#include "mixer.h"
#include <assert.h>
+#include <endian.h>
#include <epoxy/egl.h>
#include <movit/effect_chain.h>
#include <movit/effect_util.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
+#include <sys/resource.h>
#include <sys/time.h>
#include <time.h>
#include <algorithm>
+#include <chrono>
#include <cmath>
#include <condition_variable>
#include <cstddef>
#include <thread>
#include <utility>
#include <vector>
-#include <arpa/inet.h>
-#include <sys/time.h>
-#include <sys/resource.h>
#include "bmusb/bmusb.h"
+#include "bmusb/fake_capture.h"
#include "context.h"
+#include "db.h"
#include "decklink_capture.h"
#include "defs.h"
-#include "fake_capture.h"
+#include "disk_space_estimator.h"
#include "flags.h"
-#include "video_encoder.h"
#include "pbo_frame_allocator.h"
#include "ref_counted_gl_sync.h"
#include "timebase.h"
+#include "video_encoder.h"
class QOpenGLContext;
using namespace movit;
using namespace std;
+using namespace std::chrono;
using namespace std::placeholders;
+using namespace bmusb;
Mixer *global_mixer = nullptr;
bool uses_mlock = false;
namespace {
-void convert_fixed24_to_fp32(float *dst, size_t out_channels, const uint8_t *src, size_t in_channels, size_t num_samples)
-{
- assert(in_channels >= out_channels);
- for (size_t i = 0; i < num_samples; ++i) {
- for (size_t j = 0; j < out_channels; ++j) {
- uint32_t s1 = *src++;
- uint32_t s2 = *src++;
- uint32_t s3 = *src++;
- uint32_t s = s1 | (s1 << 8) | (s2 << 16) | (s3 << 24);
- dst[i * out_channels + j] = int(s) * (1.0f / 4294967296.0f);
- }
- src += 3 * (in_channels - out_channels);
- }
-}
-
-void convert_fixed32_to_fp32(float *dst, size_t out_channels, const uint8_t *src, size_t in_channels, size_t num_samples)
-{
- assert(in_channels >= out_channels);
- for (size_t i = 0; i < num_samples; ++i) {
- for (size_t j = 0; j < out_channels; ++j) {
- // Note: Assumes little-endian.
- int32_t s = *(int32_t *)src;
- dst[i * out_channels + j] = s * (1.0f / 4294967296.0f);
- src += 4;
- }
- src += 4 * (in_channels - out_channels);
- }
-}
-
void insert_new_frame(RefCountedFrame frame, unsigned field_num, bool interlaced, unsigned card_index, InputState *input_state)
{
if (interlaced) {
num_cards(num_cards),
mixer_surface(create_surface(format)),
h264_encoder_surface(create_surface(format)),
- correlation(OUTPUT_FREQUENCY),
- level_compressor(OUTPUT_FREQUENCY),
- limiter(OUTPUT_FREQUENCY),
- compressor(OUTPUT_FREQUENCY)
+ audio_mixer(num_cards),
+ correlation(OUTPUT_FREQUENCY)
{
CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF));
check_error();
movit_texel_subpixel_precision /= 2.0;
resource_pool.reset(new ResourcePool);
- theme.reset(new Theme(global_flags.theme_filename.c_str(), resource_pool.get(), num_cards));
+ theme.reset(new Theme(global_flags.theme_filename, global_flags.theme_dirs, resource_pool.get(), num_cards));
for (unsigned i = 0; i < NUM_OUTPUTS; ++i) {
output_channel[i].parent = this;
output_channel[i].channel = i;
display_chain->set_dither_bits(0); // Don't bother.
display_chain->finalize();
- video_encoder.reset(new VideoEncoder(resource_pool.get(), h264_encoder_surface, global_flags.va_display, WIDTH, HEIGHT, &httpd));
+ video_encoder.reset(new VideoEncoder(resource_pool.get(), h264_encoder_surface, global_flags.va_display, WIDTH, HEIGHT, &httpd, global_disk_space_estimator));
// Start listening for clients only once VideoEncoder has written its header, if any.
httpd.start(9095);
- // First try initializing the fake devices, then PCI devices, then USB,
- // until we have the desired number of cards.
