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

   1 /*\r
   2 * Copyright (c) 2011 Sveriges Television AB <info@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 <common/diagnostics/graph.h>\r
  29 \r
  30 #include <boost/range/adaptors.hpp>\r
  31 #include <boost/range/distance.hpp>\r
  32 \r
  33 #include <map>\r
  34 #include <stack>\r
  35 #include <vector>\r
  36 \r
  37 namespace caspar { namespace core {\r
  38 \r
  39 struct audio_item\r
  40 {\r
  41         const void*                     tag;\r
  42         frame_transform         transform;\r
  43         audio_buffer            audio_data;\r
  44 \r
  45         audio_item()\r
  46         {\r
  47         }\r
  48 \r
  49         audio_item(audio_item&& other)\r
  50                 : tag(std::move(other.tag))\r
  51                 , transform(std::move(other.transform))\r
  52                 , audio_data(std::move(other.audio_data))\r
  53         {\r
  54         }\r
  55 };\r
  56 \r
  57 typedef std::vector<float, tbb::cache_aligned_allocator<float>> audio_buffer_ps;\r
  58         \r
  59 struct audio_stream\r
  60 {\r
  61         frame_transform         prev_transform;\r
  62         audio_buffer_ps         audio_data;\r
  63 };\r
  64 \r
  65 struct audio_mixer::impl\r
  66 {\r
  67         safe_ptr<diagnostics::graph>            graph_;\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<int>                                        audio_cadence_;\r
  72         video_format_desc                                       format_desc_;\r
  73         \r
  74 public:\r
  75         impl(const safe_ptr<diagnostics::graph>& graph)\r
  76                 : graph_(graph)\r
  77         {\r
  78                 graph_->set_color("volume", diagnostics::color(1.0f, 0.8f, 0.1f));\r
  79                 transform_stack_.push(core::frame_transform());\r
  80         }\r
  81         \r
  82         void begin(core::basic_frame& frame)\r
  83         {\r
  84                 transform_stack_.push(transform_stack_.top()*frame.get_frame_transform());\r
  85         }\r
  86 \r
  87         void visit(core::write_frame& frame)\r
  88         {\r
  89                 if(transform_stack_.top().volume < 0.002 || frame.audio_data().empty())\r
  90                         return;\r
  91 \r
  92                 audio_item item;\r
  93                 item.tag                = frame.tag();\r
  94                 item.transform  = transform_stack_.top();\r
  95                 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
  96                 \r
  97                 items_.push_back(std::move(item));              \r
  98         }\r
  99 \r
 100         void begin(const core::frame_transform& transform)\r
 101         {\r
 102                 transform_stack_.push(transform_stack_.top()*transform);\r
 103         }\r
 104                 \r
 105         void end()\r
 106         {\r
 107                 transform_stack_.pop();\r
 108         }\r
 109         \r
 110         audio_buffer mix(const video_format_desc& format_desc)\r
 111         {       \r
 112                 if(format_desc_ != format_desc)\r
 113                 {\r
 114                         audio_streams_.clear();\r
 115                         audio_cadence_ = format_desc.audio_cadence;\r
 116                         format_desc_ = format_desc;\r
 117                 }               \r
 118                 \r
 119                 std::map<const void*, audio_stream>     next_audio_streams;\r
 120 \r
 121                 BOOST_FOREACH(auto& item, items_)\r
 122                 {                       \r
 123                         audio_buffer_ps next_audio;\r
 124 \r
 125                         auto next_transform = item.transform;\r
 126                         auto prev_transform = next_transform;\r
 127 \r
 128                         const auto it = audio_streams_.find(item.tag);\r
 129                         if(it != audio_streams_.end())\r
 130                         {       \r
 131                                 prev_transform  = it->second.prev_transform;\r
 132                                 next_audio              = std::move(it->second.audio_data);\r
 133                         }\r
 134                                                 \r
 135                         const float prev_volume = static_cast<float>(prev_transform.volume);\r
 136                         const float next_volume = static_cast<float>(next_transform.volume);\r
 137                                                                         \r
 138                         auto alpha = (next_volume-prev_volume)/static_cast<float>(item.audio_data.size()/format_desc.audio_channels);\r
 139                         \r
 140                         for(size_t n = 0; n < item.audio_data.size(); ++n)\r
 141                                 next_audio.push_back(item.audio_data[n] * (prev_volume + (n/format_desc_.audio_channels) * alpha));\r
 142                                                                                 \r
 143                         next_audio_streams[item.tag].prev_transform  = std::move(next_transform); // Store all active tags, inactive tags will be removed at the end.\r
 144                         next_audio_streams[item.tag].audio_data          = std::move(next_audio);                       \r
 145                 }                               \r
 146 \r
 147                 items_.clear();\r
 148 \r
 149                 audio_streams_ = std::move(next_audio_streams);\r
 150                 \r
 151                 if(audio_streams_.empty())              \r
 152                         audio_streams_[nullptr].audio_data = audio_buffer_ps(audio_cadence_.front(), 0.0f);\r
 153                                 \r
 154                 { // sanity check\r
 155 \r
 156                         auto nb_invalid_streams = boost::count_if(audio_streams_ | boost::adaptors::map_values, [&](const audio_stream& x)\r
 157                         {\r
 158                                 return x.audio_data.size() < audio_cadence_.front();\r
 159                         });\r
 160 \r
 161                         if(nb_invalid_streams > 0)              \r
 162                                 CASPAR_LOG(trace) << "[audio_mixer] Incorrect frame audio cadence detected.";                   \r
 163                 }\r
 164 \r
 165                 std::vector<float> result_ps(audio_cadence_.front(), 0.0f);\r
 166                 BOOST_FOREACH(auto& stream, audio_streams_ | boost::adaptors::map_values)\r
 167                 {\r
 168                         //CASPAR_LOG(debug) << stream.audio_data.size() << L" : " << result_ps.size();\r
 169 \r
 170                         if(stream.audio_data.size() < result_ps.size())\r
 171                         {\r
 172                                 stream.audio_data.resize(result_ps.size(), 0.0f);\r
 173                                 CASPAR_LOG(trace) << L"[audio_mixer] Appended zero samples";\r
 174                         }\r
 175 \r
 176                         auto out = boost::range::transform(result_ps, stream.audio_data, std::begin(result_ps), std::plus<float>());\r
 177                         stream.audio_data.erase(std::begin(stream.audio_data), std::begin(stream.audio_data) + std::distance(std::begin(result_ps), out));\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 \r
 186                 auto max = boost::range::max_element(result);\r
 187 \r
 188                 graph_->set_value("volume", static_cast<double>(std::abs(*max))/std::numeric_limits<int32_t>::max());\r
 189 \r
 190                 return result;\r
 191         }\r
 192 };\r
 193 \r
 194 audio_mixer::audio_mixer(const safe_ptr<diagnostics::graph>& graph) : impl_(new impl(graph)){}\r
 195 void audio_mixer::begin(core::basic_frame& frame){impl_->begin(frame);}\r
 196 void audio_mixer::visit(core::write_frame& frame){impl_->visit(frame);}\r
 197 void audio_mixer::end(){impl_->end();}\r
 198 audio_buffer audio_mixer::operator()(const video_format_desc& format_desc){return impl_->mix(format_desc);}\r
 199 \r
 200 }}