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

   1 /*\r
   2 * copyright (c) 2010 Sveriges Television AB <info@casparcg.com>\r
   3 *\r
   4 *  This file is part of CasparCG.\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 */\r
  20 #include "../../stdafx.h"\r
  21 \r
  22 #include "audio_mixer.h"\r
  23 \r
  24 #include <core/mixer/write_frame.h>\r
  25 #include <core/producer/frame/frame_transform.h>\r
  26 \r
  27 #include <tbb/cache_aligned_allocator.h>\r
  28 \r
  29 #include <boost/range.hpp>\r
  30 #include <boost/range/algorithm_ext/push_back.hpp>\r
  31 \r
  32 #include <array>\r
  33 #include <map>\r
  34 #include <stack>\r
  35 #include <vector>\r
  36 #include <valarray>\r
  37 \r
  38 namespace caspar { namespace core {\r
  39 \r
  40 struct audio_item\r
  41 {\r
  42         const void*                     tag;\r
  43         frame_transform         transform;\r
  44         audio_buffer            audio_data;\r
  45 \r
  46         audio_item()\r
  47         {\r
  48         }\r
  49 \r
  50         audio_item(audio_item&& other)\r
  51                 : tag(std::move(other.tag))\r
  52                 , transform(std::move(other.transform))\r
  53                 , audio_data(std::move(other.audio_data))\r
  54         {\r
  55         }\r
  56 };\r
  57 \r
  58 typedef std::vector<float, tbb::cache_aligned_allocator<float>> audio_buffer_ps;\r
  59         \r
  60 struct audio_stream\r
  61 {\r
  62         frame_transform         transform;\r
  63         audio_buffer_ps         audio_data;\r
  64 };\r
  65 \r
  66 struct audio_mixer::implementation\r
  67 {\r
  68         std::stack<core::frame_transform>       transform_stack_;\r
  69         std::map<const void*, audio_stream>     audio_streams_;\r
  70         std::vector<audio_item>                         items_;\r
  71         std::vector<size_t>                                     audio_cadence_;\r
  72         video_format_desc                                       format_desc_;\r
  73         \r
  74 public:\r
  75         implementation()\r
  76                 : format_desc_(video_format_desc::get(video_format::invalid))\r
  77         {\r
  78                 transform_stack_.push(core::frame_transform());\r
  79         }\r
  80         \r
  81         void begin(core::basic_frame& frame)\r
  82         {\r
  83                 transform_stack_.push(transform_stack_.top()*frame.get_frame_transform());\r
  84         }\r
  85 \r
  86         void visit(core::write_frame& frame)\r
  87         {\r
  88                 if(transform_stack_.top().volume < 0.002 || frame.audio_data().empty())\r
  89                         return;\r
  90 \r
  91                 audio_item item;\r
  92                 item.tag                = frame.tag();\r
  93                 item.transform  = transform_stack_.top();\r
  94                 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
  95                 \r
  96                 items_.push_back(std::move(item));              \r
  97         }\r
  98 \r
  99         void begin(const core::frame_transform& transform)\r
 100         {\r
 101                 transform_stack_.push(transform_stack_.top()*transform);\r
 102         }\r
 103                 \r
 104         void end()\r
 105         {\r
 106                 transform_stack_.pop();\r
 107         }\r
 108         \r
 109         audio_buffer mix(const video_format_desc& format_desc)\r
 110         {       \r
 111                 if(format_desc_ != format_desc)\r
 112                 {\r
 113                         audio_streams_.clear();\r
 114                         audio_cadence_ = format_desc.audio_cadence;\r
 115                         format_desc_ = format_desc;\r
 116                 }               \r
 117                 \r
 118                 std::map<const void*, audio_stream>     next_audio_streams_;\r
 119 \r
 120                 BOOST_FOREACH(auto& item, items_)\r
 121                 {                       \r
 122                         audio_buffer_ps next_audio;\r
 123 \r
 124                         auto next_transform = item.transform;\r
 125                         auto prev_transform = next_transform;\r
 126 \r
 127                         const auto it = audio_streams_.find(item.tag);\r
 128                         if(it != audio_streams_.end())\r
 129                         {       \r