- unsigned num_pci_devices = 0, num_usb_devices = 0;
+ // First try initializing the then PCI devices, then USB, then
+ // fill up with fake cards until we have the desired number of cards.
+ unsigned num_pci_devices = 0;
unsigned card_index = 0;
- assert(global_flags.num_fake_cards >= 0); // Enforced in flags.cpp.
- unsigned num_fake_cards = global_flags.num_fake_cards;
-
- assert(num_fake_cards <= num_cards); // Enforced in flags.cpp.
- for ( ; card_index < num_fake_cards; ++card_index) {
- configure_card(card_index, new FakeCapture(card_index), /*is_fake_capture=*/true);
- }
-
- if (global_flags.num_fake_cards > 0) {
- fprintf(stderr, "Initialized %d fake cards.\n", global_flags.num_fake_cards);
- }
-
- if (card_index < num_cards) {
+ {
IDeckLinkIterator *decklink_iterator = CreateDeckLinkIteratorInstance();
if (decklink_iterator != nullptr) {
for ( ; card_index < num_cards; ++card_index) {
break;
}
- configure_card(card_index, new DeckLinkCapture(decklink, card_index - num_fake_cards), /*is_fake_capture=*/false);
+ configure_card(card_index, new DeckLinkCapture(decklink, card_index), /*is_fake_capture=*/false);
++num_pci_devices;
}
decklink_iterator->Release();
- fprintf(stderr, "Found %d DeckLink PCI card(s).\n", num_pci_devices);
+ fprintf(stderr, "Found %u DeckLink PCI card(s).\n", num_pci_devices);
} else {
fprintf(stderr, "DeckLink drivers not found. Probing for USB cards only.\n");
}
}
- for ( ; card_index < num_cards; ++card_index) {
- BMUSBCapture *capture = new BMUSBCapture(card_index - num_pci_devices - num_fake_cards);
+ unsigned num_usb_devices = BMUSBCapture::num_cards();
+ for (unsigned usb_card_index = 0; usb_card_index < num_usb_devices && card_index < num_cards; ++usb_card_index, ++card_index) {
+ BMUSBCapture *capture = new BMUSBCapture(usb_card_index);
capture->set_card_disconnected_callback(bind(&Mixer::bm_hotplug_remove, this, card_index));
configure_card(card_index, capture, /*is_fake_capture=*/false);
- ++num_usb_devices;
+ }
+ fprintf(stderr, "Found %u USB card(s).\n", num_usb_devices);
+
+ unsigned num_fake_cards = 0;
+ for ( ; card_index < num_cards; ++card_index, ++num_fake_cards) {
+ FakeCapture *capture = new FakeCapture(WIDTH, HEIGHT, FAKE_FPS, OUTPUT_FREQUENCY, card_index, global_flags.fake_cards_audio);
+ configure_card(card_index, capture, /*is_fake_capture=*/true);
}
- if (num_usb_devices > 0) {
- has_bmusb_thread = true;
- BMUSBCapture::set_card_connected_callback(bind(&Mixer::bm_hotplug_add, this, _1));
- BMUSBCapture::start_bm_thread();
+ if (num_fake_cards > 0) {
+ fprintf(stderr, "Initialized %u fake cards.\n", num_fake_cards);
}
+ BMUSBCapture::set_card_connected_callback(bind(&Mixer::bm_hotplug_add, this, _1));
+ BMUSBCapture::start_bm_thread();
+
for (card_index = 0; card_index < num_cards; ++card_index) {
cards[card_index].queue_length_policy.reset(card_index);
cards[card_index].capture->start_bm_capture();
r128.init(2, OUTPUT_FREQUENCY);
r128.integr_start();
- locut.init(FILTER_HPF, 2);
-
- // If --flat-audio is given, turn off everything that messes with the sound,
- // except the final makeup gain.
