Libav
|
00001 /* 00002 * AAC encoder psychoacoustic model 00003 * Copyright (C) 2008 Konstantin Shishkov 00004 * 00005 * This file is part of FFmpeg. 00006 * 00007 * FFmpeg is free software; you can redistribute it and/or 00008 * modify it under the terms of the GNU Lesser General Public 00009 * License as published by the Free Software Foundation; either 00010 * version 2.1 of the License, or (at your option) any later version. 00011 * 00012 * FFmpeg is distributed in the hope that it will be useful, 00013 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00014 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00015 * Lesser General Public License for more details. 00016 * 00017 * You should have received a copy of the GNU Lesser General Public 00018 * License along with FFmpeg; if not, write to the Free Software 00019 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00020 */ 00021 00027 #include "avcodec.h" 00028 #include "aactab.h" 00029 #include "psymodel.h" 00030 00031 /*********************************** 00032 * TODOs: 00033 * thresholds linearization after their modifications for attaining given bitrate 00034 * try other bitrate controlling mechanism (maybe use ratecontrol.c?) 00035 * control quality for quality-based output 00036 **********************************/ 00037 00042 #define PSY_3GPP_SPREAD_LOW 1.5f // spreading factor for ascending threshold spreading (15 dB/Bark) 00043 #define PSY_3GPP_SPREAD_HI 3.0f // spreading factor for descending threshold spreading (30 dB/Bark) 00044 00045 #define PSY_3GPP_RPEMIN 0.01f 00046 #define PSY_3GPP_RPELEV 2.0f 00047 00054 typedef struct Psy3gppBand{ 00055 float energy; 00056 float ffac; 00057 float thr; 00058 float min_snr; 00059 float thr_quiet; 00060 }Psy3gppBand; 00061 00065 typedef struct Psy3gppChannel{ 00066 Psy3gppBand band[128]; 00067 Psy3gppBand prev_band[128]; 00068 00069 float win_energy; 00070 float iir_state[2]; 00071 uint8_t next_grouping; 00072 enum WindowSequence next_window_seq; 00073 }Psy3gppChannel; 00074 00078 typedef struct Psy3gppCoeffs{ 00079 float ath [64]; 00080 float barks [64]; 00081 float spread_low[64]; 00082 float spread_hi [64]; 00083 }Psy3gppCoeffs; 00084 00088 typedef struct Psy3gppContext{ 00089 Psy3gppCoeffs psy_coef[2]; 00090 Psy3gppChannel *ch; 00091 }Psy3gppContext; 00092 00096 static av_cold float calc_bark(float f) 00097 { 00098 return 13.3f * atanf(0.00076f * f) + 3.5f * atanf((f / 7500.0f) * (f / 7500.0f)); 00099 } 00100 00101 #define ATH_ADD 4 00102 00106 static av_cold float ath(float f, float add) 00107 { 00108 f /= 1000.0f; 00109 return 3.64 * pow(f, -0.8) 00110 - 6.8 * exp(-0.6 * (f - 3.4) * (f - 3.4)) 00111 + 6.0 * exp(-0.15 * (f - 8.7) * (f - 8.7)) 00112 + (0.6 + 0.04 * add) * 0.001 * f * f * f * f; 00113 } 00114 00115 static av_cold int psy_3gpp_init(FFPsyContext *ctx) { 00116 Psy3gppContext *pctx; 00117 float barks[1024]; 00118 int i, j, g, start; 00119 float prev, minscale, minath; 00120 00121 ctx->model_priv_data = av_mallocz(sizeof(Psy3gppContext)); 00122 pctx = (Psy3gppContext*) ctx->model_priv_data; 00123 00124 for (i = 0; i < 1024; i++) 00125 