summaryrefslogtreecommitdiff
path: root/src/audio
diff options
context:
space:
mode:
Diffstat (limited to 'src/audio')
-rw-r--r--src/audio/CMakeLists.txt2
-rw-r--r--src/audio/audio_task.cpp12
-rw-r--r--src/audio/i2s_audio_output.cpp7
-rw-r--r--src/audio/include/fir.h131
-rw-r--r--src/audio/include/resample.hpp41
-rw-r--r--src/audio/include/sink_mixer.hpp31
-rw-r--r--src/audio/include/stream_info.hpp2
-rw-r--r--src/audio/resample.cpp260
-rw-r--r--src/audio/sink_mixer.cpp211
9 files changed, 519 insertions, 178 deletions
diff --git a/src/audio/CMakeLists.txt b/src/audio/CMakeLists.txt
index 9e50f8ff..ddfc7eb4 100644
--- a/src/audio/CMakeLists.txt
+++ b/src/audio/CMakeLists.txt
@@ -5,7 +5,7 @@
idf_component_register(
SRCS "audio_task.cpp" "chunk.cpp" "fatfs_audio_input.cpp"
"stream_message.cpp" "i2s_audio_output.cpp" "stream_buffer.cpp" "track_queue.cpp"
- "stream_event.cpp" "stream_info.cpp" "audio_fsm.cpp" "sink_mixer.cpp"
+ "stream_event.cpp" "stream_info.cpp" "audio_fsm.cpp" "sink_mixer.cpp" "resample.cpp"
INCLUDE_DIRS "include"
REQUIRES "codecs" "drivers" "cbor" "result" "tasks" "span" "memory" "tinyfsm" "database" "system_fsm" "playlist" "libsamplerate")
diff --git a/src/audio/audio_task.cpp b/src/audio/audio_task.cpp
index c3498965..d7c0c209 100644
--- a/src/audio/audio_task.cpp
+++ b/src/audio/audio_task.cpp
@@ -34,6 +34,7 @@
#include "freertos/queue.h"
#include "freertos/ringbuf.h"
#include "pipeline.hpp"
+#include "sample.hpp"
#include "sink_mixer.hpp"
#include "span.hpp"
@@ -225,7 +226,7 @@ auto AudioTask::BeginDecoding(InputStream& stream) -> bool {
codecs::ICodec::OutputFormat format = res.second.value();
StreamInfo::Pcm new_format{
.channels = format.num_channels,
- .bits_per_sample = format.bits_per_sample,
+ .bits_per_sample = 32,
.sample_rate = format.sample_rate_hz,
};
@@ -255,7 +256,8 @@ auto AudioTask::ContinueDecoding(InputStream& stream) -> bool {
while (!stream.data().empty()) {
OutputStream writer{codec_buffer_.get()};
- auto res = codec_->ContinueStream(stream.data(), writer.data());
+ auto res =
+ codec_->ContinueStream(stream.data(), writer.data_as<sample::Sample>());
stream.consume(res.first);
@@ -266,7 +268,7 @@ auto AudioTask::ContinueDecoding(InputStream& stream) -> bool {
return false;
}
} else {
- writer.add(res.second->bytes_written);
+ writer.add(res.second->samples_written * sizeof(sample::Sample));
InputStream reader{codec_buffer_.get()};
SendToSink(reader);
@@ -295,12 +297,12 @@ auto AudioTask::FinishDecoding(InputStream& stream) -> void {
InputStream padded_stream{mad_buffer.get()};
OutputStream writer{codec_buffer_.get()};
- auto res = codec_->ContinueStream(stream.data(), writer.data());
+ auto res = codec_->ContinueStream(stream.data(), writer.data_as<sample::Sample>());
if (res.second.has_error()) {
return;
}
- writer.add(res.second->bytes_written);
+ writer.add(res.second->samples_written * sizeof(sample::Sample));
InputStream reader{codec_buffer_.get()};
SendToSink(reader);
diff --git a/src/audio/i2s_audio_output.cpp b/src/audio/i2s_audio_output.cpp
index e53dbe2a..4eab3e02 100644
--- a/src/audio/i2s_audio_output.cpp
+++ b/src/audio/i2s_audio_output.cpp
@@ -117,11 +117,18 @@ auto I2SAudioOutput::AdjustVolumeDown() -> bool {