- if (global_flags.flat_audio) {
- set_locut_enabled(false);
- set_gain_staging_auto(false);
- set_limiter_enabled(false);
- set_compressor_enabled(false);
- }
-
// hlen=16 is pretty low quality, but we use quite a bit of CPU otherwise,
// and there's a limit to how important the peak meter is.
peak_resampler.setup(OUTPUT_FREQUENCY, OUTPUT_FREQUENCY * 4, /*num_channels=*/2, /*hlen=*/16, /*frel=*/1.0);
{
resource_pool->release_glsl_program(cbcr_program_num);
glDeleteBuffers(1, &cbcr_vbo);
- if (has_bmusb_thread) {
- BMUSBCapture::stop_bm_thread();
- }
+ BMUSBCapture::stop_bm_thread();
for (unsigned card_index = 0; card_index < num_cards; ++card_index) {
{
if (card->surface == nullptr) {
card->surface = create_surface_with_same_format(mixer_surface);
}
- {
- unique_lock<mutex> lock(cards[card_index].audio_mutex);
- card->resampling_queue.reset(new ResamplingQueue(card_index, OUTPUT_FREQUENCY, OUTPUT_FREQUENCY, 2));
- }
while (!card->new_frames.empty()) card->new_frames.pop();
card->fractional_samples = 0;
card->last_timecode = -1;
- card->next_local_pts = 0;
card->capture->configure_card();
+
+ DeviceSpec device{InputSourceType::CAPTURE_CARD, card_index};
+ audio_mixer.reset_device(device);
+ audio_mixer.set_name(device, card->capture->get_description());
}
FrameAllocator::Frame video_frame, size_t video_offset, VideoFormat video_format,
FrameAllocator::Frame audio_frame, size_t audio_offset, AudioFormat audio_format)
{
+ DeviceSpec device{InputSourceType::CAPTURE_CARD, card_index};
CaptureCard *card = &cards[card_index];
if (is_mode_scanning[card_index]) {
is_mode_scanning[card_index] = false;
} else {
static constexpr double switch_time_s = 0.5; // Should be enough time for the signal to stabilize.
- timespec now;
- clock_gettime(CLOCK_MONOTONIC, &now);
- double sec_since_last_switch = (now.tv_sec - last_mode_scan_change[card_index].tv_sec) +
- 1e-9 * (now.tv_nsec - last_mode_scan_change[card_index].tv_nsec);
+ steady_clock::time_point now = steady_clock::now();
+ double sec_since_last_switch = duration<double>(steady_clock::now() - last_mode_scan_change[card_index]).count();
if (sec_since_last_switch > switch_time_s) {
// It isn't this mode; try the next one.
mode_scanlist_index[card_index]++;
return;
}
- int64_t local_pts = card->next_local_pts;
int dropped_frames = 0;
if (card->last_timecode != -1) {
dropped_frames = unwrap_timecode(timecode, card->last_timecode) - card->last_timecode - 1;
}
- // Convert the audio to stereo fp32 and add it.
- vector<float> audio;
- audio.resize(num_samples * 2);
- switch (audio_format.bits_per_sample) {
- case 0:
- assert(num_samples == 0);
- break;
- case 24:
- convert_fixed24_to_fp32(&audio[0], 2, audio_frame.data + audio_offset, audio_format.num_channels, num_samples);
- break;
- case 32:
- convert_fixed32_to_fp32(&audio[0], 2, audio_frame.data + audio_offset, audio_format.num_channels, num_samples);
- break;
- default:
- fprintf(stderr, "Cannot handle audio with %u bits per sample\n", audio_format.bits_per_sample);
- assert(false);
- }
+ // Number of samples per frame if we need to insert silence.
+ // (Could be nonintegral, but resampling will save us then.)
+ const int silence_samples = OUTPUT_FREQUENCY * video_format.frame_rate_den / video_format.frame_rate_nom;
- // Add the audio.
- {
- unique_lock<mutex> lock(card->audio_mutex);
-
- // Number of samples per frame if we need to insert silence.
