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/frame/audio_channel_layout.h>
  29 #include <core/monitor/monitor.h>
  30
  31 #include <common/diagnostics/graph.h>
  32 #include <common/linq.h>
  33
  34 #include <boost/range/adaptors.hpp>
  35 #include <boost/range/distance.hpp>
  36 #include <boost/lexical_cast.hpp>
  37
  38 #include <map>
  39 #include <stack>
  40 #include <vector>
  41
  42 namespace caspar { namespace core {
  43
  44 struct audio_item
  45 {
  46         const void*                             tag                             = nullptr;
  47         audio_transform                 transform;
  48         audio_buffer                    audio_data;
  49         audio_channel_layout    channel_layout  = audio_channel_layout::invalid();
  50
  51         audio_item()
  52         {
  53         }
  54
  55         audio_item(audio_item&& other)
  56                 : tag(std::move(other.tag))
  57                 , transform(std::move(other.transform))
  58                 , audio_data(std::move(other.audio_data))
  59                 , channel_layout(std::move(other.channel_layout))
  60         {
  61         }
  62 };
  63
  64 typedef cache_aligned_vector<double> audio_buffer_ps;
  65
  66 struct audio_stream
  67 {
  68         audio_transform                                                 prev_transform;
  69         audio_buffer_ps                                                 audio_data;
  70         std::unique_ptr<audio_channel_remapper> channel_remapper;
  71         bool                                                                    remapping_failed        = false;
  72         bool                                                                    is_still                        = false;
  73 };
  74
  75 struct audio_mixer::impl : boost::noncopyable
  76 {
  77         monitor::subject                                        monitor_subject_                { "/audio" };
  78         std::stack<core::audio_transform>       transform_stack_;
  79         std::map<const void*, audio_stream>     audio_streams_;
  80         std::vector<audio_item>                         items_;
  81         std::vector<int>                                        audio_cadence_;
  82         video_format_desc                                       format_desc_;
  83         audio_channel_layout                            channel_layout_                 = audio_channel_layout::invalid();
  84         float                                                           master_volume_                  = 1.0f;
  85         float                                                           previous_master_volume_ = master_volume_;
  86         spl::shared_ptr<diagnostics::graph>     graph_;
  87 public:
  88         impl(spl::shared_ptr<diagnostics::graph> graph)
  89                 : graph_(std::move(graph))
  90         {
  91                 graph_->set_color("volume", diagnostics::color(1.0f, 0.8f, 0.1f));
  92                 graph_->set_color("audio-clipping", diagnostics::color(0.3f, 0.6f, 0.3f));
  93                 transform_stack_.push(core::audio_transform());
  94         }
  95
  96         void push(const frame_transform& transform)
  97         {
  98                 transform_stack_.push(transform_stack_.top()*transform.audio_transform);
  99         }
 100
 101         void visit(const const_frame& frame)
 102         {
 103                 if(transform_stack_.top().volume < 0.002 || frame.audio_data().empty())
 104                         return;
 105
 106                 audio_item item;
 107                 item.tag                        = frame.stream_tag();
 108                 item.transform          = transform_stack_.top();
 109                 item.audio_data         = frame.audio_data();
 110                 item.channel_layout = frame.audio_channel_layout();
 111
 112                 if(item.transform.is_still)
 113                         item.transform.volume = 0.0;
 114
 115                 items_.push_back(std::move(item));
 116         }
 117
 118         void pop()
 119         {
 120                 transform_stack_.pop();
 121         }
 122
 123         void set_master_volume(float volume)
 124         {
 125                 master_volume_ = volume;
 126         }
 127
 128         float get_master_volume()
 129         {
 130                 return master_volume_;
 131         }
 132
 133         audio_buffer mix(const video_format_desc& format_desc, const audio_channel_layout& channel_layout)
 134         {
 135                 if(format_desc_ != format_desc || channel_layout_ != channel_layout)
 136                 {
 137                         audio_streams_.clear();
 138                         audio_cadence_ = format_desc.audio_cadence;
 139                         format_desc_ = format_desc;
 140                         channel_layout_ = channel_layout;
 141                 }
 142
 143                 std::map<const void*, audio_stream>     next_audio_streams;
 144
 145                 for (auto& item : items_)
 146                 {
 147                         audio_buffer_ps next_audio;
 148                         std::unique_ptr<audio_channel_remapper> channel_remapper;
 149                         bool remapping_failed = false;
 150
 151                         auto next_transform = item.transform;
 152                         auto prev_transform = next_transform;
 153
 154                         auto tag = item.tag;
 155
 156                         auto it = next_audio_streams.find(tag);
 157                         bool found = it != next_audio_streams.end();
 158
 159                         if (!found)
 160                         {
 161                                 it = audio_streams_.find(tag);
 162                                 found = it != audio_streams_.end();
 163                         }
 164
 165                         if (found)
 166                         {
 167                                 prev_transform = it->second.prev_transform;
 168                                 next_audio = std::move(it->second.audio_data);
 169                                 channel_remapper = std::move(it->second.channel_remapper);
 170                                 remapping_failed = it->second.remapping_failed;
 171                         }
 172
 173                         if (remapping_failed)
 174                         {
 175                                 CASPAR_LOG(trace) << "[audio_mixer] audio channel remapping already failed for stream.";
 176                                 next_audio_streams[tag].remapping_failed = true;
 177                                 continue;
 178                         }
 179
 180                         // Skip it if there is no existing audio stream and item has no audio-data.