auto I2SAudioOutput::PrepareFormat(const StreamInfo::Pcm& orig)
-> StreamInfo::Pcm {
+ /*
return StreamInfo::Pcm{
.channels = std::min<uint8_t>(orig.channels, 2),
.bits_per_sample = std::clamp<uint8_t>(orig.bits_per_sample, 16, 32),
.sample_rate = std::clamp<uint32_t>(orig.sample_rate, 8000, 96000),
};
+ */
+ return StreamInfo::Pcm{
+ .channels = std::min<uint8_t>(orig.channels, 2),
+ .bits_per_sample = 16,
+ .sample_rate = 48000,
+ };
}
auto I2SAudioOutput::Configure(const StreamInfo::Pcm& pcm) -> void {
diff --git a/src/audio/include/fir.h b/src/audio/include/fir.h
new file mode 100644
index 00000000..e50c3eff
--- /dev/null
+++ b/src/audio/include/fir.h
@@ -0,0 +1,131 @@
+/*
+ * FIR filter coefficients from resample-1.x smallfilter.h
+ * see Digital Audio Resampling Home Page located at
+ * http://ccrma.stanford.edu/~jos/resample/
+ */
+32767, 32766, 32764, 32760, 32755, 32749, 32741, 32731, 32721, 32708,
+32695, 32679, 32663, 32645, 32625, 32604, 32582, 32558, 32533, 32506,
+32478, 32448, 32417, 32385, 32351, 32316, 32279, 32241, 32202, 32161,
+32119, 32075, 32030, 31984, 31936, 31887, 31836, 31784, 31731, 31676,
+31620, 31563, 31504, 31444, 31383, 31320, 31256, 31191, 31124, 31056,
+30987, 30916, 30845, 30771, 30697, 30621, 30544, 30466, 30387, 30306,
+30224, 30141, 30057, 29971, 29884, 29796, 29707, 29617, 29525, 29433,
+29339, 29244, 29148, 29050, 28952, 28852, 28752, 28650, 28547, 28443,
+28338, 28232, 28125, 28017, 27908, 27797, 27686, 27574, 27461, 27346,
+27231, 27115, 26998, 26879, 26760, 26640, 26519, 26398, 26275, 26151,
+26027, 25901, 25775, 25648, 25520, 25391, 25262, 25131, 25000, 24868,
+24735, 24602, 24467, 24332, 24197, 24060, 23923, 23785, 23647, 23507,
+23368, 23227, 23086, 22944, 22802, 22659, 22515, 22371, 22226, 22081,
+21935, 21789, 21642, 21494, 21346, 21198, 21049, 20900, 20750, 20600,
+20449, 20298, 20146, 19995, 19842, 19690, 19537, 19383, 19230, 19076,
+18922, 18767, 18612, 18457, 18302, 18146, 17990, 17834, 17678, 17521,
+17365, 17208, 17051, 16894, 16737, 16579, 16422, 16264, 16106, 15949,
+15791, 15633, 15475, 15317, 15159, 15001, 14843, 14685, 14527, 14369,
+14212, 14054, 13896, 13739, 13581, 13424, 13266, 13109, 12952, 12795,
+12639, 12482, 12326, 12170, 12014, 11858, 11703, 11548, 11393, 11238,
+11084, 10929, 10776, 10622, 10469, 10316, 10164, 10011, 9860, 9708,
+9557, 9407, 9256, 9106, 8957, 8808, 8659, 8511, 8364, 8216, 8070,
+7924, 7778, 7633, 7488, 7344, 7200, 7057, 6914, 6773, 6631, 6490,
+6350, 6210, 6071, 5933, 5795, 5658, 5521, 5385, 5250, 5115, 4981,
+4848, 4716, 4584, 4452, 4322, 4192, 4063, 3935, 3807, 3680, 3554,
+3429, 3304, 3180, 3057, 2935, 2813, 2692, 2572, 2453, 2335, 2217,
+2101, 1985, 1870, 1755, 1642, 1529, 1418, 1307, 1197, 1088, 979, 872,
+765, 660, 555, 451, 348, 246, 145, 44, -54, -153, -250, -347, -443,
+-537, -631, -724, -816, -908, -998, -1087, -1175, -1263, -1349, -1435,
+-1519, -1603, -1685, -1767, -1848, -1928, -2006, -2084, -2161, -2237,
+-2312, -2386, -2459, -2531, -2603, -2673, -2742, -2810, -2878, -2944,
+-3009, -3074, -3137, -3200, -3261, -3322, -3381, -3440, -3498, -3554,
+-3610, -3665, -3719, -3772, -3824, -3875, -3925, -3974, -4022, -4069,
