aacpsy.c 10.7 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
/*
 * AAC 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
 */

/**
23
 * @file libavcodec/aacpsy.c
24 25 26 27 28
 * AAC encoder psychoacoustic model
 */

#include "avcodec.h"
#include "aactab.h"
29
#include "psymodel.h"
30 31 32 33 34 35 36 37 38 39 40 41 42 43

/***********************************
 *              TODOs:
 * thresholds linearization after their modifications for attaining given bitrate
 * try other bitrate controlling mechanism (maybe use ratecontrol.c?)
 * control quality for quality-based output
 **********************************/

/**
 * 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)
44 45 46

#define PSY_3GPP_RPEMIN      0.01f
#define PSY_3GPP_RPELEV      2.0f
47 48 49 50 51 52 53 54 55 56
/**
 * @}
 */

/**
 * information for single band used by 3GPP TS26.403-inspired psychoacoustic model
 */
typedef struct Psy3gppBand{
    float energy;    ///< band energy
    float ffac;      ///< form factor
57 58 59
    float thr;       ///< energy threshold
    float min_snr;   ///< minimal SNR
    float thr_quiet; ///< threshold in quiet
60 61
}Psy3gppBand;

62 63 64 65 66 67 68 69 70 71 72 73 74
/**
 * 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;

75 76 77 78 79 80 81 82 83 84
/**
 * psychoacoustic model frame type-dependent coefficients
 */
typedef struct Psy3gppCoeffs{
    float ath       [64]; ///< absolute threshold of hearing per bands
    float barks     [64]; ///< Bark value for each spectral band in long frame
    float spread_low[64]; ///< spreading factor for low-to-high threshold spreading in long frame
    float spread_hi [64]; ///< spreading factor for high-to-low threshold spreading in long frame
}Psy3gppCoeffs;

85 86 87 88 89 90 91 92
/**
 * 3GPP TS26.403-inspired psychoacoustic model specific data
 */
typedef struct Psy3gppContext{
    Psy3gppCoeffs psy_coef[2];
    Psy3gppChannel *ch;
}Psy3gppContext;

93 94 95
/**
 * Calculate Bark value for given line.
 */
96
static av_cold float calc_bark(float f)
97 98 99
{
    return 13.3f * atanf(0.00076f * f) + 3.5f * atanf((f / 7500.0f) * (f / 7500.0f));
}
100 101 102 103 104 105 106 107 108

#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;
109
    return    3.64 * pow(f, -0.8)
110 111 112 113 114
            - 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;
}

115
static av_cold int psy_3gpp_init(FFPsyContext *ctx) {
116 117 118 119 120 121 122 123
    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;

124
    for (i = 0; i < 1024; i++)
125 126
        barks[i] = calc_bark(i * ctx->avctx->sample_rate / 2048.0);
    minath = ath(3410, ATH_ADD);
127
    for (j = 0; j < 2; j++) {
128 129 130
        Psy3gppCoeffs *coeffs = &pctx->psy_coef[j];
        i = 0;
        prev = 0.0;
131
        for (g = 0; g < ctx->num_bands[j]; g++) {
132 133 134 135
            i += ctx->bands[j][g];
            coeffs->barks[g] = (barks[i - 1] + prev) / 2.0;
            prev = barks[i - 1];
        }
136
        for (g = 0; g < ctx->num_bands[j] - 1; g++) {
137 138 139 140
            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;
141
        for (g = 0; g < ctx->num_bands[j]; g++) {
142
            minscale = ath(ctx->avctx->sample_rate * start / 1024.0, ATH_ADD);
143
            for (i = 1; i < ctx->bands[j][g]; i++)
144
                minscale = FFMIN(minscale, ath(ctx->avctx->sample_rate * (start + i) / 1024.0 / 2.0, ATH_ADD));
145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182
            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;
183 184
    int br               = ctx->avctx->bit_rate / ctx->avctx->channels;
    int attack_ratio     = br <= 16000 ? 18 : 10;
185
    Psy3gppContext *pctx = (Psy3gppContext*) ctx->model_priv_data;
186 187
    Psy3gppChannel *pch  = &pctx->ch[channel];
    uint8_t grouping     = 0;
188 189 190
    FFPsyWindowInfo wi;