- // (Could be nonintegral, but resampling will save us then.)
- int silence_samples = OUTPUT_FREQUENCY * video_format.frame_rate_den / video_format.frame_rate_nom;
-
- if (dropped_frames > MAX_FPS * 2) {
- fprintf(stderr, "Card %d lost more than two seconds (or time code jumping around; from 0x%04x to 0x%04x), resetting resampler\n",
- card_index, card->last_timecode, timecode);
- card->resampling_queue.reset(new ResamplingQueue(card_index, OUTPUT_FREQUENCY, OUTPUT_FREQUENCY, 2));
- dropped_frames = 0;
- } else if (dropped_frames > 0) {
- // Insert silence as needed.
- fprintf(stderr, "Card %d dropped %d frame(s) (before timecode 0x%04x), inserting silence.\n",
- card_index, dropped_frames, timecode);
- vector<float> silence(silence_samples * 2, 0.0f);
- for (int i = 0; i < dropped_frames; ++i) {
- card->resampling_queue->add_input_samples(local_pts / double(TIMEBASE), silence.data(), silence_samples);
- // Note that if the format changed in the meantime, we have
- // no way of detecting that; we just have to assume the frame length
- // is always the same.
- local_pts += frame_length;
- }
- }
- if (num_samples == 0) {
- audio.resize(silence_samples * 2);
- num_samples = silence_samples;
- }
- card->resampling_queue->add_input_samples(local_pts / double(TIMEBASE), audio.data(), num_samples);
- card->next_local_pts = local_pts + frame_length;
+ if (dropped_frames > MAX_FPS * 2) {
+ fprintf(stderr, "Card %d lost more than two seconds (or time code jumping around; from 0x%04x to 0x%04x), resetting resampler\n",
+ card_index, card->last_timecode, timecode);
+ audio_mixer.reset_device(device);
+ dropped_frames = 0;
+ } else if (dropped_frames > 0) {
+ // Insert silence as needed.
+ fprintf(stderr, "Card %d dropped %d frame(s) (before timecode 0x%04x), inserting silence.\n",
+ card_index, dropped_frames, timecode);
+
+ audio_mixer.add_silence(device, silence_samples, dropped_frames, frame_length);
}
- card->last_timecode = timecode;
+ audio_mixer.add_audio(device, audio_frame.data + audio_offset, num_samples, audio_format, frame_length);
// Done with the audio, so release it.
if (audio_frame.owner) {
audio_frame.owner->release_frame(audio_frame);
}
+ card->last_timecode = timecode;
+
size_t expected_length = video_format.width * (video_format.height + video_format.extra_lines_top + video_format.extra_lines_bottom) * 2;
if (video_frame.len - video_offset == 0 ||
video_frame.len - video_offset != expected_length) {
PBOFrameAllocator::Userdata *userdata = (PBOFrameAllocator::Userdata *)video_frame.userdata;
unsigned num_fields = video_format.interlaced ? 2 : 1;
- timespec frame_upload_start;
+ steady_clock::time_point frame_upload_start;
if (video_format.interlaced) {
// Send the two fields along as separate frames; the other side will need to add
// a deinterlacer to actually get this right.
assert(frame_length % 2 == 0);
frame_length /= 2;
num_fields = 2;
- clock_gettime(CLOCK_MONOTONIC, &frame_upload_start);
+ frame_upload_start = steady_clock::now();
}
userdata->last_interlaced = video_format.interlaced;
userdata->last_has_signal = video_format.has_signal;
// against the video display, although the latter is not as critical.)
// This requires our system clock to be reasonably close to the
// video clock, but that's not an unreasonable assumption.