+-4116, -4161, -4205, -4249, -4291, -4333, -4374, -4413, -4452, -4490,
+-4527, -4563, -4599, -4633, -4666, -4699, -4730, -4761, -4791, -4820,
+-4848, -4875, -4901, -4926, -4951, -4974, -4997, -5019, -5040, -5060,
+-5080, -5098, -5116, -5133, -5149, -5164, -5178, -5192, -5205, -5217,
+-5228, -5238, -5248, -5257, -5265, -5272, -5278, -5284, -5289, -5293,
+-5297, -5299, -5301, -5303, -5303, -5303, -5302, -5300, -5298, -5295,
+-5291, -5287, -5282, -5276, -5270, -5263, -5255, -5246, -5237, -5228,
+-5217, -5206, -5195, -5183, -5170, -5157, -5143, -5128, -5113, -5097,
+-5081, -5064, -5047, -5029, -5010, -4991, -4972, -4952, -4931, -4910,
+-4889, -4867, -4844, -4821, -4797, -4774, -4749, -4724, -4699, -4673,
+-4647, -4620, -4593, -4566, -4538, -4510, -4481, -4452, -4422, -4393,
+-4363, -4332, -4301, -4270, -4238, -4206, -4174, -4142, -4109, -4076,
+-4042, -4009, -3975, -3940, -3906, -3871, -3836, -3801, -3765, -3729,
+-3693, -3657, -3620, -3584, -3547, -3510, -3472, -3435, -3397, -3360,
+-3322, -3283, -3245, -3207, -3168, -3129, -3091, -3052, -3013, -2973,
+-2934, -2895, -2855, -2816, -2776, -2736, -2697, -2657, -2617, -2577,
+-2537, -2497, -2457, -2417, -2377, -2337, -2297, -2256, -2216, -2176,
+-2136, -2096, -2056, -2016, -1976, -1936, -1896, -1856, -1817, -1777,
+-1737, -1698, -1658, -1619, -1579, -1540, -1501, -1462, -1423, -1384,
+-1345, -1306, -1268, -1230, -1191, -1153, -1115, -1077, -1040, -1002,
+-965, -927, -890, -854, -817, -780, -744, -708, -672, -636, -600,
+-565, -530, -494, -460, -425, -391, -356, -322, -289, -255, -222,
+-189, -156, -123, -91, -59, -27, 4, 35, 66, 97, 127, 158, 188, 218,
+247, 277, 306, 334, 363, 391, 419, 447, 474, 501, 528, 554, 581, 606,
+632, 657, 683, 707, 732, 756, 780, 803, 827, 850, 872, 895, 917, 939,
+960, 981, 1002, 1023, 1043, 1063, 1082, 1102, 1121, 1139, 1158, 1176,
+1194, 1211, 1228, 1245, 1262, 1278, 1294, 1309, 1325, 1340, 1354,
+1369, 1383, 1397, 1410, 1423, 1436, 1448, 1461, 1473, 1484, 1496,
+1507, 1517, 1528, 1538, 1548, 1557, 1566, 1575, 1584, 1592, 1600,
+1608, 1616, 1623, 1630, 1636, 1643, 1649, 1654, 1660, 1665, 1670,
+1675, 1679, 1683, 1687, 1690, 1694, 1697, 1700, 1702, 1704, 1706,
+1708, 1709, 1711, 1712, 1712, 1713, 1713, 1713, 1713, 1712, 1711,
+1710, 1709, 1708, 1706, 1704, 1702, 1700, 1697, 1694, 1691, 1688,
+1685, 1681, 1677, 1673, 1669, 1664, 1660, 1655, 1650, 1644, 1639,
+1633, 1627, 1621, 1615, 1609, 1602, 1596, 1589, 1582, 1575, 1567,
+1560, 1552, 1544, 1536, 1528, 1520, 1511, 1503, 1494, 1485, 1476,
+1467, 1458, 1448, 1439, 1429, 1419, 1409, 1399, 1389, 1379, 1368,
+1358, 1347, 1337, 1326, 1315, 1304, 1293, 1282, 1271, 1260, 1248,
+1237, 1225, 1213, 1202, 1190, 1178, 1166, 1154, 1142, 1130, 1118,
+1106, 1094, 1081, 1069, 1057, 1044, 1032, 1019, 1007, 994, 981, 969,
+956, 943, 931, 918, 905, 892, 879, 867, 854, 841, 828, 815, 802, 790,
+777, 764, 751, 738, 725, 713, 700, 687, 674, 662, 649, 636, 623, 611,
+598, 585, 573, 560, 548, 535, 523, 510, 498, 486, 473, 461, 449, 437,
+425, 413, 401, 389, 377, 365, 353, 341, 330, 318, 307, 295, 284, 272,
+261, 250, 239, 228, 217, 206, 195, 184, 173, 163, 152, 141, 131, 121,
+110, 100, 90, 80, 70, 60, 51, 41, 31, 22, 12, 3, -5, -14, -23, -32,
