2 #define _AUDIO_MIXER_H 1
4 // The audio mixer, dealing with extracting the right signals from
5 // each capture card, resampling signals so that they are in sync,
6 // processing them with effects (if desired), and then mixing them
7 // all together into one final audio signal.
9 // All operations on AudioMixer, except destruction and set_delay_analyzer(),
14 #include <zita-resampler/resampler.h>
25 #include "alsa_pool.h"
26 #include "correlation_measurer.h"
29 #include "ebu_r128_proc.h"
31 #include "input_mapping.h"
32 #include "resampling_queue.h"
33 #include "stereocompressor.h"
35 class DelayAnalyzerInterface;
36 class DeviceSpecProto;
42 // Convert the given audio from {16,24,32}-bit M-channel to 32-bit N-channel PCM.
43 // Assumes little-endian and chunky, signed PCM throughout.
44 std::vector<int32_t> convert_audio_to_fixed32(const uint8_t *data, unsigned num_samples, bmusb::AudioFormat audio_format, unsigned num_destination_channels);
46 // Similar, except converts to floating-point instead, and converts only one channel.
47 void convert_audio_to_fp32(float *dst, size_t out_channel, size_t out_num_channels,
48 const uint8_t *src, size_t in_channel, bmusb::AudioFormat in_audio_format,
60 AudioMixer(unsigned num_capture_cards, unsigned num_ffmpeg_inputs);
61 void reset_resampler(DeviceSpec device_spec);
64 // Add audio (or silence) to the given device's queue. Can return false if
65 // the lock wasn't successfully taken; if so, you should simply try again.
66 // (This is to avoid a deadlock where a card hangs on the mutex in add_audio()
67 // while we are trying to shut it down from another thread that also holds
69 bool add_audio(DeviceSpec device_spec, const uint8_t *data, unsigned num_samples, bmusb::AudioFormat audio_format, std::chrono::steady_clock::time_point frame_time);
70 bool add_silence(DeviceSpec device_spec, unsigned samples_per_frame, unsigned num_frames);
72 // If a given device is offline for whatever reason and cannot deliver audio
73 // (by means of add_audio() or add_silence()), you can call put it in silence mode,
74 // where it will be taken to only output silence. Note that when taking it _out_
75 // of silence mode, the resampler will be reset, so that old audio will not
76 // affect it. Same true/false behavior as add_audio().
77 bool silence_card(DeviceSpec device_spec, bool silence);
79 std::vector<float> get_output(std::chrono::steady_clock::time_point ts, unsigned num_samples, ResamplingQueue::RateAdjustmentPolicy rate_adjustment_policy);
81 float get_fader_volume(unsigned bus_index) const { return fader_volume_db[bus_index]; }
82 void set_fader_volume(unsigned bus_index, float level_db) { fader_volume_db[bus_index] = level_db; }
84 bool get_mute(unsigned bus_index) const { return mute[bus_index]; }
85 void set_mute(unsigned bus_index, bool muted) { mute[bus_index] = muted; }
90 // Note: Holds all ALSA devices (see ALSAPool::get_devices()).
91 // You will need to call set_input_mapping() to get the hold state correctly,
92 // or every card will be held forever.
95 std::map<DeviceSpec, DeviceInfo> get_devices(HoldDevices hold_devices);
97 // See comments on ALSAPool::get_card_state().
98 ALSAPool::Device::State get_alsa_card_state(unsigned index)
100 return alsa_pool.get_card_state(index);
103 // See comments on ALSAPool::create_dead_card().
104 DeviceSpec create_dead_card(const std::string &name, const std::string &info, unsigned num_channels)
106 unsigned dead_card_index = alsa_pool.create_dead_card(name, info, num_channels);
107 return DeviceSpec{InputSourceType::ALSA_INPUT, dead_card_index};
110 void set_display_name(DeviceSpec device_spec, const std::string &name);
112 // Note: The card should be held (currently this isn't enforced, though).
113 void serialize_device(DeviceSpec device_spec, DeviceSpecProto *device_spec_proto);
115 enum class MappingMode {
116 // A single bus, only from a video card (no ALSA devices),
117 // only channel 1 and 2, locked to +0 dB. Note that this is
118 // only an UI abstraction around exactly the same audio code
119 // as MULTICHANNEL; it's just less flexible.
122 // Full, arbitrary mappings.
126 // Automatically sets mapping mode to MappingMode::SIMPLE.
