git.sesse.net Git - casparcg/blob - core/mixer/audio/audio_mixer.cpp

   1 /*
   2 * Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>
   3 *
   4 * This file is part of CasparCG (www.casparcg.com).
   5 *
   6 * CasparCG is free software: you can redistribute it and/or modify
   7 * it under the terms of the GNU General Public License as published by
   8 * the Free Software Foundation, either version 3 of the License, or
   9 * (at your option) any later version.
  10 *
  11 * CasparCG is distributed in the hope that it will be useful,
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14 * GNU General Public License for more details.
  15 *
  16 * You should have received a copy of the GNU General Public License
  17 * along with CasparCG. If not, see <http://www.gnu.org/licenses/>.
  18 *
  19 * Author: Robert Nagy, ronag89@gmail.com
  20 */
  21
  22 #include "../../StdAfx.h"
  23
  24 #include "audio_mixer.h"
  25
  26 #include <core/frame/frame.h>
  27 #include <core/frame/frame_transform.h>
  28 #include <core/monitor/monitor.h>
  29
  30 #include <common/diagnostics/graph.h>
  31 #include <common/linq.h>
  32
  33 #include <boost/range/adaptors.hpp>
  34 #include <boost/range/distance.hpp>
  35 #include <boost/lexical_cast.hpp>
  36
  37 #include <map>
  38 #include <stack>
  39 #include <vector>
  40
  41 namespace caspar { namespace core {
  42
  43 struct audio_item
  44 {
  45         const void*                     tag                     = nullptr;
  46         audio_transform         transform;
  47         audio_buffer            audio_data;
  48
  49         audio_item()
  50         {
  51         }
  52
  53         audio_item(audio_item&& other)
  54                 : tag(std::move(other.tag))
  55                 , transform(std::move(other.transform))
  56                 , audio_data(std::move(other.audio_data))
  57         {
  58         }
  59 };
  60
  61 typedef cache_aligned_vector<float> audio_buffer_ps;
  62
  63 struct audio_stream
  64 {
  65         audio_transform         prev_transform;
  66         audio_buffer_ps         audio_data;
  67         bool                            is_still                = false;
  68 };
  69
  70 struct audio_mixer::impl : boost::noncopyable
  71 {
  72         monitor::subject                                        monitor_subject_                { "/audio" };
  73         std::stack<core::audio_transform>       transform_stack_;
  74         std::map<const void*, audio_stream>     audio_streams_;
  75         std::vector<audio_item>                         items_;
  76         std::vector<int>                                        audio_cadence_;
  77         video_format_desc                                       format_desc_;
  78         float                                                           master_volume_                  = 1.0f;
  79         float                                                           previous_master_volume_ = master_volume_;
  80         spl::shared_ptr<diagnostics::graph>     graph_;
  81 public:
  82         impl(spl::shared_ptr<diagnostics::graph> graph)
  83                 : graph_(std::move(graph))
  84         {
  85                 graph_->set_color("volume", diagnostics::color(1.0f, 0.8f, 0.1f));
  86                 transform_stack_.push(core::audio_transform());
  87         }
  88
  89         void push(const frame_transform& transform)
  90         {
  91                 transform_stack_.push(transform_stack_.top()*transform.audio_transform);
  92         }
  93
  94         void visit(const const_frame& frame)
  95         {
  96                 if(transform_stack_.top().volume < 0.002 || frame.audio_data().empty())
  97                         return;
  98
  99                 audio_item item;
 100                 item.tag                = frame.stream_tag();
 101                 item.transform  = transform_stack_.top();
 102                 item.audio_data = frame.audio_data();
 103
 104                 if(item.transform.is_still)
 105                         item.transform.volume = 0.0;
 106
 107                 items_.push_back(std::move(item));
 108         }
 109
 110         void begin(const core::audio_transform& transform)
 111         {
 112                 transform_stack_.push(transform_stack_.top()*transform);
 113         }
 114
 115         void pop()
 116         {
 117                 transform_stack_.pop();
 118         }
 119
 120         void set_master_volume(float volume)
 121         {
 122                 master_volume_ = volume;
 123         }
 124
 125         float get_master_volume()
 126         {
 127                 return master_volume_;
 128         }
 129
 130         audio_buffer mix(const video_format_desc& format_desc)
 131         {
 132                 if(format_desc_ != format_desc)
 133                 {
 134                         audio_streams_.clear();
 135                         audio_cadence_ = format_desc.audio_cadence;
 136                         format_desc_ = format_desc;
 137                 }
 138
 139                 std::map<const void*, audio_stream>     next_audio_streams;
 140                 std::vector<const void*> used_tags;
 141
 142                 for (auto& item : items_)
 143                 {
 144                         audio_buffer_ps next_audio;
 145
 146                         auto next_transform = item.transform;
 147                         auto prev_transform = next_transform;
 148
 149                         auto tag = item.tag;
 150
 151                         if(boost::range::find(used_tags, tag) != used_tags.end())
 152                                 continue;
 153
 154                         used_tags.push_back(tag);
 155
 156                         const auto it = audio_streams_.find(tag);
 157                         if(it != audio_streams_.end())
 158                         {
 159                                 prev_transform  = it->second.prev_transform;
 160                                 next_audio              = std::move(it->second.audio_data);
 161                         }
 162
 163                         // Skip it if there is no existing audio stream and item has no audio-data.