+-41, -50, -59, -67, -76, -84, -93, -101, -109, -117, -125, -133, -140,
+-148, -156, -163, -170, -178, -185, -192, -199, -206, -212, -219,
+-226, -232, -239, -245, -251, -257, -263, -269, -275, -280, -286,
+-291, -297, -302, -307, -312, -317, -322, -327, -332, -336, -341,
+-345, -349, -354, -358, -362, -366, -369, -373, -377, -380, -384,
+-387, -390, -394, -397, -400, -402, -405, -408, -411, -413, -416,
+-418, -420, -422, -424, -426, -428, -430, -432, -433, -435, -436,
+-438, -439, -440, -442, -443, -444, -445, -445, -446, -447, -447,
+-448, -448, -449, -449, -449, -449, -449, -449, -449, -449, -449,
+-449, -449, -448, -448, -447, -447, -446, -445, -444, -443, -443,
+-442, -441, -440, -438, -437, -436, -435, -433, -432, -430, -429,
+-427, -426, -424, -422, -420, -419, -417, -415, -413, -411, -409,
+-407, -405, -403, -400, -398, -396, -393, -391, -389, -386, -384,
+-381, -379, -376, -374, -371, -368, -366, -363, -360, -357, -355,
+-352, -349, -346, -343, -340, -337, -334, -331, -328, -325, -322,
+-319, -316, -313, -310, -307, -304, -301, -298, -294, -291, -288,
+-285, -282, -278, -275, -272, -269, -265, -262, -259, -256, -252,
+-249, -246, -243, -239, -236, -233, -230, -226, -223, -220, -217,
+-213, -210, -207, -204, -200, -197, -194, -191, -187, -184, -181,
+-178, -175, -172, -168, -165, -162, -159, -156, -153, -150, -147,
+-143, -140, -137, -134, -131, -128, -125, -122, -120, -117, -114,
+-111, -108, -105, -102, -99, -97, -94, -91, -88, -86, -83, -80, -78,
+-75, -72, -70, -67, -65, -62, -59, -57, -55, -52, -50, -47, -45, -43,
+-40, -38, -36, -33, -31, -29, -27, -25, -22, -20, -18, -16, -14, -12,
+-10, -8, -6, -4, -2, 0, 0, 2, 4, 6, 8, 9, 11, 13, 14, 16, 17, 19, 21,
+22, 24, 25, 27, 28, 29, 31, 32, 33, 35, 36, 37, 38, 40, 41, 42, 43,
+44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 56, 57, 58, 59,
+59, 60, 61, 62, 62, 63, 63, 64, 64, 65, 66, 66, 66, 67, 67, 68, 68,
+69, 69, 69, 70, 70, 70, 70, 71, 71, 71, 71, 71, 72, 72, 72, 72, 72,
+72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72, 72,
+72, 71, 71, 71, 71, 71, 70, 70, 70, 70, 69, 69, 69, 69, 68, 68, 68,
+67, 67, 67, 66, 66, 66, 65, 65, 64, 64, 64, 63, 63, 62, 62, 62, 61,
+61, 60, 60, 59, 59, 58, 58, 58, 57, 57, 56, 56, 55, 55, 54, 54, 53,
+53, 52, 52, 51, 51, 50, 50, 49, 48, 48, 47, 47, 46, 46, 45, 45, 44,
+44, 43, 43, 42, 42, 41, 41, 40, 39, 39, 38, 38, 37, 37, 36, 36, 35,
+35, 34, 34, 33, 33, 32, 32, 31, 31, 30, 30, 29, 29, 28, 28, 27, 27,
+26, 26, 25, 25, 24, 24, 23, 23, 23, 22, 22, 21, 21, 20, 20, 20, 19,
+19, 18, 18, 17, 17, 17, 16, 16, 15, 15, 15, 14, 14, 14, 13, 13, 12,
+12, 12, 11, 11, 11, 10, 10, 10, 9, 9, 9, 9, 8, 8, 8, 7, 7, 7, 7, 6, 6,
+6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 1, 1, 1,
+1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, -1, -1, -1,
+-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -2, -2, -2, -2, -2, -2, -2,
+-2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2,
+-2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2,
+-2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2, -2,
+-2, -2, -2, -2, -2, -2, -2, -2, -1, -1, -1, -1, -1, -1, -1, -1, -1,
+-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