127 void set_simple_input(unsigned card_index);
129 // If mapping mode is not representable as a MappingMode::SIMPLE type
130 // mapping, returns numeric_limits<unsigned>::max().
131 unsigned get_simple_input() const;
133 // Implicitly sets mapping mode to MappingMode::MULTICHANNEL.
134 void set_input_mapping(const InputMapping &input_mapping);
136 MappingMode get_mapping_mode() const;
137 InputMapping get_input_mapping() const;
139 unsigned num_buses() const;
141 void set_locut_cutoff(float cutoff_hz)
143 locut_cutoff_hz = cutoff_hz;
146 float get_locut_cutoff() const
148 return locut_cutoff_hz;
151 void set_locut_enabled(unsigned bus, bool enabled)
153 locut_enabled[bus] = enabled;
156 bool get_locut_enabled(unsigned bus)
158 return locut_enabled[bus];
161 bool is_mono(unsigned bus_index);
163 void set_stereo_width(unsigned bus_index, float width)
165 stereo_width[bus_index] = width;
168 float get_stereo_width(unsigned bus_index)
170 return stereo_width[bus_index];
173 void set_eq(unsigned bus_index, EQBand band, float db_gain)
175 assert(band >= 0 && band < NUM_EQ_BANDS);
176 eq_level_db[bus_index][band] = db_gain;
179 float get_eq(unsigned bus_index, EQBand band) const
181 assert(band >= 0 && band < NUM_EQ_BANDS);
182 return eq_level_db[bus_index][band];
185 float get_limiter_threshold_dbfs() const
187 return limiter_threshold_dbfs;
190 float get_compressor_threshold_dbfs(unsigned bus_index) const
192 return compressor_threshold_dbfs[bus_index];
195 void set_limiter_threshold_dbfs(float threshold_dbfs)
197 limiter_threshold_dbfs = threshold_dbfs;
200 void set_compressor_threshold_dbfs(unsigned bus_index, float threshold_dbfs)
202 compressor_threshold_dbfs[bus_index] = threshold_dbfs;
205 void set_limiter_enabled(bool enabled)
207 limiter_enabled = enabled;
210 bool get_limiter_enabled() const
212 return limiter_enabled;
215 void set_compressor_enabled(unsigned bus_index, bool enabled)
217 compressor_enabled[bus_index] = enabled;
220 bool get_compressor_enabled(unsigned bus_index) const
222 return compressor_enabled[bus_index];
225 void set_gain_staging_db(unsigned bus_index, float gain_db)
227 std::lock_guard<std::mutex> lock(compressor_mutex);
228 level_compressor_enabled[bus_index] = false;
229 gain_staging_db[bus_index] = gain_db;
232 float get_gain_staging_db(unsigned bus_index) const
234 std::lock_guard<std::mutex> lock(compressor_mutex);
235 return gain_staging_db[bus_index];
238 void set_gain_staging_auto(unsigned bus_index, bool enabled)
240 std::lock_guard<std::mutex> lock(compressor_mutex);
241 level_compressor_enabled[bus_index] = enabled;
244 bool get_gain_staging_auto(unsigned bus_index) const
246 std::lock_guard<std::mutex> lock(compressor_mutex);
247 return level_compressor_enabled[bus_index];
250 void set_final_makeup_gain_db(float gain_db)
252 std::lock_guard<std::mutex> lock(compressor_mutex);
253 final_makeup_gain_auto = false;
254 final_makeup_gain = from_db(gain_db);
257 float get_final_makeup_gain_db()
259 std::lock_guard<std::mutex> lock(compressor_mutex);
260 return to_db(final_makeup_gain);
263 void set_final_makeup_gain_auto(bool enabled)
265 std::lock_guard<std::mutex> lock(compressor_mutex);
266 final_makeup_gain_auto = enabled;
269 bool get_final_makeup_gain_auto() const
271 std::lock_guard<std::mutex> lock(compressor_mutex);
272 return final_makeup_gain_auto;
275 void reset_peak(unsigned bus_index);
278 float current_level_dbfs[2]; // Digital peak of last frame, left and right.
279 float peak_level_dbfs[2]; // Digital peak with hold, left and right.
280 float historic_peak_dbfs;
281 float gain_staging_db;
282 float compressor_attenuation_db; // A positive number; 0.0 for no attenuation.