- timespec second_field_start;
- second_field_start.tv_nsec = frame_upload_start.tv_nsec +
- frame_length * 1000000000 / TIMEBASE;
- second_field_start.tv_sec = frame_upload_start.tv_sec +
- second_field_start.tv_nsec / 1000000000;
- second_field_start.tv_nsec %= 1000000000;
-
- while (clock_nanosleep(CLOCK_MONOTONIC, TIMER_ABSTIME,
- &second_field_start, nullptr) == -1 &&
- errno == EINTR) ;
+ steady_clock::time_point second_field_start = frame_upload_start +
+ nanoseconds(frame_length * 1000000000 / TIMEBASE);
+ this_thread::sleep_until(second_field_start);
}
{
exit(1);
}
- struct timespec start, now;
- clock_gettime(CLOCK_MONOTONIC, &start);
+ steady_clock::time_point start, now;
+ start = steady_clock::now();
int frame = 0;
int stats_dropped_frames = 0;
while (!should_quit) {
- CaptureCard::NewFrame new_frames[MAX_CARDS];
- bool has_new_frame[MAX_CARDS] = { false };
- int num_samples[MAX_CARDS] = { 0 };
+ CaptureCard::NewFrame new_frames[MAX_VIDEO_CARDS];
+ bool has_new_frame[MAX_VIDEO_CARDS] = { false };
+ int num_samples[MAX_VIDEO_CARDS] = { 0 };
unsigned master_card_index = theme->map_signal(master_clock_channel);
assert(master_card_index < num_cards);
}
}
- int64_t duration = new_frames[master_card_index].length;
- render_one_frame(duration);
+ int64_t frame_duration = new_frames[master_card_index].length;
+ render_one_frame(frame_duration);
++frame;
- pts_int += duration;
+ pts_int += frame_duration;
- clock_gettime(CLOCK_MONOTONIC, &now);
- double elapsed = now.tv_sec - start.tv_sec +
- 1e-9 * (now.tv_nsec - start.tv_nsec);
+ now = steady_clock::now();
+ double elapsed = duration<double>(now - start).count();
if (frame % 100 == 0) {
- // check our memory usage, to see if we are close to our mlockall()
- // limit (if at all set).
- rusage used;
- if (getrusage(RUSAGE_SELF, &used) == -1) {
- perror("getrusage(RUSAGE_SELF)");
- assert(false);
- }
-
- rlimit limit;
- if (getrlimit(RLIMIT_MEMLOCK, &limit) == -1) {
- perror("getrlimit(RLIMIT_MEMLOCK)");
- assert(false);
- }
-
printf("%d frames (%d dropped) in %.3f seconds = %.1f fps (%.1f ms/frame)",
frame, stats_dropped_frames, elapsed, frame / elapsed,
1e3 * elapsed / frame);
// chain->print_phase_timing();
- if (uses_mlock) {
- // Check our memory usage, to see if we are close to our mlockall()
- // limit (if at all set).
- rusage used;
- if (getrusage(RUSAGE_SELF, &used) == -1) {
- perror("getrusage(RUSAGE_SELF)");
- assert(false);
- }
+ // Check our memory usage, to see if we are close to our mlockall()
+ // limit (if at all set).
+ rusage used;
+ if (getrusage(RUSAGE_SELF, &used) == -1) {
+ perror("getrusage(RUSAGE_SELF)");
+ assert(false);
+ }
+ if (uses_mlock) {
rlimit limit;
if (getrlimit(RLIMIT_MEMLOCK, &limit) == -1) {
perror("getrlimit(RLIMIT_MEMLOCK)");
long(used.ru_maxrss / 1024),
long(limit.rlim_cur / 1048576),
float(100.0 * (used.ru_maxrss * 1024.0) / limit.rlim_cur));
+ } else {
+ printf(", using %ld MB memory (not locked)",
+ long(used.ru_maxrss / 1024));
}
printf("\n");
resource_pool->clean_context();
}
-void Mixer::get_one_frame_from_each_card(unsigned master_card_index, CaptureCard::NewFrame new_frames[MAX_CARDS], bool has_new_frame[MAX_CARDS], int num_samples[MAX_CARDS])
+void Mixer::get_one_frame_from_each_card(unsigned master_card_index, CaptureCard::NewFrame new_frames[MAX_VIDEO_CARDS], bool has_new_frame[MAX_VIDEO_CARDS], int num_samples[MAX_VIDEO_CARDS])
{
start:
// The first card is the master timer, so wait for it to have a new frame.