diff --git a/src/audio/include/resample.hpp b/src/audio/include/resample.hpp
new file mode 100644
index 00000000..d7933470
--- /dev/null
+++ b/src/audio/include/resample.hpp
@@ -0,0 +1,41 @@
+#pragma once
+
+#include <sys/_stdint.h>
+#include <vector>
+
+#include "span.hpp"
+
+#include "sample.hpp"
+
+namespace audio {
+
+class Channel;
+
+class Resampler {
+ public:
+ Resampler(uint32_t source_sample_rate,
+ uint32_t target_sample_rate,
+ uint8_t num_channels);
+
+ ~Resampler();
+
+ auto source_sample_rate() -> uint32_t { return source_sample_rate_; }
+ auto target_sample_rate() -> uint32_t { return target_sample_rate_; }
+ auto channels() -> uint_fast8_t { return num_channels_; }
+
+ auto Process(cpp::span<const sample::Sample> input,
+ cpp::span<sample::Sample> output,
+ bool end_of_data) -> std::pair<size_t,size_t>;
+
+ private:
+ auto ApplyDither(cpp::span<sample::Sample>) -> void;
+
+ uint32_t source_sample_rate_;
+ uint32_t target_sample_rate_;
+ uint32_t factor_;
+
+ uint8_t num_channels_;
+ std::vector<Channel> channels_;
+};
+
+} // namespace audio \ No newline at end of file
diff --git a/src/audio/include/sink_mixer.hpp b/src/audio/include/sink_mixer.hpp
index 632ffa2e..1bf12016 100644
--- a/src/audio/include/sink_mixer.hpp
+++ b/src/audio/include/sink_mixer.hpp
@@ -10,6 +10,8 @@
#include <cstdint>
#include <memory>
+#include "resample.hpp"
+#include "sample.hpp"
#include "samplerate.h"
#include "audio_decoder.hpp"
@@ -38,12 +40,10 @@ class SinkMixer {
auto SetTargetFormat(const StreamInfo::Pcm& format) -> void;
auto HandleBytes() -> void;
- template <typename T>
- auto ConvertFixedToFloating(InputStream&, OutputStream&) -> void;
- auto Resample(float, int, InputStream&, OutputStream&) -> void;
- template <typename T>
- auto Quantise(InputStream&) -> std::size_t;
-
+ auto Resample(InputStream&, OutputStream&) -> bool;
+ auto ApplyDither(cpp::span<sample::Sample> samples, uint_fast8_t bits) -> void;
+ auto Downscale(cpp::span<sample::Sample>, cpp::span<int16_t>) -> void;
+
enum class Command {
kReadBytes,
kSetSourceFormat,
@@ -58,31 +58,14 @@ class SinkMixer {
QueueHandle_t commands_;
SemaphoreHandle_t is_idle_;
- SRC_STATE* resampler_;
+ std::unique_ptr<Resampler> resampler_;
std::unique_ptr<RawStream> input_stream_;
- std::unique_ptr<RawStream> floating_point_stream_;
std::unique_ptr<RawStream> resampled_stream_;
- cpp::span<std::byte> quantisation_buffer_;
- cpp::span<short> quantisation_buffer_as_shorts_;
- cpp::span<int> quantisation_buffer_as_ints_;
-
StreamInfo::Pcm target_format_;
StreamBufferHandle_t source_;
StreamBufferHandle_t sink_;
};
-template <>
-auto SinkMixer::ConvertFixedToFloating<short>(InputStream&, OutputStream&)
- -> void;
-template <>
-auto SinkMixer::ConvertFixedToFloating<int>(InputStream&, OutputStream&)
- -> void;
-
-template <>
-auto SinkMixer::Quantise<short>(InputStream&) -> std::size_t;
-template <>
-auto SinkMixer::Quantise<int>(InputStream&) -> std::size_t;
-
} // namespace audio
diff --git a/src/audio/include/stream_info.hpp b/src/audio/include/stream_info.hpp
index 7cf9e847..01dd282a 100644
--- a/src/audio/include/stream_info.hpp
+++ b/src/audio/include/stream_info.hpp
@@ -77,7 +77,7 @@ class StreamInfo {
// The sample rate.