285 typedef std::function<void(float level_lufs, float peak_db,
286 std::vector<BusLevel> bus_levels,
287 float global_level_lufs, float range_low_lufs, float range_high_lufs,
288 float final_makeup_gain_db,
289 float correlation)> audio_level_callback_t;
290 void set_audio_level_callback(audio_level_callback_t callback)
292 audio_level_callback = callback;
295 typedef std::function<void()> state_changed_callback_t;
296 void set_state_changed_callback(state_changed_callback_t callback)
298 state_changed_callback = callback;
301 state_changed_callback_t get_state_changed_callback() const
303 return state_changed_callback;
306 void trigger_state_changed_callback()
308 if (state_changed_callback != nullptr) {
309 state_changed_callback();
313 // A combination of all settings for a bus. Useful if you want to get
314 // or store them as a whole without bothering to call all of the get_*
315 // or set_* functions for that bus.
317 float fader_volume_db;
321 float eq_level_db[NUM_EQ_BANDS];
322 float gain_staging_db;
323 bool level_compressor_enabled;
324 float compressor_threshold_dbfs;
325 bool compressor_enabled;
327 static BusSettings get_default_bus_settings();
328 BusSettings get_bus_settings(unsigned bus_index) const;
329 void set_bus_settings(unsigned bus_index, const BusSettings &settings);
331 // Does not take ownership. Not thread-safe (so only call when the mixer is being created).
332 void set_delay_analyzer(DelayAnalyzerInterface *delay_analyzer)
334 this->delay_analyzer = delay_analyzer;
339 std::unique_ptr<ResamplingQueue> resampling_queue;
340 std::string display_name;
341 unsigned capture_frequency = OUTPUT_FREQUENCY;
342 // Which channels we consider interesting (ie., are part of some input_mapping).
343 std::set<unsigned> interesting_channels;
344 bool silenced = false;
346 // Positive means the audio is delayed, negative means we try to have it earlier
347 // (although we can't time-travel!). Stored together with the input mapping.
348 double extra_delay_ms = 0.0;
351 const AudioDevice *find_audio_device(DeviceSpec device_spec) const
353 return const_cast<AudioMixer *>(this)->find_audio_device(device_spec);
356 AudioDevice *find_audio_device(DeviceSpec device_spec);
358 void find_sample_src_from_device(const std::map<DeviceSpec, std::vector<float>> &samples_card, DeviceSpec device_spec, int source_channel, const float **srcptr, unsigned *stride);
359 void fill_audio_bus(const std::map<DeviceSpec, std::vector<float>> &samples_card, const InputMapping::Bus &bus, unsigned num_samples, float stereo_width, float *output);
360 void reset_resampler_mutex_held(DeviceSpec device_spec);
361 void apply_eq(unsigned bus_index, std::vector<float> *samples_bus);
362 void update_meters(const std::vector<float> &samples);
363 void add_bus_to_master(unsigned bus_index, const std::vector<float> &samples_bus, std::vector<float> *samples_out);
364 void measure_bus_levels(unsigned bus_index, const std::vector<float> &left, const std::vector<float> &right);
365 void send_audio_level_callback();
366 std::vector<DeviceSpec> get_active_devices() const;
367 void set_input_mapping_lock_held(const InputMapping &input_mapping);
369 unsigned num_capture_cards, num_ffmpeg_inputs;
371 mutable std::timed_mutex audio_mutex;
374 AudioDevice video_cards[MAX_VIDEO_CARDS]; // Under audio_mutex.
375 AudioDevice alsa_inputs[MAX_ALSA_CARDS]; // Under audio_mutex.
376 std::unique_ptr<AudioDevice[]> ffmpeg_inputs; // Under audio_mutex.
378 std::atomic<float> locut_cutoff_hz{120};
379 StereoFilter locut[MAX_BUSES]; // Default cutoff 120 Hz, 24 dB/oct.
380 std::atomic<bool> locut_enabled[MAX_BUSES];
381 StereoFilter eq[MAX_BUSES][NUM_EQ_BANDS]; // The one for EQBand::MID isn't actually used (see comments in apply_eq()).
383 // First compressor; takes us up to about -12 dBFS.
384 mutable std::mutex compressor_mutex;
385 std::unique_ptr<StereoCompressor> level_compressor[MAX_BUSES]; // Under compressor_mutex. Used to set/override gain_staging_db if <level_compressor_enabled>.
386 float gain_staging_db[MAX_BUSES]; // Under compressor_mutex.