barks[i] = calc_bark(i * ctx->avctx->sample_rate / 2048.0); 00126 minath = ath(3410, ATH_ADD); 00127 for (j = 0; j < 2; j++) { 00128 Psy3gppCoeffs *coeffs = &pctx->psy_coef[j]; 00129 i = 0; 00130 prev = 0.0; 00131 for (g = 0; g < ctx->num_bands[j]; g++) { 00132 i += ctx->bands[j][g]; 00133 coeffs->barks[g] = (barks[i - 1] + prev) / 2.0; 00134 prev = barks[i - 1]; 00135 } 00136 for (g = 0; g < ctx->num_bands[j] - 1; g++) { 00137 coeffs->spread_low[g] = pow(10.0, -(coeffs->barks[g+1] - coeffs->barks[g]) * PSY_3GPP_SPREAD_LOW); 00138 coeffs->spread_hi [g] = pow(10.0, -(coeffs->barks[g+1] - coeffs->barks[g]) * PSY_3GPP_SPREAD_HI); 00139 } 00140 start = 0; 00141 for (g = 0; g < ctx->num_bands[j]; g++) { 00142 minscale = ath(ctx->avctx->sample_rate * start / 1024.0, ATH_ADD); 00143 for (i = 1; i < ctx->bands[j][g]; i++) 00144 minscale = FFMIN(minscale, ath(ctx->avctx->sample_rate * (start + i) / 1024.0 / 2.0, ATH_ADD)); 00145 coeffs->ath[g] = minscale - minath; 00146 start += ctx->bands[j][g]; 00147 } 00148 } 00149 00150 pctx->ch = av_mallocz(sizeof(Psy3gppChannel) * ctx->avctx->channels); 00151 return 0; 00152 } 00153 00157 static float iir_filter(int in, float state[2]) 00158 { 00159 float ret; 00160 00161 ret = 0.7548f * (in - state[0]) + 0.5095f * state[1]; 00162 state[0] = in; 00163 state[1] = ret; 00164 return ret; 00165 } 00166 00170 static const uint8_t window_grouping[9] = { 00171 0xB6, 0x6C, 0xD8, 0xB2, 0x66, 0xC6, 0x96, 0x36, 0x36 00172 }; 00173 00178 static FFPsyWindowInfo psy_3gpp_window(FFPsyContext *ctx, 00179 const int16_t *audio, const int16_t *la, 00180 int channel, int prev_type) 00181 { 00182 int i, j; 00183 int br = ctx->avctx->bit_rate / ctx->avctx->channels; 00184 int attack_ratio = br <= 16000 ? 18 : 10; 00185 Psy3gppContext *pctx = (Psy3gppContext*) ctx->model_priv_data; 00186 Psy3gppChannel *pch = &pctx->ch[channel]; 00187 uint8_t grouping = 0; 00188 FFPsyWindowInfo wi; 00189 00190 memset(&wi, 0, sizeof(wi)); 00191 if (la) { 00192 float s[8], v; 00193 int switch_to_eight = 0; 00194 float sum = 0.0, sum2 = 0.0; 00195 int attack_n = 0; 00196 for (i = 0; i < 8; i++) { 00197 for (j = 0; j < 128; j++) { 00198 v = iir_filter(audio[(i*128+j)*ctx->avctx->channels], pch->iir_state); 00199 sum += v*v; 00200 } 00201 s[i] = sum; 00202 sum2 += sum; 00203 } 00204 for (i = 0; i < 8; i++) { 00205 if (s[i] > pch->win_energy * attack_ratio) { 00206 attack_n = i + 1; 00207 switch_to_eight = 1; 00208 break; 00209 } 00210 } 00211 pch->win_energy = pch->win_energy*7/8 + sum2/64; 00212 00213 wi.window_type[1] = prev_type; 00214 switch (prev_type) { 00215 case ONLY_LONG_SEQUENCE: 00216 wi.window_type[0] = switch_to_eight ? LONG_START_SEQUENCE : ONLY_LONG_SEQUENCE; 00217 break; 00218 case LONG_START_SEQUENCE: 00219 wi.window_type[0] = EIGHT_SHORT_SEQUENCE; 00220 grouping = pch->next_grouping; 00221 break; 00222 case LONG_STOP_SEQUENCE: 00223 wi.window_type[0] = ONLY_LONG_SEQUENCE; 00224 break; 00225 case EIGHT_SHORT_SEQUENCE: 00226 wi.window_type[0] = switch_to_eight ? EIGHT_SHORT_SEQUENCE : LONG_STOP_SEQUENCE; 00227 grouping = switch_to_eight ? pch->next_grouping : 0; 