uint32_t sample_rate;
- auto real_bytes_per_sample() const -> uint8_t {
+ auto bytes_per_sample() const -> uint8_t {
return bits_per_sample == 16 ? 2 : 4;
}
diff --git a/src/audio/resample.cpp b/src/audio/resample.cpp
new file mode 100644
index 00000000..93ea1034
--- /dev/null
+++ b/src/audio/resample.cpp
@@ -0,0 +1,260 @@
+#include "resample.hpp"
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <algorithm>
+#include <numeric>
+
+#include "esp_log.h"
+
+#include "sample.hpp"
+#include "stream_info.hpp"
+
+namespace audio {
+
+static constexpr size_t kFilterSize = 1536;
+
+constexpr auto calc_deltas(const std::array<int32_t, kFilterSize>& filter)
+ -> std::array<int32_t, kFilterSize> {
+ std::array<int32_t, kFilterSize> deltas;
+ for (size_t n = 0; n < kFilterSize - 1; n++)
+ deltas[n] = filter[n + 1] - filter[n];
+ return deltas;
+}
+
+static const std::array<int32_t, kFilterSize> kFilter{
+#include "fir.h"
+};
+
+static const std::array<int32_t, kFilterSize> kFilterDeltas =
+ calc_deltas(kFilter);
+
+class Channel {
+ public:
+ Channel(uint32_t src_rate,
+ uint32_t dest_rate,
+ size_t chunk_size,
+ size_t skip);
+ ~Channel();
+
+ auto output_chunk_size() -> size_t { return output_chunk_size_; }
+
+ auto FlushSamples(cpp::span<sample::Sample> out) -> size_t;
+ auto AddSample(sample::Sample, cpp::span<sample::Sample> out) -> std::size_t;
+ auto ApplyFilter() -> sample::Sample;
+
+ private:
+ size_t output_chunk_size_;
+ size_t skip_;
+
+ uint32_t factor_; /* factor */
+
+ uint32_t time_; /* time */
+
+ uint32_t time_per_filter_iteration_; /* output step */
+ uint32_t filter_step_; /* filter step */
+ uint32_t filter_end_; /* filter end */
+
+ int32_t unity_scale_; /* unity scale */
+
+ int32_t samples_per_filter_wing_; /* extra samples */
+ int32_t latest_sample_; /* buffer index */
+ cpp::span<int32_t> sample_buffer_; /* the buffer */
+};
+
+enum {
+ Nl = 8, /* 2^Nl samples per zero crossing in fir */
+ Nη = 8, /* phase bits for filter interpolation */
+ kPhaseBits = Nl + Nη, /* phase bits (fract of fixed point) */
+ One = 1 << kPhaseBits,
+};
+
+Channel::Channel(uint32_t irate, uint32_t orate, size_t count, size_t skip)
+ : skip_(skip) {
+ factor_ = ((uint64_t)orate << kPhaseBits) / irate;
+ if (factor_ != One) {
+ time_per_filter_iteration_ = ((uint64_t)irate << kPhaseBits) / orate;
+ filter_step_ = 1 << (Nl + Nη);
+ filter_end_ = kFilterSize << Nη;
+ samples_per_filter_wing_ = 1 + (filter_end_ / filter_step_);
+ unity_scale_ = 13128; /* unity scale factor for fir */
+ if (factor_ < One) {
+ unity_scale_ *= factor_;
+ unity_scale_ >>= kPhaseBits;
+ filter_step_ *= factor_;
+ filter_step_ >>= kPhaseBits;
+ samples_per_filter_wing_ *= time_per_filter_iteration_;
+ samples_per_filter_wing_ >>= kPhaseBits;
+ }
+ latest_sample_ = samples_per_filter_wing_;
+ time_ = latest_sample_ << kPhaseBits;
+
+ size_t buf_size = samples_per_filter_wing_ * 2 + count;
+ int32_t* buf = new int32_t[buf_size];
+ sample_buffer_ = {buf, buf_size};
+ count += buf_size; /* account for buffer accumulation */
+ }
+ output_chunk_size_ = ((uint64_t)count * factor_) >> kPhaseBits;
+}
+
+Channel::~Channel() {
+ delete sample_buffer_.data();
+}
+
+auto Channel::ApplyFilter() -> sample::Sample {
+ uint32_t iteration, p, i;
+ int32_t *sample, a;
+
+ int64_t value = 0;
+
+ // I did my best, but I'll be honest with you I've no idea about any of this
+ // maths stuff.
