2 * Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>
4 * This file is part of CasparCG (www.casparcg.com).
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.
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.
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/>.
19 * Author: Robert Nagy, ronag89@gmail.com
22 #include "../../StdAfx.h"
24 #include "audio_mixer.h"
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>
31 #include <common/diagnostics/graph.h>
32 #include <common/linq.h>
34 #include <boost/range/adaptors.hpp>
35 #include <boost/range/distance.hpp>
36 #include <boost/lexical_cast.hpp>
42 namespace caspar { namespace core {
46 const void* tag = nullptr;
47 audio_transform transform;
48 audio_buffer audio_data;
49 audio_channel_layout channel_layout = audio_channel_layout::invalid();
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))
64 typedef cache_aligned_vector<double> audio_buffer_ps;
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;
75 struct audio_mixer::impl : boost::noncopyable
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_;
88 impl(spl::shared_ptr<diagnostics::graph> graph)
89 : graph_(std::move(graph))
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());
96 void push(const frame_transform& transform)
98 transform_stack_.push(transform_stack_.top()*transform.audio_transform);
101 void visit(const const_frame& frame)
103 if(transform_stack_.top().volume < 0.002 || frame.audio_data().empty())
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();
112 if(item.transform.is_still)
113 item.transform.volume = 0.0;
115 items_.push_back(std::move(item));
118 void begin(const core::audio_transform& transform)
120 transform_stack_.push(transform_stack_.top()*transform);
125 transform_stack_.pop();
128 void set_master_volume(float volume)
130 master_volume_ = volume;
133 float get_master_volume()
135 return master_volume_;
138 audio_buffer mix(const video_format_desc& format_desc, const audio_channel_layout& channel_layout)
140 if(format_desc_ != format_desc || channel_layout_ != channel_layout)
142 audio_streams_.clear();
143 audio_cadence_ = format_desc.audio_cadence;
144 format_desc_ = format_desc;
145 channel_layout_ = channel_layout;
148 std::map<const void*, audio_stream> next_audio_streams;
149 std::vector<const void*> used_tags;
151 for (auto& item : items_)
153 audio_buffer_ps next_audio;
154 std::unique_ptr<audio_channel_remapper> channel_remapper;
155 bool remapping_failed = false;
157 auto next_transform = item.transform;
158 auto prev_transform = next_transform;
162 if(boost::range::find(used_tags, tag) != used_tags.end())
165 used_tags.push_back(tag);
167 const auto it = audio_streams_.find(tag);
168 if (it != audio_streams_.end())
170 prev_transform = it->second.prev_transform;
171 next_audio = std::move(it->second.audio_data);
172 channel_remapper = std::move(it->second.channel_remapper);
173 remapping_failed = it->second.remapping_failed;
176 if (remapping_failed)
178 CASPAR_LOG(trace) << "[audio_mixer] audio channel remapping already failed for stream.";
179 next_audio_streams[tag].remapping_failed = true;
183 // Skip it if there is no existing audio stream and item has no audio-data.
184 if(it == audio_streams_.end() && item.audio_data.empty())
187 if (item.channel_layout == audio_channel_layout::invalid())
189 CASPAR_LOG(debug) << "[audio_mixer] invalid audio channel layout for item";
193 if (!channel_remapper)
197 channel_remapper.reset(new audio_channel_remapper(item.channel_layout, channel_layout_));
201 CASPAR_LOG_CURRENT_EXCEPTION();
202 CASPAR_LOG(error) << "[audio_mixer] audio channel remapping failed for stream.";
203 next_audio_streams[tag].remapping_failed = true;
208 item.audio_data = channel_remapper->mix_and_rearrange(item.audio_data);
210 const double prev_volume = prev_transform.volume * previous_master_volume_;
211 const double next_volume = next_transform.volume * master_volume_;
213 // TODO: Move volume mixing into code below, in order to support audio sample counts not corresponding to frame audio samples.
214 auto alpha = (next_volume-prev_volume)/static_cast<double>(item.audio_data.size()/channel_layout_.num_channels);
216 for(size_t n = 0; n < item.audio_data.size(); ++n)
218 auto sample_multiplier = (prev_volume + (n / channel_layout_.num_channels) * alpha);
219 next_audio.push_back(item.audio_data.data()[n] * sample_multiplier);
222 next_audio_streams[tag].prev_transform = std::move(next_transform); // Store all active tags, inactive tags will be removed at the end.