00228 break; 00229 } 00230 pch->next_grouping = window_grouping[attack_n]; 00231 } else { 00232 for (i = 0; i < 3; i++) 00233 wi.window_type[i] = prev_type; 00234 grouping = (prev_type == EIGHT_SHORT_SEQUENCE) ? window_grouping[0] : 0; 00235 } 00236 00237 wi.window_shape = 1; 00238 if (wi.window_type[0] != EIGHT_SHORT_SEQUENCE) { 00239 wi.num_windows = 1; 00240 wi.grouping[0] = 1; 00241 } else { 00242 int lastgrp = 0; 00243 wi.num_windows = 8; 00244 for (i = 0; i < 8; i++) { 00245 if (!((grouping >> i) & 1)) 00246 lastgrp = i; 00247 wi.grouping[lastgrp]++; 00248 } 00249 } 00250 00251 return wi; 00252 } 00253 00257 static void psy_3gpp_analyze(FFPsyContext *ctx, int channel, 00258 const float *coefs, FFPsyWindowInfo *wi) 00259 { 00260 Psy3gppContext *pctx = (Psy3gppContext*) ctx->model_priv_data; 00261 Psy3gppChannel *pch = &pctx->ch[channel]; 00262 int start = 0; 00263 int i, w, g; 00264 const int num_bands = ctx->num_bands[wi->num_windows == 8]; 00265 const uint8_t* band_sizes = ctx->bands[wi->num_windows == 8]; 00266 Psy3gppCoeffs *coeffs = &pctx->psy_coef[wi->num_windows == 8]; 00267 00268 //calculate energies, initial thresholds and related values - 5.4.2 "Threshold Calculation" 00269 for (w = 0; w < wi->num_windows*16; w += 16) { 00270 for (g = 0; g < num_bands; g++) { 00271 Psy3gppBand *band = &pch->band[w+g]; 00272 band->energy = 0.0f; 00273 for (i = 0; i < band_sizes[g]; i++) 00274 band->energy += coefs[start+i] * coefs[start+i]; 00275 band->energy *= 1.0f / (512*512); 00276 band->thr = band->energy * 0.001258925f; 00277 start += band_sizes[g]; 00278 00279 ctx->psy_bands[channel*PSY_MAX_BANDS+w+g].energy = band->energy; 00280 } 00281 } 00282 //modify thresholds - spread, threshold in quiet - 5.4.3 "Spreaded Energy Calculation" 00283 for (w = 0; w < wi->num_windows*16; w += 16) { 00284 Psy3gppBand *band = &pch->band[w]; 00285 for (g = 1; g < num_bands; g++) 00286 band[g].thr = FFMAX(band[g].thr, band[g-1].thr * coeffs->spread_low[g-1]); 00287 for (g = num_bands - 2; g >= 0; g--) 00288 band[g].thr = FFMAX(band[g].thr, band[g+1].thr * coeffs->spread_hi [g]); 00289 for (g = 0; g < num_bands; g++) { 00290 band[g].thr_quiet = FFMAX(band[g].thr, coeffs->ath[g]); 00291 if (wi->num_windows != 8 && wi->window_type[1] != EIGHT_SHORT_SEQUENCE) 00292 band[g].thr_quiet = FFMAX(PSY_3GPP_RPEMIN*band[g].thr_quiet, 00293 FFMIN(band[g].thr_quiet, 00294 PSY_3GPP_RPELEV*pch->prev_band[w+g].thr_quiet)); 00295 band[g].thr = FFMAX(band[g].thr, band[g].thr_quiet * 0.25); 00296 00297 ctx->psy_bands[channel*PSY_MAX_BANDS+w+g].threshold = band[g].thr; 00298 } 00299 } 00300 memcpy(pch->prev_band, pch->band, sizeof(pch->band)); 00301 } 00302 00303 static av_cold void psy_3gpp_end(FFPsyContext *apc) 00304 { 00305 Psy3gppContext *pctx = (Psy3gppContext*) apc->model_priv_data; 00306 av_freep(&pctx->ch); 00307 av_freep(&apc->model_priv_data); 00308 } 00309 00310 00311 const FFPsyModel ff_aac_psy_model = 00312 { 00313 .name = "3GPP TS 26.403-inspired model", 00314 .init = psy_3gpp_init, 00315 .window = psy_3gpp_window, 00316 .analyze = psy_3gpp_analyze, 00317 .end = psy_3gpp_end, 00318 };