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

   1 /*\r
   2 * Copyright 2013 Sveriges Television AB http://casparcg.com/\r
   3 *\r
   4 * This file is part of CasparCG (www.casparcg.com).\r
   5 *\r
   6 * CasparCG is free software: you can redistribute it and/or modify\r
   7 * it under the terms of the GNU General Public License as published by\r
   8 * the Free Software Foundation, either version 3 of the License, or\r
   9 * (at your option) any later version.\r
  10 *\r
  11 * CasparCG is distributed in the hope that it will be useful,\r
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of\r
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the\r
  14 * GNU General Public License for more details.\r
  15 *\r
  16 * You should have received a copy of the GNU General Public License\r
  17 * along with CasparCG. If not, see <http://www.gnu.org/licenses/>.\r
  18 *\r
  19 * Author: Robert Nagy, ronag89@gmail.com\r
  20 */\r
  21 \r
  22 #include "../../stdafx.h"\r
  23 \r
  24 #include "audio_mixer.h"\r
  25 \r
  26 #include <core/mixer/write_frame.h>\r
  27 #include <core/producer/frame/frame_transform.h>\r
  28 #include <core/monitor/monitor.h>\r
  29 #include <common/diagnostics/graph.h>\r
  30 #include "audio_util.h"\r
  31 \r
  32 #include <tbb/cache_aligned_allocator.h>\r
  33 \r
  34 #include <boost/range/adaptors.hpp>\r
  35 #include <boost/range/distance.hpp>\r
  36 \r
  37 #include <map>\r
  38 #include <stack>\r
  39 #include <vector>\r
  40 \r
  41 namespace caspar { namespace core {\r
  42 \r
  43 struct audio_item\r
  44 {\r
  45         const void*                     tag;\r
  46         frame_transform         transform;\r
  47         audio_buffer            audio_data;\r
  48 \r
  49         audio_item()\r
  50         {\r
  51         }\r
  52 \r
  53         audio_item(audio_item&& other)\r
  54                 : tag(std::move(other.tag))\r
  55                 , transform(std::move(other.transform))\r
  56                 , audio_data(std::move(other.audio_data))\r
  57         {\r
  58         }\r
  59 };\r
  60 \r
  61 typedef std::vector<float, tbb::cache_aligned_allocator<float>> audio_buffer_ps;\r
  62         \r
  63 struct audio_stream\r
  64 {\r
  65         frame_transform prev_transform;\r
  66         audio_buffer_ps audio_data;\r
  67 };\r
  68 \r
  69 struct audio_mixer::implementation\r
  70 {\r
  71         safe_ptr<diagnostics::graph>            graph_;\r
  72         std::stack<core::frame_transform>       transform_stack_;\r
  73         std::map<const void*, audio_stream>     audio_streams_;\r
  74         std::vector<audio_item>                         items_;\r
  75         std::vector<size_t>                                     audio_cadence_;\r
  76         video_format_desc                                       format_desc_;\r
  77         channel_layout                                          channel_layout_;\r
  78         float                                                           master_volume_;\r
  79         float                                                           previous_master_volume_;\r
  80         monitor::subject                                        monitor_subject_;\r
  81         \r
  82 public:\r
  83         implementation(const safe_ptr<diagnostics::graph>& graph)\r
  84                 : graph_(graph)\r
  85                 , format_desc_(video_format_desc::get(video_format::invalid))\r
  86                 , channel_layout_(channel_layout::stereo())\r
  87                 , master_volume_(1.0f)\r
  88                 , previous_master_volume_(master_volume_)\r
  89                 , monitor_subject_("/audio")\r
  90         {\r
  91                 graph_->set_color("volume", diagnostics::color(1.0f, 0.8f, 0.1f));\r
  92                 transform_stack_.push(core::frame_transform());\r
  93         }\r
  94         \r
  95         void begin(core::basic_frame& frame)\r
  96         {\r
  97                 transform_stack_.push(transform_stack_.top()*frame.get_frame_transform());\r
  98         }\r
  99 \r
 100         void visit(core::write_frame& frame)\r
 101         {\r
 102                 if(transform_stack_.top().volume < 0.002 || frame.audio_data().empty())\r
 103                         return;\r
 104 \r
 105                 audio_item item;\r
 106                 item.tag                = frame.tag();\r
 107                 item.transform  = transform_stack_.top();\r
 108 \r
 109                 if (needs_rearranging(frame.get_channel_layout(), channel_layout_))\r
 110                 {\r
 111                         auto src_view = frame.get_multichannel_view();\r
 112                         \r
 113                         audio_buffer rearranged_buffer;\r
 114                         rearranged_buffer.resize(\r
 115                                         src_view.num_samples() * channel_layout_.num_channels);\r
 116 \r
 117                         auto dst_view = make_multichannel_view<int32_t>(\r
 118                                         rearranged_buffer.begin(),\r
 119                                         rearranged_buffer.end(),\r