223 next_audio_streams[tag].audio_data = std::move(next_audio);
224 next_audio_streams[tag].channel_remapper = std::move(channel_remapper);
225 next_audio_streams[tag].remapping_failed = remapping_failed;
226 next_audio_streams[tag].is_still = item.transform.is_still;
229 previous_master_volume_ = master_volume_;
232 audio_streams_ = std::move(next_audio_streams);
234 if(audio_streams_.empty())
235 audio_streams_[nullptr].audio_data = audio_buffer_ps(audio_size(audio_cadence_.front()), 0.0);
239 auto nb_invalid_streams = cpplinq::from(audio_streams_)
241 .where([&](const audio_stream& x)
243 return !x.remapping_failed && x.audio_data.size() < audio_size(audio_cadence_.front());
247 if(nb_invalid_streams > 0)
248 CASPAR_LOG(trace) << "[audio_mixer] Incorrect frame audio cadence detected.";
251 audio_buffer_ps result_ps(audio_size(audio_cadence_.front()), 0.0);
252 for (auto& stream : audio_streams_ | boost::adaptors::map_values)
254 if(stream.audio_data.size() < result_ps.size())
255 stream.audio_data.resize(result_ps.size(), 0.0f);
257 auto out = boost::range::transform(result_ps, stream.audio_data, std::begin(result_ps), std::plus<double>());
258 stream.audio_data.erase(std::begin(stream.audio_data), std::begin(stream.audio_data) + std::distance(std::begin(result_ps), out));
261 boost::range::rotate(audio_cadence_, std::begin(audio_cadence_)+1);
263 auto result_owner = spl::make_shared<mutable_audio_buffer>();
264 auto& result = *result_owner;
265 result.reserve(result_ps.size());
266 const int32_t min_amplitude = std::numeric_limits<int32_t>::min();
267 const int32_t max_amplitude = std::numeric_limits<int32_t>::max();
268 bool clipping = false;
269 boost::range::transform(result_ps, std::back_inserter(result), [&](double sample)
271 if (sample > max_amplitude)
274 return max_amplitude;
276 else if (sample < min_amplitude)
279 return min_amplitude;
282 return static_cast<int32_t>(sample);
286 graph_->set_tag(diagnostics::tag_severity::WARNING, "audio-clipping");
288 const int num_channels = channel_layout_.num_channels;
289 monitor_subject_ << monitor::message("/nb_channels") % num_channels;
291 auto max = std::vector<int32_t>(num_channels, std::numeric_limits<int32_t>::min());
293 for (size_t n = 0; n < result.size(); n += num_channels)
294 for (int ch = 0; ch < num_channels; ++ch)
295 max[ch] = std::max(max[ch], std::abs(result[n + ch]));
297 // Makes the dBFS of silence => -dynamic range of 32bit LPCM => about -192 dBFS
298 // Otherwise it would be -infinity
299 static const auto MIN_PFS = 0.5f / static_cast<float>(std::numeric_limits<int32_t>::max());
301 for (int i = 0; i < num_channels; ++i)
303 const auto pFS = max[i] / static_cast<float>(std::numeric_limits<int32_t>::max());
304 const auto dBFS = 20.0f * std::log10(std::max(MIN_PFS, pFS));
306 auto chan_str = boost::lexical_cast<std::string>(i + 1);
308 monitor_subject_ << monitor::message("/" + chan_str + "/pFS") % pFS;
309 monitor_subject_ << monitor::message("/" + chan_str + "/dBFS") % dBFS;
312 graph_->set_value("volume", static_cast<double>(*boost::max_element(max)) / std::numeric_limits<int32_t>::max());
314 return caspar::array<int32_t>(result.data(), result.size(), true, std::move(result_owner));
317 size_t audio_size(size_t num_samples) const
319 return num_samples * channel_layout_.num_channels;
323 audio_mixer::audio_mixer(spl::shared_ptr<diagnostics::graph> graph) : impl_(new impl(std::move(graph))){}
324 void audio_mixer::push(const frame_transform& transform){impl_->push(transform);}
325 void audio_mixer::visit(const const_frame& frame){impl_->visit(frame);}
326 void audio_mixer::pop(){impl_->pop();}
327 void audio_mixer::set_master_volume(float volume) { impl_->set_master_volume(volume); }
328 float audio_mixer::get_master_volume() { return impl_->get_master_volume(); }
329 audio_buffer audio_mixer::operator()(const video_format_desc& format_desc, const audio_channel_layout& channel_layout){ return impl_->mix(format_desc, channel_layout); }
330 monitor::subject& audio_mixer::monitor_output(){ return impl_->monitor_subject_; }