+
+ // Left wing of the filter.
+ sample = &sample_buffer_[time_ >> kPhaseBits];
+ p = time_ & ((1 << kPhaseBits) - 1);
+ iteration = factor_ < One ? (factor_ * p) >> kPhaseBits : p;
+ while (iteration < filter_end_) {
+ i = iteration >> Nη;
+ a = iteration & ((1 << Nη) - 1);
+ iteration += filter_step_;
+ a *= kFilterDeltas[i];
+ a >>= Nη;
+ a += kFilter[i];
+ value += static_cast<int64_t>(*--sample) * a;
+ }
+
+ // Right wing of the filter.
+ sample = &sample_buffer_[time_ >> kPhaseBits];
+ p = (One - p) & ((1 << kPhaseBits) - 1);
+ iteration = factor_ < One ? (factor_ * p) >> kPhaseBits : p;
+ if (p == 0) /* skip h[0] as it was already been summed above if p == 0 */
+ iteration += filter_step_;
+ while (iteration < filter_end_) {
+ i = iteration >> Nη;
+ a = iteration & ((1 << Nη) - 1);
+ iteration += filter_step_;
+ a *= kFilterDeltas[i];
+ a >>= Nη;
+ a += kFilter[i];
+ value += static_cast<int64_t>(*sample++) * a;
+ }
+
+ /* scale */
+ value >>= 2;
+ value *= unity_scale_;
+ value >>= 27;
+
+ return sample::Clip(value);
+}
+
+auto Channel::FlushSamples(cpp::span<sample::Sample> out) -> size_t {
+ size_t zeroes_needed = (2 * samples_per_filter_wing_) - latest_sample_;
+ size_t produced = 0;
+ while (zeroes_needed > 0) {
+ produced += AddSample(0, out.subspan(produced));
+ zeroes_needed--;
+ }
+ return produced;
+}
+
+auto Channel::AddSample(sample::Sample in, cpp::span<sample::Sample> out)
+ -> size_t {
+ // Add the latest sample to our working buffer.
+ sample_buffer_[latest_sample_++] = in;
+
+ // If we don't have enough samples to run the filter, then bail out and wait
+ // for more.
+ if (latest_sample_ < 2 * samples_per_filter_wing_) {
+ return 0;
+ }
+
+ // Apply the filter to the buffered samples. First, we work out how long (in
+ // samples) we can run the filter for before running out. This isn't as
+ // trivial as it might look; e.g. depending on the resampling factor we might
+ // be doubling the number of samples, or halving them.
+ uint32_t max_time = (latest_sample_ - samples_per_filter_wing_) << kPhaseBits;
+ size_t samples_output = 0;
+ while (time_ < max_time) {
+ out[skip_ * samples_output++] = ApplyFilter();
+ time_ += time_per_filter_iteration_;
+ }
+
+ // If we are approaching the end of our buffer, we need to shift all the data
+ // in it down to the front to make room for more samples.
+ int32_t current_sample = time_ >> kPhaseBits;
+ if (current_sample >= (sample_buffer_.size() - samples_per_filter_wing_)) {
+ // NB: bit shifting back and forth means we're only modifying `time` by
+ // whole samples.
