information(s, cpe, chans); + put_bits(&s->pb, 3, tag); + put_bits(&s->pb, 4, chan_el_counter[tag]++); + if(chans == 2){ + put_bits(&s->pb, 1, cpe->common_window); + if(cpe->common_window){ + put_ics_info(s, &cpe->ch[0].ics); + encode_ms_info(&s->pb, cpe); + } + } + for(j = 0; j < chans; j++){ + s->cur_channel = start_ch + j; + ff_psy_set_band_info(&s->psy, s->cur_channel, cpe->ch[j].coeffs, &wi[j]); + encode_individual_channel(avctx, s, &cpe->ch[j], cpe->common_window); + } + start_ch += chans; + } + + put_bits(&s->pb, 3, TYPE_END); + flush_put_bits(&s->pb); + avctx->frame_bits = put_bits_count(&s->pb); + + // rate control stuff + if(!(avctx->flags & CODEC_FLAG_QSCALE)){ + float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits; + s->lambda *= ratio; + } + + if (avctx->frame_bits > 6144*avctx->channels) { + av_log(avctx, AV_LOG_ERROR, "input buffer violation %d > %d.\n", avctx->frame_bits, 6144*avctx->channels); + } + + if(!data) + s->last_frame = 1; + memcpy(s->samples, s->samples + 1024 * avctx->channels, 1024 * avctx->channels * sizeof(s->samples[0])); + return put_bits_count(&s->pb)>>3; +} + static av_cold int aac_encode_end(AVCodecContext *avctx) { AACEncContext *s = avctx->priv_data; ff_mdct_end(&s->mdct1024); ff_mdct_end(&s->mdct128); - ff_aac_psy_end(&s->psy); + ff_psy_end(&s->psy); + ff_psy_preprocess_end(s->psypp); av_freep(&s->samples); av_freep(&s->cpe); return 0; diff --git a/libavcodec/aacenc.h b/libavcodec/aacenc.h new file mode 100644 index 0000000000..458d3d5d99 --- /dev/null +++ b/libavcodec/aacenc.h @@ -0,0 +1,71 @@ +/* + * AAC encoder + * Copyright (C) 2008 Konstantin Shishkov + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVCODEC_AACENC_H +#define AVCODEC_AACENC_H + +#include "avcodec.h" +#include "put_bits.h" +#include "dsputil.h" + +#include "aac.h" + +#include "psymodel.h" + +struct AACEncContext; + +typedef struct AACCoefficientsEncoder{ + void (*search_for_quantizers)(AVCodecContext *avctx, struct AACEncContext *s, + SingleChannelElement *sce, const float lambda); + void (*encode_window_bands_info)(struct AACEncContext *s, SingleChannelElement *sce, + int win, int group_len, const float lambda); + void (*quantize_and_encode_band)(struct AACEncContext *s, PutBitContext *pb, const float *in, int size, + int scale_idx, int cb, const float lambda); + void (*search_for_ms)(struct AACEncContext *s, ChannelElement *cpe, const float lambda); +}AACCoefficientsEncoder; + +extern AACCoefficientsEncoder ff_aac_coders[]; + +/** + * AAC encoder context + */ +typedef struct AACEncContext { + PutBitContext pb; + MDCTContext mdct1024; ///< long (1024 samples) frame transform context + MDCTContext mdct128; ///< short (128 samples) frame transform context + DSPContext dsp; + DECLARE_ALIGNED_16(FFTSample, output[2048]); ///< temporary buffer for MDCT input coefficients + int16_t* samples; ///< saved preprocessed input + + int samplerate_index; ///< MPEG-4 samplerate index + + ChannelElement *cpe; ///< channel elements + FFPsyContext psy; + struct FFPsyPreprocessContext* psypp; + AACCoefficientsEncoder *coder; + int cur_channel; + int last_frame; + float lambda; + DECLARE_ALIGNED_16(int, qcoefs[96][2]); ///< quantized coefficients + DECLARE_ALIGNED_16(float, scoefs[1024]); ///< scaled coefficients +} AACEncContext; + +#endif /* AVCODEC_AACENC_H */ diff --git a/libavcodec/aacpsy.c b/libavcodec/aacpsy.c index 45fcad46be..3880266784 100644 --- a/libavcodec/aacpsy.c +++ b/libavcodec/aacpsy.c @@ -25,54 +25,25 @@ */ #include "avcodec.h" -#include "aacpsy.h" #include "aactab.h" +#include "psymodel.h" /*********************************** * TODOs: - * General: - * better audio preprocessing (add DC highpass filter?) - * more psy models - * maybe improve coefficient quantization function in some way - * - * 3GPP-based psy model: * thresholds linearization after their modifications for attaining given bitrate * try other bitrate controlling mechanism (maybe use ratecontrol.c?) * control quality for quality-based output **********************************/ -/** - * Quantize one coefficient. - * @return absolute value of the quantized coefficient - * @see 3GPP TS26.403 5.6.2 "Scalefactor determination" - */ -static av_always_inline int quant(float coef, const float Q) -{ - return av_clip((int)(pow(fabsf(coef) * Q, 0.75) + 0.4054), 0, 8191); -} - -static inline float get_approximate_quant_error(float *c, int size, int scale_idx) -{ - int i; - int q; - float coef, unquant, sum = 0.0f; - const float Q = ff_aac_pow2sf_tab[200 - scale_idx + SCALE_ONE_POS - SCALE_DIV_512]; - const float IQ = ff_aac_pow2sf_tab[200 + scale_idx - SCALE_ONE_POS + SCALE_DIV_512]; - for(i = 0; i < size; i++){ - coef = fabs(c[i]); - q = quant(c[i], Q); - unquant = (q * cbrt(q)) * IQ; - sum += (coef - unquant) * (coef - unquant); - } - return sum; -} - /** * constants for 3GPP AAC psychoacoustic model * @{ */ #define PSY_3GPP_SPREAD_LOW 1.5f // spreading factor for ascending threshold spreading (15 dB/Bark) #define PSY_3GPP_SPREAD_HI 3.0f // spreading factor for descending threshold spreading (30 dB/Bark) + +#define PSY_3GPP_RPEMIN 0.01f +#define PSY_3GPP_RPELEV 2.0f /** * @} */ @@ -83,8 +54,24 @@ static inline float get_approximate_quant_error(float *c, int size, int scale_id typedef struct Psy3gppBand{ float energy; ///< band energy float ffac; ///< form factor + float thr; ///< energy threshold + float min_snr; ///< minimal SNR + float thr_quiet; ///< threshold in quiet }Psy3gppBand; +/** + * single/pair channel context for psychoacoustic model + */ +typedef struct Psy3gppChannel{ + Psy3gppBand band[128]; ///< bands information + Psy3gppBand prev_band[128]; ///< bands information from the previous frame + + float win_energy; ///< sliding average of channel energy + float iir_state[2]; ///< hi-pass IIR filter state + uint8_t next_grouping; ///< stored grouping scheme for the next frame (in case of 8 short window sequence) + enum WindowSequence next_window_seq; ///< window sequence to be used in the next frame +}Psy3gppChannel; + /** * psychoacoustic model frame type-dependent coefficients */ @@ -95,10 +82,241 @@ typedef struct Psy3gppCoeffs{ float spread_hi [64]; ///< spreading factor for high-to-low threshold spreading in long frame }Psy3gppCoeffs; +/** + * 3GPP TS26.403-inspired psychoacoustic model specific data + */ +typedef struct Psy3gppContext{ + Psy3gppCoeffs psy_coef[2]; + Psy3gppChannel *ch; +}Psy3gppContext; + /** * Calculate Bark value for given line. */ -static inline float calc_bark(float f) +static av_cold float calc_bark(float f) { return 13.3f * atanf(0.00076f * f) + 3.5f * atanf((f / 7500.0f) * (f / 7500.0f)); } + +#define ATH_ADD 4 +/** + * Calculate ATH value for given frequency. + * Borrowed from Lame. + */ +static av_cold float ath(float f, float add) +{ + f /= 1000.0f; + return 3.64 * pow(f, -0.8) + - 6.8 * exp(-0.6 * (f - 3.4) * (f - 3.4)) + + 6.0 * exp(-0.15 * (f - 8.7) * (f - 8.7)) + + (0.6 + 0.04 * add) * 0.001 * f * f * f * f; +} + +static av_cold int psy_3gpp_init(FFPsyContext *ctx){ + Psy3gppContext *pctx; + float barks[1024]; + int i, j, g, start; + float prev, minscale, minath; + + ctx->model_priv_data = av_mallocz(sizeof(Psy3gppContext)); + pctx = (Psy3gppContext*) ctx->model_priv_data; + + for(i = 0; i < 1024; i++) + barks[i] = calc_bark(i * ctx->avctx->sample_rate / 2048.0); + minath = ath(3410, ATH_ADD); + for(j = 0; j < 2; j++){ + Psy3gppCoeffs *coeffs = &pctx->psy_coef[j]; + i = 0; + prev = 0.0; + for(g = 0; g < ctx->num_bands[j]; g++){ + i += ctx->bands[j][g]; + coeffs->barks[g] = (barks[i - 1] + prev) / 2.0; + prev = barks[i - 1]; + } + for(g = 0; g < ctx->num_bands[j] - 1; g++){ + coeffs->spread_low[g] = pow(10.0, -(coeffs->barks[g+1] - coeffs->barks[g]) * PSY_3GPP_SPREAD_LOW); + coeffs->spread_hi [g] = pow(10.0, -(coeffs->barks[g+1] - coeffs->barks[g]) * PSY_3GPP_SPREAD_HI); + } + start = 0; + for(g = 0; g < ctx->num_bands[j]; g++){ + minscale = ath(ctx->avctx->sample_rate * start / 1024.0, ATH_ADD); + for(i = 1; i < ctx->bands[j][g]; i++){ + minscale = fminf(minscale, ath(ctx->avctx->sample_rate * (start + i) / 1024.0 / 2.0, ATH_ADD)); + } + coeffs->ath[g] = minscale - minath; + start += ctx->bands[j][g]; + } + } + + pctx->ch = av_mallocz(sizeof(Psy3gppChannel) * ctx->avctx->channels); + return 0; +} + +/** + * IIR filter used in block switching decision + */ +static float iir_filter(int in, float state[2]) +{ + float ret; + + ret = 0.7548f * (in - state[0]) + 0.5095f * state[1]; + state[0] = in; + state[1] = ret; + return ret; +} + +/** + * window grouping information stored as bits (0 - new group, 1 - group continues) + */ +static const uint8_t window_grouping[9] = { + 0xB6, 0x6C, 0xD8, 0xB2, 0x66, 0xC6, 0x96, 0x36, 0x36 +}; + +/** + * Tell encoder which window types to use. + * @see 3GPP TS26.403 5.4.1 "Blockswitching" + */ +static FFPsyWindowInfo psy_3gpp_window(FFPsyContext *ctx, + const int16_t *audio, const int16_t *la, + int channel, int prev_type) +{ + int i, j; + int br = ctx->avctx->bit_rate / ctx->avctx->channels; + int attack_ratio = br <= 16000 ? 18 : 10; + Psy3gppContext *pctx = (Psy3gppContext*) ctx->model_priv_data; + Psy3gppChannel *pch = &pctx->ch[channel]; + uint8_t grouping = 0; + FFPsyWindowInfo wi; + + memset(&wi, 0, sizeof(wi)); + if(la){ + float s[8], v; + int switch_to_eight = 0; + float sum = 0.0, sum2 = 0.0; + int attack_n = 0; + for(i = 0; i < 8; i++){ + for(j = 0; j < 128; j++){ + v = iir_filter(audio[(i*128+j)*ctx->avctx->channels], pch->iir_state); + sum += v*v; + } + s[i] = sum; + sum2 += sum; + } + for(i = 0; i < 8; i++){ + if(s[i] > pch->win_energy * attack_ratio){ + attack_n = i + 1; + switch_to_eight = 1; + break; + } + } + pch->win_energy = pch->win_energy*7/8 + sum2/64; + + wi.window_type[1] = prev_type; + switch(prev_type){ + case ONLY_LONG_SEQUENCE: + wi.window_type[0] = switch_to_eight ? LONG_START_SEQUENCE : ONLY_LONG_SEQUENCE; + break; + case LONG_START_SEQUENCE: + wi.window_type[0] = EIGHT_SHORT_SEQUENCE; + grouping = pch->next_grouping; + break; + case LONG_STOP_SEQUENCE: + wi.window_type[0] = ONLY_LONG_SEQUENCE; + break; + case EIGHT_SHORT_SEQUENCE: + wi.window_type[0] = switch_to_eight ? EIGHT_SHORT_SEQUENCE : LONG_STOP_SEQUENCE; + grouping = switch_to_eight ? pch->next_grouping : 0; + break; + } + pch->next_grouping = window_grouping[attack_n]; + }else{ + for(i = 0; i < 3; i++) + wi.window_type[i] = prev_type; + grouping = (prev_type == EIGHT_SHORT_SEQUENCE) ? window_grouping[0] : 0; + } + + wi.window_shape = 1; + if(wi.window_type[0] != EIGHT_SHORT_SEQUENCE){ + wi.num_windows = 1; + wi.grouping[0] = 1; + }else{ + int lastgrp = 0; + wi.num_windows = 8; + for(i = 0; i < 8; i++){ + if(!