 181                         if(!found && item.audio_data.empty())
 182                                 continue;
 183
 184                         if (item.channel_layout == audio_channel_layout::invalid())
 185                         {
 186                                 CASPAR_LOG(warning) << "[audio_mixer] invalid audio channel layout for item";
 187                                 next_audio_streams[tag].remapping_failed = true;
 188                                 continue;
 189                         }
 190
 191                         if (!channel_remapper)
 192                         {
 193                                 try
 194                                 {
 195                                         channel_remapper.reset(new audio_channel_remapper(item.channel_layout, channel_layout_));
 196                                 }
 197                                 catch (...)
 198                                 {
 199                                         CASPAR_LOG_CURRENT_EXCEPTION();
 200                                         CASPAR_LOG(error) << "[audio_mixer] audio channel remapping failed for stream.";
 201                                         next_audio_streams[tag].remapping_failed = true;
 202                                         continue;
 203                                 }
 204                         }
 205
 206                         item.audio_data = channel_remapper->mix_and_rearrange(item.audio_data);
 207
 208                         const double prev_volume = prev_transform.volume * previous_master_volume_;
 209                         const double next_volume = next_transform.volume * master_volume_;
 210
 211                         // TODO: Move volume mixing into code below, in order to support audio sample counts not corresponding to frame audio samples.
 212                         auto alpha = (next_volume-prev_volume)/static_cast<double>(item.audio_data.size()/channel_layout_.num_channels);
 213
 214                         for(size_t n = 0; n < item.audio_data.size(); ++n)
 215                         {
 216                                 auto sample_multiplier = (prev_volume + (n / channel_layout_.num_channels) * alpha);
 217                                 next_audio.push_back(item.audio_data.data()[n] * sample_multiplier);
 218                         }
 219
 220                         next_audio_streams[tag].prev_transform          = std::move(next_transform); // Store all active tags, inactive tags will be removed at the end.
 221                         next_audio_streams[tag].audio_data                      = std::move(next_audio);
 222                         next_audio_streams[tag].channel_remapper        = std::move(channel_remapper);
 223                         next_audio_streams[tag].remapping_failed        = remapping_failed;
 224                         next_audio_streams[tag].is_still                        = item.transform.is_still;
 225                 }
 226
 227                 previous_master_volume_ = master_volume_;
 228                 items_.clear();
 229
 230                 audio_streams_ = std::move(next_audio_streams);
 231
 232                 if(audio_streams_.empty())
 233                         audio_streams_[nullptr].audio_data = audio_buffer_ps(audio_size(audio_cadence_.front()), 0.0);
 234
 235                 { // sanity check
 236
 237                         auto nb_invalid_streams = cpplinq::from(audio_streams_)
 238                                 .select(values())
 239                                 .where([&](const audio_stream& x)
 240                                 {
 241                                         return !x.remapping_failed && x.audio_data.size() < audio_size(audio_cadence_.front());
 242                                 })
 243                                 .count();
 244
 245                         if(nb_invalid_streams > 0)
 246                                 CASPAR_LOG(trace) << "[audio_mixer] Incorrect frame audio cadence detected.";
 247                 }
 248
 249                 audio_buffer_ps result_ps(audio_size(audio_cadence_.front()), 0.0);
 250                 for (auto& stream : audio_streams_ | boost::adaptors::map_values)
 251                 {
 252                         if (stream.audio_data.size() < result_ps.size())
 253                         {
 254                                 auto samples = (result_ps.size() - stream.audio_data.size()) / channel_layout_.num_channels;
 255                                 CASPAR_LOG(trace) << L"[audio_mixer] Appended " << samples << L" zero samples";