CaptureCard *card = &cards[card_index];
if (card->capture->get_disconnected()) {
fprintf(stderr, "Card %u went away, replacing with a fake card.\n", card_index);
- configure_card(card_index, new FakeCapture(card_index), /*is_fake_capture=*/true);
+ FakeCapture *capture = new FakeCapture(WIDTH, HEIGHT, FAKE_FPS, OUTPUT_FREQUENCY, card_index, global_flags.fake_cards_audio);
+ configure_card(card_index, capture, /*is_fake_capture=*/true);
card->queue_length_policy.reset(card_index);
card->capture->start_bm_capture();
}
// Resample the audio as needed, including from previously dropped frames.
assert(num_cards > 0);
for (unsigned frame_num = 0; frame_num < dropped_frames + 1; ++frame_num) {
+ const bool dropped_frame = (frame_num != dropped_frames);
{
// Signal to the audio thread to process this frame.
+ // Note that if the frame is a dropped frame, we signal that
+ // we don't want to use this frame as base for adjusting
+ // the resampler rate. The reason for this is that the timing
+ // of these frames is often way too late; they typically don't
+ // “arrive” before we synthesize them. Thus, we could end up
+ // in a situation where we have inserted e.g. five audio frames
+ // into the queue before we then start pulling five of them
+ // back out. This makes ResamplingQueue overestimate the delay,
+ // causing undue resampler changes. (We _do_ use the last,
+ // non-dropped frame; perhaps we should just discard that as well,
+ // since dropped frames are expected to be rare, and it might be
+ // better to just wait until we have a slightly more normal situation).
unique_lock<mutex> lock(audio_mutex);
- audio_task_queue.push(AudioTask{pts_int, num_samples_per_frame});
+ bool adjust_rate = !dropped_frame;
+ audio_task_queue.push(AudioTask{pts_int, num_samples_per_frame, adjust_rate});
audio_task_queue_changed.notify_one();
}
- if (frame_num != dropped_frames) {
+ if (dropped_frame) {
// For dropped frames, increase the pts. Note that if the format changed
// in the meantime, we have no way of detecting that; we just have to
// assume the frame length is always the same.
return;
}
- unique_lock<mutex> lock(compressor_mutex);
+ unique_lock<mutex> lock(audio_measure_mutex);
double loudness_s = r128.loudness_S();
double loudness_i = r128.integrated();
double loudness_range_low = r128.range_min();
double loudness_range_high = r128.range_max();
- audio_level_callback(loudness_s, 20.0 * log10(peak),
+ audio_level_callback(loudness_s, to_db(peak),
loudness_i, loudness_range_low, loudness_range_high,
- gain_staging_db, 20.0 * log10(final_makeup_gain),
+ audio_mixer.get_gain_staging_db(),
+ audio_mixer.get_final_makeup_gain_db(),
correlation.get_correlation());
}
audio_task_queue.pop();
}
- process_audio_one_frame(task.pts_int, task.num_samples);
+ ResamplingQueue::RateAdjustmentPolicy rate_adjustment_policy =
+ task.adjust_rate ? ResamplingQueue::ADJUST_RATE : ResamplingQueue::DO_NOT_ADJUST_RATE;
+ process_audio_one_frame(task.pts_int, task.num_samples, rate_adjustment_policy);
}
}
-void Mixer::process_audio_one_frame(int64_t frame_pts_int, int num_samples)
+void Mixer::process_audio_one_frame(int64_t frame_pts_int, int num_samples, ResamplingQueue::RateAdjustmentPolicy rate_adjustment_policy)
{
- vector<float> samples_card;
- vector<float> samples_out;
-
- // TODO: Allow mixing audio from several sources.