((grouping >> i) & 1)) + lastgrp = i; + wi.grouping[lastgrp]++; + } + } + + return wi; +} + +/** + * Calculate band thresholds as suggested in 3GPP TS26.403 + */ +static void psy_3gpp_analyze(FFPsyContext *ctx, int channel, const float *coefs, + FFPsyWindowInfo *wi) +{ + Psy3gppContext *pctx = (Psy3gppContext*) ctx->model_priv_data; + Psy3gppChannel *pch = &pctx->ch[channel]; + int start = 0; + int i, w, g; + const int num_bands = ctx->num_bands[wi->num_windows == 8]; + const uint8_t* band_sizes = ctx->bands[wi->num_windows == 8]; + Psy3gppCoeffs *coeffs = &pctx->psy_coef[wi->num_windows == 8]; + + //calculate energies, initial thresholds and related values - 5.4.2 "Threshold Calculation" + for(w = 0; w < wi->num_windows*16; w += 16){ + for(g = 0; g < num_bands; g++){ + Psy3gppBand *band = &pch->band[w+g]; + band->energy = 0.0f; + for(i = 0; i < band_sizes[g]; i++) + band->energy += coefs[start+i] * coefs[start+i]; + band->energy *= 1.0f / (512*512); + band->thr = band->energy * 0.001258925f; + start += band_sizes[g]; + + ctx->psy_bands[channel*PSY_MAX_BANDS+w+g].energy = band->energy; + } + } + //modify thresholds - spread, threshold in quiet - 5.4.3 "Spreaded Energy Calculation" + for(w = 0; w < wi->num_windows*16; w += 16){ + Psy3gppBand *band = &pch->band[w]; + for(g = 1; g < num_bands; g++){ + band[g].thr = FFMAX(band[g].thr, band[g-1].thr * coeffs->spread_low[g-1]); + } + for(g = num_bands - 2; g >= 0; g--){ + band[g].thr = FFMAX(band[g].thr, band[g+1].thr * coeffs->spread_hi [g]); + } + for(g = 0; g < num_bands; g++){ + band[g].thr_quiet = FFMAX(band[g].thr, coeffs->ath[g]); + if(wi->num_windows != 8 && wi->window_type[1] != EIGHT_SHORT_SEQUENCE){ + band[g].thr_quiet = fmaxf(PSY_3GPP_RPEMIN*band[g].thr_quiet, + fminf(band[g].thr_quiet, + PSY_3GPP_RPELEV*pch->prev_band[w+g].thr_quiet)); + } + band[g].thr = FFMAX(band[g].thr, band[g].thr_quiet * 0.25); + + ctx->psy_bands[channel*PSY_MAX_BANDS+w+g].threshold = band[g].thr; + } + } + memcpy(pch->prev_band, pch->band, sizeof(pch->band)); +} + +static av_cold void psy_3gpp_end(FFPsyContext *apc) +{ + Psy3gppContext *pctx = (Psy3gppContext*) apc->model_priv_data; + av_freep(&pctx->ch); + av_freep(&apc->model_priv_data); +} + + +const FFPsyModel ff_aac_psy_model = +{ + .name = "3GPP TS 26.403-inspired model", + .init = psy_3gpp_init, + .window = psy_3gpp_window, + .analyze = psy_3gpp_analyze, + .end = psy_3gpp_end, +}; diff --git a/libavcodec/allcodecs.c b/libavcodec/allcodecs.c index 6a171147f8..c289523a8f 100644 --- a/libavcodec/allcodecs.c +++ b/libavcodec/allcodecs.c @@ -195,7 +195,7 @@ void avcodec_register_all(void) REGISTER_ENCDEC (ZMBV, zmbv); /* audio codecs */ - REGISTER_DECODER (AAC, aac); + REGISTER_ENCDEC (AAC, aac); REGISTER_ENCDEC (AC3, ac3); REGISTER_ENCDEC (ALAC, alac); REGISTER_DECODER (APE, ape); diff --git a/libavcodec/psymodel.c b/libavcodec/psymodel.c new file mode 100644 index 0000000000..623254e531 --- /dev/null +++ b/libavcodec/psymodel.c @@ -0,0 +1,130 @@ +/* + * audio encoder psychoacoustic model + * Copyright (C) 2008 Konstantin Shishkov + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "avcodec.h" +#include "psymodel.h" +#include "iirfilter.h" + +extern const FFPsyModel ff_aac_psy_model; + +av_cold int ff_psy_init(FFPsyContext *ctx, AVCodecContext *avctx, + int num_lens, + const uint8_t **bands, const int* num_bands) +{ + ctx->avctx = avctx; + ctx->psy_bands = av_mallocz(sizeof(FFPsyBand) * PSY_MAX_BANDS * avctx->channels); + ctx->bands = av_malloc (sizeof(ctx->bands[0]) * num_lens); + ctx->num_bands = av_malloc (sizeof(ctx->num_bands[0]) * num_lens); + memcpy(ctx->bands, bands, sizeof(ctx->bands[0]) * num_lens); + memcpy(ctx->num_bands, num_bands, sizeof(ctx->num_bands[0]) * num_lens); + switch(ctx->avctx->codec_id){ + case CODEC_ID_AAC: + ctx->model = &ff_aac_psy_model; + break; + } + if(ctx->model->init) + return ctx->model->init(ctx); + return 0; +} + +FFPsyWindowInfo ff_psy_suggest_window(FFPsyContext *ctx, + const int16_t *audio, const int16_t *la, + int channel, int prev_type) +{ + return ctx->model->window(ctx, audio, la, channel, prev_type); +} + +void ff_psy_set_band_info(FFPsyContext *ctx, int channel, + const float *coeffs, FFPsyWindowInfo *wi) +{ + ctx->model->analyze(ctx, channel, coeffs, wi); +} + +av_cold void ff_psy_end(FFPsyContext *ctx) +{ + if(ctx->model->end) + ctx->model->end(ctx); + av_freep(&ctx->bands); + av_freep(&ctx->num_bands); + av_freep(&ctx->psy_bands); +} + +typedef struct FFPsyPreprocessContext{ + AVCodecContext *avctx; + float stereo_att; + struct FFIIRFilterCoeffs *fcoeffs; + struct FFIIRFilterState **fstate; +}FFPsyPreprocessContext; + +#define FILT_ORDER 4 + +av_cold struct FFPsyPreprocessContext* ff_psy_preprocess_init(AVCodecContext *avctx) +{ + FFPsyPreprocessContext *ctx; + int i; + float cutoff_coeff; + ctx = av_mallocz(sizeof(FFPsyPreprocessContext)); + ctx->avctx = avctx; + + if(avctx->flags & CODEC_FLAG_QSCALE) + cutoff_coeff = 1.0f / av_clip(1 + avctx->global_quality / FF_QUALITY_SCALE, 1, 8); + else + cutoff_coeff = avctx->bit_rate / (4.0f * avctx->sample_rate * avctx->channels); + + ctx->fcoeffs = ff_iir_filter_init_coeffs(FF_FILTER_TYPE_BUTTERWORTH, FF_FILTER_MODE_LOWPASS, + FILT_ORDER, cutoff_coeff, 0.0, 0.0); + if(ctx->fcoeffs){ + ctx->fstate = av_mallocz(sizeof(ctx->fstate[0]) * avctx->channels); + for(i = 0; i < avctx->channels; i++) + ctx->fstate[i] = ff_iir_filter_init_state(FILT_ORDER); + } + return ctx; +} + +void ff_psy_preprocess(struct FFPsyPreprocessContext *ctx, + const int16_t *audio, int16_t *dest, + int tag, int channels) +{ + int ch, i; + if(ctx->fstate){ + for(ch = 0; ch < channels; ch++){ + ff_iir_filter(ctx->fcoeffs, ctx->fstate[tag+ch], ctx->avctx->frame_size, + audio + ch, ctx->avctx->channels, + dest + ch, ctx->avctx->channels); + } + }else{ + for(ch = 0; ch < channels; ch++){ + for(i = 0; i < ctx->avctx->frame_size; i++) + dest[i*ctx->avctx->channels + ch] = audio[i*ctx->avctx->channels + ch]; + } + } +} + +av_cold void ff_psy_preprocess_end(struct FFPsyPreprocessContext *ctx) +{ + int i; + ff_iir_filter_free_coeffs(ctx->fcoeffs); + if (ctx->fstate) + for (i = 0; i < ctx->avctx->channels; i++) + ff_iir_filter_free_state(ctx->fstate[i]); + av_freep(&ctx->fstate); +} + diff --git a/libavcodec/psymodel.h b/libavcodec/psymodel.h new file mode 100644 index 0000000000..5bcf556f0b --- /dev/null +++ b/libavcodec/psymodel.h @@ -0,0 +1,158 @@ +/* + * audio encoder psychoacoustic model + * Copyright (C) 2008 Konstantin Shishkov + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVCODEC_PSYMODEL_H +#define AVCODEC_PSYMODEL_H + +#include "avcodec.h" + +/** maximum possible number of bands */ +#define PSY_MAX_BANDS 128 + +/** + * single band psychoacoustic information + */ +typedef struct FFPsyBand{ + int bits; + float energy; + float threshold; + float distortion; + float perceptual_weight; +}FFPsyBand; + +/** + * windowing related information + */ +typedef struct FFPsyWindowInfo{ + int window_type[3]; ///< window type (short/long/transitional, etc.) - current, previous and next + int window_shape; ///< window shape (sine/KBD/whatever) + int num_windows; ///< number of windows in a frame + int grouping[8]; ///< window grouping (for e.g. AAC) + int *window_sizes; ///< sequence