 164                         if(it == audio_streams_.end() && item.audio_data.empty())
 165                                 continue;
 166
 167                         const float prev_volume = static_cast<float>(prev_transform.volume) * previous_master_volume_;
 168                         const float next_volume = static_cast<float>(next_transform.volume) * master_volume_;
 169
 170                         // TODO: Move volume mixing into code below, in order to support audio sample counts not corresponding to frame audio samples.
 171                         auto alpha = (next_volume-prev_volume)/static_cast<float>(item.audio_data.size()/format_desc.audio_channels);
 172
 173                         for(size_t n = 0; n < item.audio_data.size(); ++n)
 174                         {
 175                                 auto sample_multiplier = (prev_volume + (n/format_desc_.audio_channels) * alpha);
 176                                 next_audio.push_back(item.audio_data[n] * sample_multiplier);
 177                         }
 178
 179                         next_audio_streams[tag].prev_transform  = std::move(next_transform); // Store all active tags, inactive tags will be removed at the end.
 180                         next_audio_streams[tag].audio_data              = std::move(next_audio);
 181                         next_audio_streams[tag].is_still                = item.transform.is_still;
 182                 }
 183
 184                 previous_master_volume_ = master_volume_;
 185                 items_.clear();
 186
 187                 audio_streams_ = std::move(next_audio_streams);
 188
 189                 if(audio_streams_.empty())
 190                         audio_streams_[nullptr].audio_data = audio_buffer_ps(audio_size(audio_cadence_.front()), 0.0f);
 191
 192                 { // sanity check
 193
 194                         auto nb_invalid_streams = cpplinq::from(audio_streams_)
 195                                 .select(values())
 196                                 .where([&](const audio_stream& x) { return x.audio_data.size() < audio_size(audio_cadence_.front()); })
 197                                 .count();
 198
 199                         if(nb_invalid_streams > 0)
 200                                 CASPAR_LOG(trace) << "[audio_mixer] Incorrect frame audio cadence detected.";
 201                 }
 202
 203                 std::vector<float> result_ps(audio_size(audio_cadence_.front()), 0.0f);
 204                 for (auto& stream : audio_streams_ | boost::adaptors::map_values)
 205                 {
 206                         if(stream.audio_data.size() < result_ps.size())
 207                                 stream.audio_data.resize(result_ps.size(), 0.0f);
 208
 209                         auto out = boost::range::transform(result_ps, stream.audio_data, std::begin(result_ps), std::plus<float>());
 210                         stream.audio_data.erase(std::begin(stream.audio_data), std::begin(stream.audio_data) + std::distance(std::begin(result_ps), out));
 211                 }
 212
 213                 boost::range::rotate(audio_cadence_, std::begin(audio_cadence_)+1);
 214
 215                 audio_buffer result;
 216                 result.reserve(result_ps.size());
 217                 boost::range::transform(result_ps, std::back_inserter(result), [](float sample){return static_cast<int32_t>(sample);});
 218
 219                 const int num_channels = format_desc_.audio_channels;
 220                 monitor_subject_ << monitor::message("/nb_channels") % num_channels;
 221
 222                 auto max = std::vector<int32_t>(num_channels, std::numeric_limits<int32_t>::min());
 223
 224                 for (size_t n = 0; n < result.size(); n += num_channels)
 225                         for (int ch = 0; ch < num_channels; ++ch)
 226                                 max[ch] = std::max(max[ch], std::abs(result[n + ch]));
 227
 228                 // Makes the dBFS of silence => -dynamic range of 32bit LPCM => about -192 dBFS
 229                 // Otherwise it would be -infinity
 230                 static const auto MIN_PFS = 0.5f / static_cast<float>(std::numeric_limits<int32_t>::max());
 231
 232                 for (int i = 0; i < num_channels; ++i)
 233                 {
 234                         const auto pFS = max[i] / static_cast<float>(std::numeric_limits<int32_t>::max());
 235                         const auto dBFS = 20.0f * std::log10(std::max(MIN_PFS, pFS));
 236
 237                         auto chan_str = boost::lexical_cast<std::string>(i + 1);
 238
 239                         monitor_subject_ << monitor::message("/" + chan_str + "/pFS") % pFS;
 240                         monitor_subject_ << monitor::message("/" + chan_str + "/dBFS") % dBFS;
 241                 }
 242
 243                 graph_->set_value("volume", static_cast<double>(*boost::max_element(max)) / std::numeric_limits<int32_t>::max());
 244
 245                 return result;
 246         }
 247
 248         size_t audio_size(size_t num_samples) const
 249         {
 250                 return num_samples * format_desc_.audio_channels;
 251         }
 252 };
 253
 254 audio_mixer::audio_mixer(spl::shared_ptr<diagnostics::graph> graph) : impl_(new impl(std::move(graph))){}
 255 void audio_mixer::push(const frame_transform& transform){impl_->push(transform);}
 256 void audio_mixer::visit(const const_frame& frame){impl_->visit(frame);}
 257 void audio_mixer::pop(){impl_->pop();}
 258 void audio_mixer::set_master_volume(float volume) { impl_->set_master_volume(volume); }
 259 float audio_mixer::get_master_volume() { return impl_->get_master_volume(); }
 260 audio_buffer audio_mixer::operator()(const video_format_desc& format_desc){ return impl_->mix(format_desc); }
 261 monitor::subject& audio_mixer::monitor_output(){ return impl_->monitor_subject_; }
 262
 263 }}