 120                                         channel_layout_);\r
 121 \r
 122                         bool rearrange_success = rearrange_or_rearrange_and_mix(\r
 123                                         src_view, dst_view, default_mix_config_repository());\r
 124 \r
 125                         if (!rearrange_success)\r
 126                         {\r
 127                                 failed_rearrange(item.tag, src_view.channel_layout());\r
 128                         }\r
 129 \r
 130                         item.audio_data = std::move(rearranged_buffer);\r
 131                 }\r
 132                 else\r
 133                 {\r
 134                         item.audio_data = std::move(frame.audio_data()); // Note: We don't need to care about upper/lower since audio_data is removed/moved from the last field.\r
 135                 }\r
 136                 \r
 137                 items_.push_back(std::move(item));              \r
 138         }\r
 139 \r
 140         void begin(const core::frame_transform& transform)\r
 141         {\r
 142                 transform_stack_.push(transform_stack_.top()*transform);\r
 143         }\r
 144                 \r
 145         void end()\r
 146         {\r
 147                 transform_stack_.pop();\r
 148         }\r
 149 \r
 150         float get_master_volume() const\r
 151         {\r
 152                 return master_volume_;\r
 153         }\r
 154 \r
 155         void set_master_volume(float volume)\r
 156         {\r
 157                 master_volume_ = volume;\r
 158         }\r
 159         \r
 160         audio_buffer mix(const video_format_desc& format_desc, const channel_layout& layout)\r
 161         {       \r
 162                 if(format_desc_ != format_desc)\r
 163                 {\r
 164                         audio_streams_.clear();\r
 165                         audio_cadence_ = format_desc.audio_cadence;\r
 166                         format_desc_ = format_desc;\r
 167                         channel_layout_ = layout;\r
 168                 }\r
 169                 \r
 170                 std::map<const void*, audio_stream>     next_audio_streams;\r
 171 \r
 172                 BOOST_FOREACH(auto& item, items_)\r
 173                 {                       \r
 174                         audio_buffer_ps next_audio;\r
 175 \r
 176                         auto next_transform = item.transform;\r
 177                         auto prev_transform = next_transform;\r
 178 \r
 179                         const auto it = audio_streams_.find(item.tag);\r
 180                         if(it != audio_streams_.end())\r
 181                         {       \r
 182                                 prev_transform  = it->second.prev_transform;\r
 183                                 next_audio              = std::move(it->second.audio_data);\r
 184                         }\r
 185 \r
 186                         if(prev_transform.volume < 0.001 && next_transform.volume < 0.001)\r
 187                                 continue;\r
 188                         \r
 189                         const float prev_volume = static_cast<float>(prev_transform.volume) * previous_master_volume_;\r
 190                         const float next_volume = static_cast<float>(next_transform.volume) * master_volume_;\r
 191                                                                         \r
 192                         auto alpha = (next_volume-prev_volume)/static_cast<float>(item.audio_data.size()/channel_layout_.num_channels);\r
 193                         \r
 194                         for(size_t n = 0; n < item.audio_data.size(); ++n)\r
 195                         {\r
 196                                 auto sample_multiplier = (prev_volume + (n/channel_layout_.num_channels) * alpha);\r
 197                                 next_audio.push_back(item.audio_data[n] * sample_multiplier);\r
 198                         }\r
 199                                                                                 \r
 200                         next_audio_streams[item.tag].prev_transform  = std::move(next_transform); // Store all active tags, inactive tags will be removed at the end.\r
 201                         next_audio_streams[item.tag].audio_data          = std::move(next_audio);                       \r
 202                 }\r
 203 \r
 204                 previous_master_volume_ = master_volume_;\r
 205                 items_.clear();\r
 206 \r
 207                 audio_streams_ = std::move(next_audio_streams);\r
 208                 \r
 209                 if(audio_streams_.empty())              \r
 210                         audio_streams_[nullptr].audio_data = audio_buffer_ps(audio_size(audio_cadence_.front()), 0.0f);\r
 211                                 \r
 212                 { // sanity check\r
 213 \r
 214                         auto nb_invalid_streams = boost::count_if(audio_streams_ | boost::adaptors::map_values, [&](const audio_stream& x)\r
 215                         {\r
 216                                 return x.audio_data.size() < audio_size(audio_cadence_.front());\r
