Subversion Repositories Kolibri OS

Rev

Blame | Last modification | View Log | RSS feed

  1. /*
  2.  * Copyright (C) 2001-2010 Krzysztof Foltman, Markus Schmidt, Thor Harald Johansen and others
  3.  * Copyright (c) 2015 Paul B Mahol
  4.  *
  5.  * This file is part of FFmpeg.
  6.  *
  7.  * FFmpeg is free software; you can redistribute it and/or
  8.  * modify it under the terms of the GNU Lesser General Public
  9.  * License as published by the Free Software Foundation; either
  10.  * version 2.1 of the License, or (at your option) any later version.
  11.  *
  12.  * FFmpeg is distributed in the hope that it will be useful,
  13.  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15.  * Lesser General Public License for more details.
  16.  *
  17.  * You should have received a copy of the GNU Lesser General Public
  18.  * License along with FFmpeg; if not, write to the Free Software
  19.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20.  */
  21.  
  22. /**
  23.  * @file
  24.  * Sidechain compressor filter
  25.  */
  26.  
  27. #include "libavutil/avassert.h"
  28. #include "libavutil/channel_layout.h"
  29. #include "libavutil/common.h"
  30. #include "libavutil/opt.h"
  31.  
  32. #include "audio.h"
  33. #include "avfilter.h"
  34. #include "formats.h"
  35. #include "internal.h"
  36.  
  37. typedef struct SidechainCompressContext {
  38.     const AVClass *class;
  39.  
  40.     double attack, attack_coeff;
  41.     double release, release_coeff;
  42.     double lin_slope;
  43.     double ratio;
  44.     double threshold;
  45.     double makeup;
  46.     double thres;
  47.     double knee;
  48.     double knee_start;
  49.     double knee_stop;
  50.     double lin_knee_start;
  51.     double compressed_knee_stop;
  52.     int link;
  53.     int detection;
  54.  
  55.     AVFrame *input_frame[2];
  56. } SidechainCompressContext;
  57.  
  58. #define OFFSET(x) offsetof(SidechainCompressContext, x)
  59. #define A AV_OPT_FLAG_AUDIO_PARAM
  60. #define F AV_OPT_FLAG_FILTERING_PARAM
  61.  
  62. static const AVOption sidechaincompress_options[] = {
  63.     { "threshold", "set threshold",    OFFSET(threshold), AV_OPT_TYPE_DOUBLE, {.dbl=0.125}, 0.000976563,    1, A|F },
  64.     { "ratio",     "set ratio",        OFFSET(ratio),     AV_OPT_TYPE_DOUBLE, {.dbl=2},               1,   20, A|F },
  65.     { "attack",    "set attack",       OFFSET(attack),    AV_OPT_TYPE_DOUBLE, {.dbl=20},           0.01, 2000, A|F },
  66.     { "release",   "set release",      OFFSET(release),   AV_OPT_TYPE_DOUBLE, {.dbl=250},          0.01, 9000, A|F },
  67.     { "makeup",    "set make up gain", OFFSET(makeup),    AV_OPT_TYPE_DOUBLE, {.dbl=2},               1,   64, A|F },
  68.     { "knee",      "set knee",         OFFSET(knee),      AV_OPT_TYPE_DOUBLE, {.dbl=2.82843},         1,    8, A|F },
  69.     { "link",      "set link type",    OFFSET(link),      AV_OPT_TYPE_INT,    {.i64=0},               0,    1, A|F, "link" },
  70.     {   "average", 0,                  0,                 AV_OPT_TYPE_CONST,  {.i64=0},               0,    0, A|F, "link" },
  71.     {   "maximum", 0,                  0,                 AV_OPT_TYPE_CONST,  {.i64=1},               0,    0, A|F, "link" },
  72.     { "detection", "set detection",    OFFSET(detection), AV_OPT_TYPE_INT,    {.i64=1},               0,    1, A|F, "detection" },
  73.     {   "peak",    0,                  0,                 AV_OPT_TYPE_CONST,  {.i64=0},               0,    0, A|F, "detection" },
  74.     {   "rms",     0,                  0,                 AV_OPT_TYPE_CONST,  {.i64=1},               0,    0, A|F, "detection" },
  75.     { NULL }
  76. };
  77.  
  78. AVFILTER_DEFINE_CLASS(sidechaincompress);
  79.  
  80. static av_cold int init(AVFilterContext *ctx)
  81. {
  82.     SidechainCompressContext *s = ctx->priv;
  83.  
  84.     s->thres = log(s->threshold);
  85.     s->lin_knee_start = s->threshold / sqrt(s->knee);
  86.     s->knee_start = log(s->lin_knee_start);
  87.     s->knee_stop = log(s->threshold * sqrt(s->knee));
  88.     s->compressed_knee_stop = (s->knee_stop - s->thres) / s->ratio + s->thres;
  89.  
  90.     return 0;
  91. }
  92.  
  93. static inline float hermite_interpolation(float x, float x0, float x1,
  94.                                           float p0, float p1,
  95.                                           float m0, float m1)
  96. {
  97.     float width = x1 - x0;
  98.     float t = (x - x0) / width;
  99.     float t2, t3;
  100.     float ct0, ct1, ct2, ct3;
  101.  