- unsigned selected_audio_card = theme->map_signal(audio_source_channel);
- assert(selected_audio_card < num_cards);
-
- for (unsigned card_index = 0; card_index < num_cards; ++card_index) {
- samples_card.resize(num_samples * 2);
- {
- unique_lock<mutex> lock(cards[card_index].audio_mutex);
- cards[card_index].resampling_queue->get_output_samples(double(frame_pts_int) / TIMEBASE, &samples_card[0], num_samples);
- }
- if (card_index == selected_audio_card) {
- samples_out = move(samples_card);
- }
- }
-
- // Cut away everything under 120 Hz (or whatever the cutoff is);
- // we don't need it for voice, and it will reduce headroom
- // and confuse the compressor. (In particular, any hums at 50 or 60 Hz
- // should be dampened.)
- if (locut_enabled) {
- locut.render(samples_out.data(), samples_out.size() / 2, locut_cutoff_hz * 2.0 * M_PI / OUTPUT_FREQUENCY, 0.5f);
- }
-
- // Apply a level compressor to get the general level right.
- // Basically, if it's over about -40 dBFS, we squeeze it down to that level
- // (or more precisely, near it, since we don't use infinite ratio),
- // then apply a makeup gain to get it to -14 dBFS. -14 dBFS is, of course,
- // entirely arbitrary, but from practical tests with speech, it seems to
- // put ut around -23 LUFS, so it's a reasonable starting point for later use.
- {
- unique_lock<mutex> lock(compressor_mutex);
- if (level_compressor_enabled) {
- float threshold = 0.01f; // -40 dBFS.
- float ratio = 20.0f;
- float attack_time = 0.5f;
- float release_time = 20.0f;
- float makeup_gain = pow(10.0f, (ref_level_dbfs - (-40.0f)) / 20.0f); // +26 dB.
- level_compressor.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain);
- gain_staging_db = 20.0 * log10(level_compressor.get_attenuation() * makeup_gain);
- } else {
- // Just apply the gain we already had.
- float g = pow(10.0f, gain_staging_db / 20.0f);
- for (size_t i = 0; i < samples_out.size(); ++i) {
- samples_out[i] *= g;
- }
- }
- }
-
-#if 0
- printf("level=%f (%+5.2f dBFS) attenuation=%f (%+5.2f dB) end_result=%+5.2f dB\n",
- level_compressor.get_level(), 20.0 * log10(level_compressor.get_level()),
- level_compressor.get_attenuation(), 20.0 * log10(level_compressor.get_attenuation()),
- 20.0 * log10(level_compressor.get_level() * level_compressor.get_attenuation() * makeup_gain));
-#endif
-
-// float limiter_att, compressor_att;
-
- // The real compressor.
- if (compressor_enabled) {
- float threshold = pow(10.0f, compressor_threshold_dbfs / 20.0f);
- float ratio = 20.0f;
- float attack_time = 0.005f;
- float release_time = 0.040f;
- float makeup_gain = 2.0f; // +6 dB.
- compressor.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain);
-// compressor_att = compressor.get_attenuation();
- }
-
- // Finally a limiter at -4 dB (so, -10 dBFS) to take out the worst peaks only.
- // Note that since ratio is not infinite, we could go slightly higher than this.
- if (limiter_enabled) {
- float threshold = pow(10.0f, limiter_threshold_dbfs / 20.0f);
- float ratio = 30.0f;
- float attack_time = 0.0f; // Instant.
- float release_time = 0.020f;
- float makeup_gain = 1.0f; // 0 dB.
- limiter.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain);
-// limiter_att = limiter.get_attenuation();
- }
-
-// printf("limiter=%+5.1f compressor=%+5.1f\n", 20.0*log10(limiter_att), 20.0*log10(compressor_att));
+ vector<float> samples_out = audio_mixer.get_output(double(frame_pts_int) / TIMEBASE, num_samples, rate_adjustment_policy);
// Upsample 4x to find interpolated peak.
peak_resampler.inp_data = samples_out.data();
vector<float> interpolated_samples_out;
interpolated_samples_out.resize(samples_out.size());
- while (peak_resampler.inp_count > 0) { // About four iterations.
