Make it possible to override the level compressor with the gain staging knob.

[nageru] / mixer.cpp

diff --git a/mixer.cpp b/mixer.cpp
index 61050067b7b9dd60c652a855336eea54f68dec9d..72022026f771056e0211ccac874e69c8ee0ed22a 100644 (file)
--- a/mixer.cpp
+++ b/mixer.cpp
@@ -74,10 +74,27 @@ void insert_new_frame(RefCountedFrame frame, unsigned field_num, bool interlaced
        }
 }
 
+string generate_local_dump_filename(int frame)
+{
+       time_t now = time(NULL);
+       tm now_tm;
+       localtime_r(&now, &now_tm);
+
+       char timestamp[256];
+       strftime(timestamp, sizeof(timestamp), "%F-%T%z", &now_tm);
+
+       // Use the frame number to disambiguate between two cuts starting
+       // on the same second.
+       char filename[256];
+       snprintf(filename, sizeof(filename), "%s%s-f%02d%s",
+               LOCAL_DUMP_PREFIX, timestamp, frame % 100, LOCAL_DUMP_SUFFIX);
+       return filename;
+}
+
 }  // namespace
 
 Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards)
-       : httpd(LOCAL_DUMP_FILE_NAME, WIDTH, HEIGHT),
+       : httpd(WIDTH, HEIGHT),
          num_cards(num_cards),
          mixer_surface(create_surface(format)),
          h264_encoder_surface(create_surface(format)),
@@ -85,6 +102,7 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards)
          limiter(OUTPUT_FREQUENCY),
          compressor(OUTPUT_FREQUENCY)
 {
+       httpd.open_output_file(generate_local_dump_filename(/*frame=*/0).c_str());
        httpd.start(9095);
 
        CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF));
@@ -204,7 +222,7 @@ float find_peak(const float *samples, size_t num_samples)
 {
        float m = fabs(samples[0]);
        for (size_t i = 1; i < num_samples; ++i) {
-               m = std::max(m, fabs(samples[i]));
+               m = max(m, fabs(samples[i]));
        }
        return m;
 }
@@ -508,7 +526,7 @@ void Mixer::thread_func()
                }
 
                if (audio_level_callback != nullptr) {
-                       unique_lock<mutex> lock(r128_mutex);
+                       unique_lock<mutex> lock(compressor_mutex);
                        double loudness_s = r128.loudness_S();
                        double loudness_i = r128.integrated();
                        double loudness_range_low = r128.range_min();
@@ -516,7 +534,7 @@ void Mixer::thread_func()
 
                        audio_level_callback(loudness_s, 20.0 * log10(peak),
                                             loudness_i, loudness_range_low, loudness_range_high,
-                                            last_gain_staging_db);
+                                            gain_staging_db);
                }
 
                for (unsigned card_index = 1; card_index < num_cards; ++card_index) {
@@ -626,6 +644,15 @@ void Mixer::thread_func()
                //      chain->print_phase_timing();
                }
 
+               if (should_cut.exchange(false)) {  // Test and clear.
+                       string filename = generate_local_dump_filename(frame);
+                       printf("Starting new recording: %s\n", filename.c_str());
+                       h264_encoder->shutdown();
+                       httpd.close_output_file();
+                       httpd.open_output_file(filename.c_str());
+                       h264_encoder.reset(new H264Encoder(h264_encoder_surface, WIDTH, HEIGHT, &httpd));
+               }
+
 #if 0
                // Reset every 100 frames, so that local variations in frame times
                // (especially for the first few frames, when the shaders are
@@ -688,15 +715,23 @@ void Mixer::process_audio_one_frame(int64_t frame_pts_int, int num_samples)
        // 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.
-       float ref_level_dbfs = -14.0f;
        {
-               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);
-               last_gain_staging_db = 20.0 * log10(level_compressor.get_attenuation() * makeup_gain);
+               unique_lock<mutex> lock(level_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
@@ -752,7 +787,7 @@ void Mixer::process_audio_one_frame(int64_t frame_pts_int, int num_samples)
        deinterleave_samples(samples_out, &left, &right);
        float *ptrs[] = { left.data(), right.data() };
        {
-               unique_lock<mutex> lock(r128_mutex);
+               unique_lock<mutex> lock(compressor_mutex);
                r128.process(left.size(), ptrs);
        }