4 // Takes in samples from an input source, possibly with jitter, and outputs a fixed number
5 // of samples every iteration. Used to a) change sample rates if needed, and b) deal with
6 // input sources that don't have audio locked to video. For every input video
7 // frame, you call add_input_samples() with the pts (measured in seconds) of the video frame,
8 // taken to be the start point of the frame's audio. When you want to _output_ a finished
9 // frame with audio, you get_output_samples() with the number of samples you want, and will
10 // get exactly that number of samples back. If the input and output clocks are not in sync,
11 // the audio will be stretched for you. (If they are _very_ out of sync, this will come through
12 // as a pitch shift.) Of course, the process introduces some delay; you specify a target delay
13 // (typically measured in milliseconds, although more is fine) and the algorithm works to
14 // provide exactly that.
16 // A/V sync is a much harder problem than one would intuitively assume. This implementation
17 // is based on a 2012 paper by Fons Adriaensen, “Controlling adaptive resampling”
18 // (http://kokkinizita.linuxaudio.org/papers/adapt-resamp.pdf). The paper gives an algorithm
19 // that converges to jitter of <100 ns; the basic idea is to measure the _rate_ the input
20 // queue fills and is drained (as opposed to the length of the queue itself), and smoothly
21 // adjust the resampling rate so that it reaches steady state at the desired delay.
23 // Parts of the code is adapted from Adriaensen's project Zita-ajbridge (based on the same
24 // algorithm), although it has been heavily reworked for this use case. Original copyright follows:
26 // Copyright (C) 2012-2015 Fons Adriaensen <fons@linuxaudio.org>
28 // This program is free software; you can redistribute it and/or modify
29 // it under the terms of the GNU General Public License as published by
30 // the Free Software Foundation; either version 3 of the License, or
31 // (at your option) any later version.
33 // This program is distributed in the hope that it will be useful,
34 // but WITHOUT ANY WARRANTY; without even the implied warranty of
35 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
36 // GNU General Public License for more details.
38 // You should have received a copy of the GNU General Public License
39 // along with this program. If not, see <http://www.gnu.org/licenses/>.
43 #include <zita-resampler/vresampler.h>
49 Resampler(unsigned freq_in, unsigned freq_out, unsigned num_channels = 2);
51 // Note: pts is always in seconds.
52 void add_input_samples(double pts, const float *samples, ssize_t num_samples);
53 bool get_output_samples(double pts, float *samples, ssize_t num_samples); // Returns false if underrun.
56 void init_loop_filter(double bandwidth_hz);
58 VResampler vresampler;
60 unsigned freq_in, freq_out, num_channels;
62 bool first_input = true, first_output = true;
63 double last_input_pts; // Start of last input block, in seconds.
64 double last_output_pts;
66 ssize_t k_a0 = 0; // Total amount of samples inserted _before_ the last call to add_input_samples().
67 ssize_t k_a1 = 0; // Total amount of samples inserted _after_ the last call to add_input_samples().
69 ssize_t total_consumed_samples = 0;
71 // Duration of last input block, in seconds.
72 double last_input_len;
74 // Filter state for the loop filter.
75 double z1 = 0.0, z2 = 0.0, z3 = 0.0;
77 // Ratio between the two frequencies.
80 // How much delay we are expected to have, in input samples.
81 // If actual delay drifts too much away from this, we will start
82 // changing the resampling ratio to compensate.
83 double expected_delay = 4800.0;
85 // Input samples not yet fed into the resampler.
86 // TODO: Use a circular buffer instead, for efficiency.
87 std::deque<float> buffer;
90 #endif // !defined(_RESAMPLER_H)