 256                                 CASPAR_LOG(trace) << L"[audio_mixer] Actual number of samples " << stream.audio_data.size() / channel_layout_.num_channels;
 257                                 CASPAR_LOG(trace) << L"[audio_mixer] Wanted number of samples " << result_ps.size() / channel_layout_.num_channels;
 258                                 stream.audio_data.resize(result_ps.size(), 0.0);
 259                         }
 260
 261                         auto out = boost::range::transform(result_ps, stream.audio_data, std::begin(result_ps), std::plus<double>());
 262                         stream.audio_data.erase(std::begin(stream.audio_data), std::begin(stream.audio_data) + std::distance(std::begin(result_ps), out));
 263                 }
 264
 265                 boost::range::rotate(audio_cadence_, std::begin(audio_cadence_)+1);
 266
 267                 auto result_owner = spl::make_shared<mutable_audio_buffer>();
 268                 auto& result = *result_owner;
 269                 result.reserve(result_ps.size());
 270                 const int32_t min_amplitude = std::numeric_limits<int32_t>::min();
 271                 const int32_t max_amplitude = std::numeric_limits<int32_t>::max();
 272                 bool clipping = false;
 273                 boost::range::transform(result_ps, std::back_inserter(result), [&](double sample)
 274                 {
 275                         if (sample > max_amplitude)
 276                         {
 277                                 clipping = true;
 278                                 return max_amplitude;
 279                         }
 280                         else if (sample < min_amplitude)
 281                         {
 282                                 clipping = true;
 283                                 return min_amplitude;
 284                         }
 285                         else
 286                                 return static_cast<int32_t>(sample);
 287                 });
 288
 289                 if (clipping)
 290                         graph_->set_tag(diagnostics::tag_severity::WARNING, "audio-clipping");
 291
 292                 const int num_channels = channel_layout_.num_channels;
 293                 monitor_subject_ << monitor::message("/nb_channels") % num_channels;
 294
 295                 auto max = std::vector<int32_t>(num_channels, std::numeric_limits<int32_t>::min());
 296
 297                 for (size_t n = 0; n < result.size(); n += num_channels)
 298                         for (int ch = 0; ch < num_channels; ++ch)
 299                                 max[ch] = std::max(max[ch], std::abs(result[n + ch]));
 300
 301                 // Makes the dBFS of silence => -dynamic range of 32bit LPCM => about -192 dBFS
 302                 // Otherwise it would be -infinity
 303                 static const auto MIN_PFS = 0.5f / static_cast<float>(std::numeric_limits<int32_t>::max());
 304
 305                 for (int i = 0; i < num_channels; ++i)
 306                 {
 307                         const auto pFS = max[i] / static_cast<float>(std::numeric_limits<int32_t>::max());
 308                         const auto dBFS = 20.0f * std::log10(std::max(MIN_PFS, pFS));
 309
 310                         auto chan_str = boost::lexical_cast<std::string>(i + 1);
 311
 312                         monitor_subject_ << monitor::message("/" + chan_str + "/pFS") % pFS;
 313                         monitor_subject_ << monitor::message("/" + chan_str + "/dBFS") % dBFS;
 314                 }
 315
 316                 graph_->set_value("volume", static_cast<double>(*boost::max_element(max)) / std::numeric_limits<int32_t>::max());
 317
 318                 return caspar::array<int32_t>(result.data(), result.size(), true, std::move(result_owner));
 319         }
 320
 321         size_t audio_size(size_t num_samples) const
 322         {
 323                 return num_samples * channel_layout_.num_channels;
 324         }
 325 };
 326
 327 audio_mixer::audio_mixer(spl::shared_ptr<diagnostics::graph> graph) : impl_(new impl(std::move(graph))){}
 328 void audio_mixer::push(const frame_transform& transform){impl_->push(transform);}
 329 void audio_mixer::visit(const const_frame& frame){impl_->visit(frame);}
 330 void audio_mixer::pop(){impl_->pop();}
 331 void audio_mixer::set_master_volume(float volume) { impl_->set_master_volume(volume); }
 332 float audio_mixer::get_master_volume() { return impl_->get_master_volume(); }
 333 audio_buffer audio_mixer::operator()(const video_format_desc& format_desc, const audio_channel_layout& channel_layout){ return impl_->mix(format_desc, channel_layout); }
 334 monitor::subject& audio_mixer::monitor_output(){ return impl_->monitor_subject_; }
 335
 336 }}