 217                         });\r
 218 \r
 219                         if(nb_invalid_streams > 0)              \r
 220                                 CASPAR_LOG(trace) << "[audio_mixer] Incorrect frame audio cadence detected.";                   \r
 221                 }\r
 222 \r
 223                 std::vector<float> result_ps(audio_size(audio_cadence_.front()), 0.0f);\r
 224 \r
 225                 BOOST_FOREACH(auto& stream, audio_streams_ | boost::adaptors::map_values)\r
 226                 {\r
 227                         //CASPAR_LOG(debug) << stream.audio_data.size() << L" : " << result_ps.size();\r
 228 \r
 229                         if(stream.audio_data.size() < result_ps.size())\r
 230                         {\r
 231                                 stream.audio_data.resize(result_ps.size(), 0.0f);\r
 232                                 CASPAR_LOG(trace) << L"[audio_mixer] Appended zero samples";\r
 233                         }\r
 234 \r
 235                         auto out = boost::range::transform(result_ps, stream.audio_data, std::begin(result_ps), std::plus<float>());\r
 236                         stream.audio_data.erase(std::begin(stream.audio_data), std::begin(stream.audio_data) + std::distance(std::begin(result_ps), out));\r
 237                 }\r
 238                 \r
 239                 boost::range::rotate(audio_cadence_, std::begin(audio_cadence_)+1);\r
 240                 \r
 241 \r
 242                 audio_buffer result;\r
 243                 result.reserve(result_ps.size());\r
 244                 boost::range::transform(result_ps, std::back_inserter(result), [](float sample){return static_cast<int32_t>(sample);});         \r
 245                 \r
 246                 const int num_channels = channel_layout_.num_channels;\r
 247                 monitor_subject_ << monitor::message("/nb_channels") % num_channels;\r
 248 \r
 249                 auto max = std::vector<int32_t>(num_channels, std::numeric_limits<int32_t>::min());\r
 250 \r
 251                 for (size_t n = 0; n < result.size(); n += num_channels)\r
 252                         for (int ch = 0; ch < num_channels; ++ch)\r
 253                                 max[ch] = std::max(max[ch], std::abs(result[n + ch]));\r
 254                 \r
 255                 // Makes the dBFS of silence => -dynamic range of 32bit LPCM => about -192 dBFS\r
 256                 // Otherwise it would be -infinity\r
 257                 static const auto MIN_PFS = 0.5f / static_cast<float>(std::numeric_limits<int32_t>::max());\r
 258 \r
 259                 for (int i = 0; i < num_channels; ++i)\r
 260                 {\r
 261                         const auto pFS  = max[i] / static_cast<float>(std::numeric_limits<int32_t>::max());\r
 262                         const auto dBFS = 20.0f * std::log10(std::max(MIN_PFS, pFS));\r
 263                         \r
 264                         auto chan_str = boost::lexical_cast<std::string>(i + 1);\r
 265 \r
 266                         monitor_subject_ << monitor::message("/" + chan_str + "/pFS") % pFS;\r
 267                         monitor_subject_ << monitor::message("/" + chan_str + "/dBFS") % dBFS;\r
 268                 }\r
 269 \r
 270                 graph_->set_value("volume", static_cast<double>(*boost::max_element(max)) / std::numeric_limits<int32_t>::max());\r
 271 \r
 272                 return result;\r
 273         }\r
 274 \r
 275         size_t audio_size(size_t num_samples) const\r
 276         {\r
 277                 return num_samples * channel_layout_.num_channels;\r
 278         }\r
 279 \r
 280         void failed_rearrange(const void* tag, const channel_layout& layout)\r
 281         {\r
 282                 if (audio_streams_.find(tag) != audio_streams_.end())\r
 283                         return; // We don't want to flood the logs.\r
 284 \r
 285                 CASPAR_LOG(warning)\r
 286                                 << L"[audio_mixer] Could not satisfactory down/upmix from " \r
 287                                 << layout.name << L" to " << channel_layout_.name \r
 288                                 << L" because no mix config was found for " \r
 289                                 << layout.layout_type << L" => " << channel_layout_.layout_type \r
 290                                 << L". This might cause audio to be lost.";\r
 291         }\r
 292 };\r
 293 \r
 294 audio_mixer::audio_mixer(const safe_ptr<diagnostics::graph>& graph) : impl_(new implementation(graph)){}\r
 295 void audio_mixer::begin(core::basic_frame& frame){impl_->begin(frame);}\r
 296 void audio_mixer::visit(core::write_frame& frame){impl_->visit(frame);}\r
 297 void audio_mixer::end(){impl_->end();}\r
 298 float audio_mixer::get_master_volume() const { return impl_->get_master_volume(); }\r
 299 void audio_mixer::set_master_volume(float volume) { impl_->set_master_volume(volume); }\r
 300 audio_buffer audio_mixer::operator()(const video_format_desc& format_desc, const channel_layout& layout){return impl_->mix(format_desc, layout);}\r
 301 monitor::subject& audio_mixer::monitor_output(){return impl_->monitor_subject_;}\r
 302 \r
 303 }}