of window sizes inside one frame (for eg. WMA) +}FFPsyWindowInfo; + +/** + * context used by psychoacoustic model + */ +typedef struct FFPsyContext{ + AVCodecContext *avctx; ///< encoder context + const struct FFPsyModel *model; ///< encoder-specific model functions + + FFPsyBand *psy_bands; ///< frame bands information + + uint8_t **bands; ///< scalefactor band sizes for possible frame sizes + int *num_bands; ///< number of scalefactor bands for possible frame sizes + int num_lens; ///< number of scalefactor band sets + + void* model_priv_data; ///< psychoacoustic model implementation private data +}FFPsyContext; + +/** + * codec-specific psychoacoustic model implementation + */ +typedef struct FFPsyModel { + const char *name; + int (*init) (FFPsyContext *apc); + FFPsyWindowInfo (*window)(FFPsyContext *ctx, const int16_t *audio, const int16_t *la, int channel, int prev_type); + void (*analyze)(FFPsyContext *ctx, int channel, const float *coeffs, FFPsyWindowInfo *wi); + void (*end) (FFPsyContext *apc); +}FFPsyModel; + +/** + * Initialize psychoacoustic model. + * + * @param ctx model context + * @param avctx codec context + * @param num_lens number of possible frame lengths + * @param bands scalefactor band lengths for all frame lengths + * @param num_bands number of scalefactor bands for all frame lengths + * + * @return zero if successful, a negative value if not + */ +av_cold int ff_psy_init(FFPsyContext *ctx, AVCodecContext *avctx, + int num_lens, + const uint8_t **bands, const int* num_bands); + +/** + * Suggest window sequence for channel. + * + * @param ctx model context + * @param audio samples for the current frame + * @param la lookahead samples (NULL when unavailable) + * @param channel number of channel element to analyze + * @param prev_type previous window type + * + * @return suggested window information in a structure + */ +FFPsyWindowInfo ff_psy_suggest_window(FFPsyContext *ctx, + const int16_t *audio, const int16_t *la, + int channel, int prev_type); + + +/** + * Perform psychoacoustic analysis and set band info (threshold, energy). + * + * @param ctx model context + * @param channel audio channel number + * @param coeffs pointer to the transformed coefficients + * @param wi window information + */ +void ff_psy_set_band_info(FFPsyContext *ctx, int channel, const float *coeffs, + FFPsyWindowInfo *wi); + +/** + * Cleanup model context at the end. + * + * @param ctx model context + */ +av_cold void ff_psy_end(FFPsyContext *ctx); + + +/************************************************************************** + * Audio preprocessing stuff. * + * This should be moved into some audio filter eventually. * + **************************************************************************/ +struct FFPsyPreprocessContext; + +/** + * psychoacoustic model audio preprocessing initialization + */ +av_cold struct FFPsyPreprocessContext* ff_psy_preprocess_init(AVCodecContext *avctx); + +/** + * Preprocess several channel in audio frame in order to compress it better. + * + * @param ctx preprocessing context + * @param audio samples to preprocess + * @param dest place to put filtered samples + * @param tag channel number + * @param channels number of channel to preprocess (some additional work may be done on stereo pair) + */ +void ff_psy_preprocess(struct FFPsyPreprocessContext *ctx, + const int16_t *audio, int16_t *dest, + int tag, int channels); + +/** + * Cleanup audio preprocessing module. + */ +av_cold void ff_psy_preprocess_end(struct FFPsyPreprocessContext *ctx); + +#endif /* AVCODEC_PSYMODEL_H */