  102.     m0 *= width;
  103.     m1 *= width;
  104.  
  105.     t2 = t*t;
  106.     t3 = t2*t;
  107.     ct0 = p0;
  108.     ct1 = m0;
  109.  
  110.     ct2 = -3 * p0 - 2 * m0 + 3 * p1 - m1;
  111.     ct3 = 2 * p0 + m0  - 2 * p1 + m1;
  112.  
  113.     return ct3 * t3 + ct2 * t2 + ct1 * t + ct0;
  114. }
  115.  
  116. // A fake infinity value (because real infinity may break some hosts)
  117. #define FAKE_INFINITY (65536.0 * 65536.0)
  118.  
  119. // Check for infinity (with appropriate-ish tolerance)
  120. #define IS_FAKE_INFINITY(value) (fabs(value-FAKE_INFINITY) < 1.0)
  121.  
  122. static double output_gain(double lin_slope, double ratio, double thres,
  123.                           double knee, double knee_start, double knee_stop,
  124.                           double compressed_knee_stop, int detection)
  125. {
  126.     double slope = log(lin_slope);
  127.     double gain = 0.0;
  128.     double delta = 0.0;
  129.  
  130.     if (detection)
  131.         slope *= 0.5;
  132.  
  133.     if (IS_FAKE_INFINITY(ratio)) {
  134.         gain = thres;
  135.         delta = 0.0;
  136.     } else {
  137.         gain = (slope - thres) / ratio + thres;
  138.         delta = 1.0 / ratio;
  139.     }
  140.  
  141.     if (knee > 1.0 && slope < knee_stop)
  142.         gain = hermite_interpolation(slope, knee_start, knee_stop,
  143.                                      knee_start, compressed_knee_stop,
  144.                                      1.0, delta);
  145.  
  146.     return exp(gain - slope);
  147. }
  148.  
  149. static int filter_frame(AVFilterLink *link, AVFrame *frame)
  150. {
  151.     AVFilterContext *ctx = link->dst;
  152.     SidechainCompressContext *s = ctx->priv;
  153.     AVFilterLink *sclink = ctx->inputs[1];
  154.     AVFilterLink *outlink = ctx->outputs[0];
  155.     const double makeup = s->makeup;
  156.     const double *scsrc;
  157.     double *sample;
  158.     int nb_samples;
  159.     int ret, i, c;
  160.  
  161.     for (i = 0; i < 2; i++)
  162.         if (link == ctx->inputs[i])
  163.             break;
  164.     av_assert0(i < 2 && !s->input_frame[i]);
  165.     s->input_frame[i] = frame;
  166.  
  167.     if (!s->input_frame[0] || !s->input_frame[1])
  168.         return 0;
  169.  
  170.     nb_samples = FFMIN(s->input_frame[0]->nb_samples,
  171.                        s->input_frame[1]->nb_samples);
  172.  
  173.     sample = (double *)s->input_frame[0]->data[0];
  174.     scsrc = (const double *)s->input_frame[1]->data[0];
  175.  
  176.     for (i = 0; i < nb_samples; i++) {
  177.         double abs_sample, gain = 1.0;
  178.  
  179.         abs_sample = FFABS(scsrc[0]);
  180.  
  181.         if (s->link == 1) {
  182.             for (c = 1; c < sclink->channels; c++)
  183.                 abs_sample = FFMAX(FFABS(scsrc[c]), abs_sample);
  184.         } else {
  185.             for (c = 1; c < sclink->channels; c++)
  186.                 abs_sample += FFABS(scsrc[c]);
  187.  
  188.             abs_sample /= sclink->channels;
  189.         }
  190.  
  191.         if (s->detection)
  192.             abs_sample *= abs_sample;
  193.  
  194.         s->lin_slope += (abs_sample - s->lin_slope) * (abs_sample > s->lin_slope ? s->attack_coeff : s->release_coeff);
  195.  
  196.         if (s->lin_slope > 0.0 && s->lin_slope > s->lin_knee_start)
  197.             gain = output_gain(s->lin_slope, s->ratio, s->thres, s->knee,
  198.                                s->knee_start, s->knee_stop,
  199.                                s->compressed_knee_stop, s->detection);
  200.  
  201.         for (c = 0; c < outlink->channels; c++)
  202.             sample[c] *= gain * makeup;
  203.  
  204.         sample += outlink->channels;
  205.         scsrc += sclink->channels;
  206.     }
  207.  
  208.     ret = ff_filter_frame(outlink, s->input_frame[0]);
  209.  
  210.     s->input_frame[0] = NULL;
  211.     av_frame_free(&s->input_frame[1]);
  212.  
  213.     return ret;
  214. }
  215.  
  216. static int request_frame(AVFilterLink *outlink)
  217. {
  218.     AVFilterContext *ctx = outlink->src;
  219.     SidechainCompressContext *s = ctx->priv;
  220.     int i, ret;
  221.  