    memset(&wi, 0, sizeof(wi));
191
    if (la) {
192 193 194 195
        float s[8], v;
        int switch_to_eight = 0;
        float sum = 0.0, sum2 = 0.0;
        int attack_n = 0;
196 197
        for (i = 0; i < 8; i++) {
            for (j = 0; j < 128; j++) {
198 199 200
                v = iir_filter(audio[(i*128+j)*ctx->avctx->channels], pch->iir_state);
                sum += v*v;
            }
201
            s[i]  = sum;
202 203
            sum2 += sum;
        }
204 205
        for (i = 0; i < 8; i++) {
            if (s[i] > pch->win_energy * attack_ratio) {
206
                attack_n        = i + 1;
207 208 209 210 211 212 213
                switch_to_eight = 1;
                break;
            }
        }
        pch->win_energy = pch->win_energy*7/8 + sum2/64;

        wi.window_type[1] = prev_type;
214
        switch (prev_type) {
215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230
        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];
231 232
    } else {
        for (i = 0; i < 3; i++)
233 234 235 236 237
            wi.window_type[i] = prev_type;
        grouping = (prev_type == EIGHT_SHORT_SEQUENCE) ? window_grouping[0] : 0;
    }

    wi.window_shape   = 1;
238
    if (wi.window_type[0] != EIGHT_SHORT_SEQUENCE) {
239 240
        wi.num_windows = 1;
        wi.grouping[0] = 1;
241
    } else {
242 243
        int lastgrp = 0;
        wi.num_windows = 8;
244 245
        for (i = 0; i < 8; i++) {
            if (!((grouping >> i) & 1))
246 247 248 249 250 251 252 253 254 255 256
                lastgrp = i;
            wi.grouping[lastgrp]++;
        }
    }

    return wi;
}

/**
 * Calculate band thresholds as suggested in 3GPP TS26.403
 */
257 258
static void psy_3gpp_analyze(FFPsyContext *ctx, int channel,
                             const float *coefs, FFPsyWindowInfo *wi)
259 260
{
    Psy3gppContext *pctx = (Psy3gppContext*) ctx->model_priv_data;
261
    Psy3gppChannel *pch  = &pctx->ch[channel];
262 263
    int start = 0;
    int i, w, g;
264
    const int num_bands       = ctx->num_bands[wi->num_windows == 8];
265
    const uint8_t* band_sizes = ctx->bands[wi->num_windows == 8];
266
    Psy3gppCoeffs *coeffs     = &pctx->psy_coef[wi->num_windows == 8];
267 268

    //calculate energies, initial thresholds and related values - 5.4.2 "Threshold Calculation"
269 270
    for (w = 0; w < wi->num_windows*16; w += 16) {
        for (g = 0; g < num_bands; g++) {
271 272
            Psy3gppBand *band = &pch->band[w+g];
            band->energy = 0.0f;
273
            for (i = 0; i < band_sizes[g]; i++)
274 275
                band->energy += coefs[start+i] * coefs[start+i];
            band->energy *= 1.0f / (512*512);
276 277
            band->thr     = band->energy * 0.001258925f;
            start        += band_sizes[g];
278 279 280 281 282

            ctx->psy_bands[channel*PSY_MAX_BANDS+w+g].energy = band->energy;
        }
    }
    //modify thresholds - spread, threshold in quiet - 5.4.3 "Spreaded Energy Calculation"
283
    for (w = 0; w < wi->num_windows*16; w += 16) {
284
        Psy3gppBand *band = &pch->band[w];
285
        for (g = 1; g < num_bands; g++)
286
            band[g].thr = FFMAX(band[g].thr, band[g-1].thr * coeffs->spread_low[g-1]);
287
        for (g = num_bands - 2; g >= 0; g--)
288
            band[g].thr = FFMAX(band[g].thr, band[g+1].thr * coeffs->spread_hi [g]);
289
        for (g = 0; g < num_bands; g++) {
290
            band[g].thr_quiet = FFMAX(band[g].thr, coeffs->ath[g]);
291
            if (wi->num_windows != 8 && wi->window_type[1] != EIGHT_SHORT_SEQUENCE)
292 293
                band[g].thr_quiet = FFMAX(PSY_3GPP_RPEMIN*band[g].thr_quiet,
                                          FFMIN(band[g].thr_quiet,
294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318
                                          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,
};