 130                                 prev_transform  = it->second.transform;\r
 131                                 next_audio              = std::move(it->second.audio_data);\r
 132                         }\r
 133 \r
 134                         if(prev_transform.volume < 0.001 && next_transform.volume < 0.001)\r
 135                                 continue;\r
 136                         \r
 137                         const float prev_volume = static_cast<float>(prev_transform.volume);\r
 138                         const float next_volume = static_cast<float>(next_transform.volume);\r
 139                                                                         \r
 140                         auto alpha              = (next_volume-prev_volume)/static_cast<float>(item.audio_data.size()/format_desc.audio_channels);\r
 141                         \r
 142                         audio_buffer_ps result;\r
 143                         result.reserve(item.audio_data.size());\r
 144                         for(size_t n = 0; n < item.audio_data.size()/2; ++n)\r
 145                         {\r
 146                                 result.push_back(item.audio_data[n*2+0] * (prev_volume + n * alpha));\r
 147                                 result.push_back(item.audio_data[n*2+1] * (prev_volume + n * alpha));\r
 148                         }\r
 149                                                 \r
 150                         boost::range::push_back(next_audio, std::move(result));\r
 151                                 \r
 152                         next_audio_streams_[item.tag].transform  = std::move(next_transform); // Store all active tags, inactive tags will be removed at the end.\r
 153                         next_audio_streams_[item.tag].audio_data = std::move(next_audio);                       \r
 154                 }                               \r
 155                 items_.clear();\r
 156 \r
 157                 audio_streams_  = std::move(next_audio_streams_);\r
 158                 next_audio_streams_.clear();\r
 159                 \r
 160                 if(audio_streams_.empty())              \r
 161                         audio_streams_[nullptr].audio_data = audio_buffer_ps(audio_cadence_.front(), 0.0f);\r
 162                                 \r
 163                 std::vector<float> result_ps(audio_cadence_.front(), 0.0f);\r
 164                 BOOST_FOREACH(auto& stream, audio_streams_)\r
 165                 {\r
 166                         auto& audio_data        = stream.second.audio_data;\r
 167                         \r
 168                         if(audio_data.size() < audio_cadence_.front())\r
 169                         {\r
 170                                 CASPAR_LOG(warning) << "[audio_mixer] Incorrect frame audio cadence, prepending silence: " << audio_cadence_.front()-audio_data.size();\r
 171                                 boost::range::push_back(audio_data, audio_buffer_ps(audio_cadence_.front()-audio_data.size(), 0.0f));\r
 172                         }\r
 173 \r
 174                         for(size_t n = 0; n < audio_cadence_.front(); ++n)\r
 175                                 result_ps[n] += audio_data[n];\r
 176 \r
 177                         audio_data.erase(std::begin(audio_data), std::begin(audio_data) + audio_cadence_.front());\r
 178                 }               \r
 179                 \r
 180                 boost::range::rotate(audio_cadence_, std::begin(audio_cadence_)+1);\r
 181                 \r
 182                 audio_buffer result;\r
 183                 result.reserve(result_ps.size());\r
 184                 boost::range::transform(result_ps, std::back_inserter(result), [](float sample){return static_cast<int32_t>(sample);});                                         \r
 185                 return result;\r
 186         }\r
 187 };\r
 188 \r
 189 audio_mixer::audio_mixer() : impl_(new implementation()){}\r
 190 void audio_mixer::begin(core::basic_frame& frame){impl_->begin(frame);}\r
 191 void audio_mixer::visit(core::write_frame& frame){impl_->visit(frame);}\r
 192 void audio_mixer::end(){impl_->end();}\r
 193 audio_buffer audio_mixer::mix(const video_format_desc& format_desc){return impl_->mix(format_desc);}\r
 194 audio_mixer& audio_mixer::operator=(audio_mixer&& other)\r
 195 {\r
 196         impl_ = std::move(other.impl_);\r
 197         return *this;\r
 198 }\r
 199 \r
 200 }}