- peak_resampler.out_data = &interpolated_samples_out[0];
- peak_resampler.out_count = interpolated_samples_out.size() / 2;
- peak_resampler.process();
- size_t out_stereo_samples = interpolated_samples_out.size() / 2 - peak_resampler.out_count;
- peak = max<float>(peak, find_peak(interpolated_samples_out.data(), out_stereo_samples * 2));
- peak_resampler.out_data = nullptr;
- }
-
- // At this point, we are most likely close to +0 LU, but all of our
- // measurements have been on raw sample values, not R128 values.
- // So we have a final makeup gain to get us to +0 LU; the gain
- // adjustments required should be relatively small, and also, the
- // offset shouldn't change much (only if the type of audio changes
- // significantly). Thus, we shoot for updating this value basically
- // “whenever we process buffers”, since the R128 calculation isn't exactly
- // something we get out per-sample.
- //
- // Note that there's a feedback loop here, so we choose a very slow filter
- // (half-time of 100 seconds).
- double target_loudness_factor, alpha;
{
- unique_lock<mutex> lock(compressor_mutex);
- double loudness_lu = r128.loudness_M() - ref_level_lufs;
- double current_makeup_lu = 20.0f * log10(final_makeup_gain);
- target_loudness_factor = pow(10.0f, -loudness_lu / 20.0f);
-
- // If we're outside +/- 5 LU uncorrected, we don't count it as
- // a normal signal (probably silence) and don't change the
- // correction factor; just apply what we already have.
- if (fabs(loudness_lu - current_makeup_lu) >= 5.0 || !final_makeup_gain_auto) {
- alpha = 0.0;
- } else {
- // Formula adapted from
- // https://en.wikipedia.org/wiki/Low-pass_filter#Simple_infinite_impulse_response_filter.
- const double half_time_s = 100.0;
- const double fc_mul_2pi_delta_t = 1.0 / (half_time_s * OUTPUT_FREQUENCY);
- alpha = fc_mul_2pi_delta_t / (fc_mul_2pi_delta_t + 1.0);
- }
-
- double m = final_makeup_gain;
- for (size_t i = 0; i < samples_out.size(); i += 2) {
- samples_out[i + 0] *= m;
- samples_out[i + 1] *= m;
- m += (target_loudness_factor - m) * alpha;
+ unique_lock<mutex> lock(audio_measure_mutex);
+
+ while (peak_resampler.inp_count > 0) { // About four iterations.
+ peak_resampler.out_data = &interpolated_samples_out[0];
+ peak_resampler.out_count = interpolated_samples_out.size() / 2;
+ peak_resampler.process();
+ size_t out_stereo_samples = interpolated_samples_out.size() / 2 - peak_resampler.out_count;
+ peak = max<float>(peak, find_peak(interpolated_samples_out.data(), out_stereo_samples * 2));
+ peak_resampler.out_data = nullptr;
}
- final_makeup_gain = m;
}
// Find R128 levels and L/R correlation.
deinterleave_samples(samples_out, &left, &right);
float *ptrs[] = { left.data(), right.data() };
{
- unique_lock<mutex> lock(compressor_mutex);
+ unique_lock<mutex> lock(audio_measure_mutex);
r128.process(left.size(), ptrs);
+ audio_mixer.set_current_loudness(r128.loudness_M());
correlation.process_samples(samples_out);
}
void Mixer::reset_meters()
{
+ unique_lock<mutex> lock(audio_measure_mutex);
peak_resampler.reset();
peak = 0.0f;
r128.reset();
assert(!mode_scanlist[card_index].empty());
mode_scanlist_index[card_index] = 0;
cards[card_index].capture->set_video_mode(mode_scanlist[card_index][0]);
- clock_gettime(CLOCK_MONOTONIC, &last_mode_scan_change[card_index]);
+ last_mode_scan_change[card_index] = steady_clock::now();
}
Mixer::OutputChannel::~OutputChannel()