387 float last_gain_staging_db[MAX_BUSES]; // Under compressor_mutex.
388 bool level_compressor_enabled[MAX_BUSES]; // Under compressor_mutex.
390 static constexpr float ref_level_dbfs = -14.0f; // Chosen so that we end up around 0 LU in practice.
391 static constexpr float ref_level_lufs = -23.0f; // 0 LU, more or less by definition.
393 StereoCompressor limiter;
394 std::atomic<float> limiter_threshold_dbfs{ref_level_dbfs + 4.0f}; // 4 dB.
395 std::atomic<bool> limiter_enabled{true};
396 std::unique_ptr<StereoCompressor> compressor[MAX_BUSES];
397 std::atomic<float> compressor_threshold_dbfs[MAX_BUSES];
398 std::atomic<bool> compressor_enabled[MAX_BUSES];
400 // Note: The values here are not in dB.
402 float current_level = 0.0f; // Peak of the last frame.
403 float historic_peak = 0.0f; // Highest peak since last reset; no falloff.
404 float current_peak = 0.0f; // Current peak of the peak meter.
405 float last_peak = 0.0f;
406 float age_seconds = 0.0f; // Time since "last_peak" was set.
408 PeakHistory peak_history[MAX_BUSES][2]; // Separate for each channel. Under audio_mutex.
410 double final_makeup_gain = 1.0; // Under compressor_mutex. Read/write by the user. Note: Not in dB, we want the numeric precision so that we can change it slowly.
411 bool final_makeup_gain_auto = true; // Under compressor_mutex.
413 MappingMode current_mapping_mode; // Under audio_mutex.
414 InputMapping input_mapping; // Under audio_mutex.
415 std::atomic<float> fader_volume_db[MAX_BUSES] {{ 0.0f }};
416 std::atomic<bool> mute[MAX_BUSES] {{ false }};
417 float last_fader_volume_db[MAX_BUSES] { 0.0f }; // Under audio_mutex.
418 std::atomic<float> stereo_width[MAX_BUSES] {{ 0.0f }}; // Default 1.0f (is set in constructor).
419 std::atomic<float> eq_level_db[MAX_BUSES][NUM_EQ_BANDS] {{{ 0.0f }}};
420 float last_eq_level_db[MAX_BUSES][NUM_EQ_BANDS] {{ 0.0f }};
422 audio_level_callback_t audio_level_callback = nullptr;
423 state_changed_callback_t state_changed_callback = nullptr;
424 mutable std::mutex audio_measure_mutex;
425 Ebu_r128_proc r128; // Under audio_measure_mutex.
426 CorrelationMeasurer correlation; // Under audio_measure_mutex.
427 Resampler peak_resampler; // Under audio_measure_mutex.
428 std::atomic<float> peak{0.0f};
431 std::atomic<double> metric_audio_loudness_short_lufs{0.0 / 0.0};
432 std::atomic<double> metric_audio_loudness_integrated_lufs{0.0 / 0.0};
433 std::atomic<double> metric_audio_loudness_range_low_lufs{0.0 / 0.0};
434 std::atomic<double> metric_audio_loudness_range_high_lufs{0.0 / 0.0};
435 std::atomic<double> metric_audio_peak_dbfs{0.0 / 0.0};
436 std::atomic<double> metric_audio_final_makeup_gain_db{0.0};
437 std::atomic<double> metric_audio_correlation{0.0};
439 // These are all gauges corresponding to the elements of BusLevel.
440 // In a sense, they'd probably do better as histograms, but that's an
441 // awful lot of time series when you have many buses.
443 std::vector<std::pair<std::string, std::string>> labels;
444 std::atomic<double> current_level_dbfs[2]{{0.0/0.0},{0.0/0.0}};
445 std::atomic<double> peak_level_dbfs[2]{{0.0/0.0},{0.0/0.0}};
446 std::atomic<double> historic_peak_dbfs{0.0/0.0};
447 std::atomic<double> gain_staging_db{0.0/0.0};
448 std::atomic<double> compressor_attenuation_db{0.0/0.0};
450 std::unique_ptr<BusMetrics[]> bus_metrics; // One for each bus in <input_mapping>.
452 DelayAnalyzerInterface *delay_analyzer = nullptr;
455 extern AudioMixer *global_audio_mixer;
457 #endif // !defined(_AUDIO_MIXER_H)