  222.     /* get a frame on each input */
  223.     for (i = 0; i < 2; i++) {
  224.         AVFilterLink *inlink = ctx->inputs[i];
  225.         if (!s->input_frame[i] &&
  226.             (ret = ff_request_frame(inlink)) < 0)
  227.             return ret;
  228.  
  229.         /* request the same number of samples on all inputs */
  230.         if (i == 0)
  231.             ctx->inputs[1]->request_samples = s->input_frame[0]->nb_samples;
  232.     }
  233.  
  234.     return 0;
  235. }
  236.  
  237. static int query_formats(AVFilterContext *ctx)
  238. {
  239.     AVFilterFormats *formats;
  240.     AVFilterChannelLayouts *layouts = NULL;
  241.     static const enum AVSampleFormat sample_fmts[] = {
  242.         AV_SAMPLE_FMT_DBL,
  243.         AV_SAMPLE_FMT_NONE
  244.     };
  245.     int ret, i;
  246.  
  247.     if (!ctx->inputs[0]->in_channel_layouts ||
  248.         !ctx->inputs[0]->in_channel_layouts->nb_channel_layouts) {
  249.         av_log(ctx, AV_LOG_WARNING,
  250.                "No channel layout for input 1\n");
  251.             return AVERROR(EAGAIN);
  252.     }
  253.  
  254.     ff_add_channel_layout(&layouts, ctx->inputs[0]->in_channel_layouts->channel_layouts[0]);
  255.     if (!layouts)
  256.         return AVERROR(ENOMEM);
  257.     ff_channel_layouts_ref(layouts, &ctx->outputs[0]->in_channel_layouts);
  258.  
  259.     for (i = 0; i < 2; i++) {
  260.         layouts = ff_all_channel_layouts();
  261.         if (!layouts)
  262.             return AVERROR(ENOMEM);
  263.         ff_channel_layouts_ref(layouts, &ctx->inputs[i]->out_channel_layouts);
  264.     }
  265.  
  266.     formats = ff_make_format_list(sample_fmts);
  267.     if (!formats)
  268.         return AVERROR(ENOMEM);
  269.     ret = ff_set_common_formats(ctx, formats);
  270.     if (ret < 0)
  271.         return ret;
  272.  
  273.     formats = ff_all_samplerates();
  274.     if (!formats)
  275.         return AVERROR(ENOMEM);
  276.     return ff_set_common_samplerates(ctx, formats);
  277. }
  278.  
  279. static int config_output(AVFilterLink *outlink)
  280. {
  281.     AVFilterContext *ctx = outlink->src;
  282.     SidechainCompressContext *s = ctx->priv;
  283.  
  284.     if (ctx->inputs[0]->sample_rate != ctx->inputs[1]->sample_rate) {
  285.         av_log(ctx, AV_LOG_ERROR,
  286.                "Inputs must have the same sample rate "
  287.                "%d for in0 vs %d for in1\n",
  288.                ctx->inputs[0]->sample_rate, ctx->inputs[1]->sample_rate);
  289.         return AVERROR(EINVAL);
  290.     }
  291.  
  292.     outlink->sample_rate = ctx->inputs[0]->sample_rate;
  293.     outlink->time_base   = ctx->inputs[0]->time_base;
  294.     outlink->channel_layout = ctx->inputs[0]->channel_layout;
  295.     outlink->channels = ctx->inputs[0]->channels;
  296.  
  297.     s->attack_coeff = FFMIN(1.f, 1.f / (s->attack * outlink->sample_rate / 4000.f));
  298.     s->release_coeff = FFMIN(1.f, 1.f / (s->release * outlink->sample_rate / 4000.f));
  299.  
  300.     return 0;
  301. }
  302.  
  303. static const AVFilterPad sidechaincompress_inputs[] = {
  304.     {
  305.         .name           = "main",
  306.         .type           = AVMEDIA_TYPE_AUDIO,
  307.         .filter_frame   = filter_frame,
  308.         .needs_writable = 1,
  309.         .needs_fifo     = 1,
  310.     },{
  311.         .name           = "sidechain",
  312.         .type           = AVMEDIA_TYPE_AUDIO,
  313.         .filter_frame   = filter_frame,
  314.         .needs_fifo     = 1,
  315.     },
  316.     { NULL }
  317. };
  318.  
  319. static const AVFilterPad sidechaincompress_outputs[] = {
  320.     {
  321.         .name          = "default",
  322.         .type          = AVMEDIA_TYPE_AUDIO,
  323.         .config_props  = config_output,
  324.         .request_frame = request_frame,
  325.     },
  326.     { NULL }
  327. };
  328.  
  329. AVFilter ff_af_sidechaincompress = {
  330.     .name           = "sidechaincompress",
  331.     .description    = NULL_IF_CONFIG_SMALL("Sidechain compressor."),
  332.     .priv_size      = sizeof(SidechainCompressContext),
  333.     .priv_class     = &sidechaincompress_class,
  334.     .init           = init,
  335.     .query_formats  = query_formats,
  336.     .inputs         = sidechaincompress_inputs,
  337.     .outputs        = sidechaincompress_outputs,
  338. };
  339.