+ time_ -= current_sample << kPhaseBits;
+ time_ += samples_per_filter_wing_ << kPhaseBits;
+
+ int32_t new_current_sample = time_ >> kPhaseBits;
+ new_current_sample -= samples_per_filter_wing_;
+ current_sample -= samples_per_filter_wing_;
+
+ int32_t samples_to_move = latest_sample_ - current_sample;
+ if (samples_to_move > 0) {
+ auto samples = sample_buffer_.subspan(current_sample, samples_to_move);
+ std::copy_backward(samples.begin(), samples.end(),
+ sample_buffer_.first(new_current_sample).end());
+ latest_sample_ = new_current_sample + samples_to_move;
+ } else {
+ latest_sample_ = new_current_sample;
+ }
+ }
+
+ return samples_output;
+}
+
+static const size_t kChunkSizeSamples = 256;
+
+Resampler::Resampler(uint32_t source_sample_rate,
+ uint32_t target_sample_rate,
+ uint8_t num_channels)
+ : source_sample_rate_(source_sample_rate),
+ target_sample_rate_(target_sample_rate),
+ factor_(((uint64_t)target_sample_rate << kPhaseBits) /
+ source_sample_rate),
+ num_channels_(num_channels),
+ channels_() {
+ for (int i = 0; i < num_channels; i++) {
+ channels_.emplace_back(source_sample_rate, target_sample_rate,
+ kChunkSizeSamples, num_channels);
+ }
+}
+
+Resampler::~Resampler() {}
+
+auto Resampler::Process(cpp::span<const sample::Sample> input,
+ cpp::span<sample::Sample> output,
+ bool end_of_data) -> std::pair<size_t, size_t> {
+ size_t samples_used = 0;
+ std::vector<size_t> samples_produced = {num_channels_, 0};
+ size_t total_samples_produced = 0;
+
+ size_t slop = (factor_ >> kPhaseBits) + 1;
+
+ uint_fast8_t cur_channel = 0;
+
+ while (input.size() > samples_used &&
+ output.size() > total_samples_produced + slop) {
+ // Work out where the next set of samples should be placed.
+ size_t next_output_index =
+ (samples_produced[cur_channel] * num_channels_) + cur_channel;
+
+ // Generate the next samples
+ size_t new_samples = channels_[cur_channel].AddSample(
+ input[samples_used++], output.subspan(next_output_index));
+
+ samples_produced[cur_channel] += new_samples;
+ total_samples_produced += new_samples;
+
+ cur_channel = (cur_channel + 1) % num_channels_;
+ }
+
+ return {samples_used, total_samples_produced};
+}
+
+} // namespace audio
diff --git a/src/audio/sink_mixer.cpp b/src/audio/sink_mixer.cpp
index 79e6f3d3..ba306626 100644
--- a/src/audio/sink_mixer.cpp
+++ b/src/audio/sink_mixer.cpp
@@ -13,6 +13,8 @@
#include "esp_log.h"
#include "freertos/portmacro.h"
#include "freertos/projdefs.h"
+#include "resample.hpp"
+#include "sample.hpp"
#include "samplerate.h"
#include "stream_info.hpp"
@@ -21,10 +23,7 @@
static constexpr char kTag[] = "mixer";
static constexpr std::size_t kSourceBufferLength = 2 * 1024;
-static constexpr std::size_t kInputBufferLength = 2 * 1024;
-static constexpr std::size_t kReformatBufferLength = 8 * 1024;
-static constexpr std::size_t kResampleBufferLength = kReformatBufferLength;
-static constexpr std::size_t kQuantisedBufferLength = 1 * 1024;
+static constexpr std::size_t kSampleBufferLength = 4 * 1024;
namespace audio {
@@ -34,20 +33,8 @@ SinkMixer::SinkMixer(StreamBufferHandle_t dest)
resampler_(nullptr),
source_(xStreamBufferCreate(kSourceBufferLength, 1)),
sink_(dest) {
- input_stream_.reset(new RawStream(kInputBufferLength));
- floating_point_stream_.reset(new RawStream(kReformatBufferLength, MALLOC_CAP_SPIRAM));
- resampled_stream_.reset(new RawStream(kResampleBufferLength, MALLOC_CAP_SPIRAM));
-
- quantisation_buffer_ = {
- reinterpret_cast<std::byte*>(heap_caps_malloc(
- kQuantisedBufferLength, MALLOC_CAP_INTERNAL | MALLOC_CAP_8BIT)),
- kQuantisedBufferLength};
- quantisation_buffer_as_ints_ = {
- reinterpret_cast<int*>(quantisation_buffer_.data()),
- quantisation_buffer_.size_bytes() / 4};
- quantisation_buffer_as_shorts_ = {
- reinterpret_cast<short*>(quantisation_buffer_.data()),
- quantisation_buffer_.size_bytes() / 2};
+ input_stream_.reset(new RawStream(kSampleBufferLength));
+ resampled_stream_.reset(new RawStream(kSampleBufferLength));
tasks::StartPersistent<tasks::Type::kMixer>([&]() { Main(); });
}
@@ -56,10 +43,6 @@ SinkMixer::~SinkMixer() {
vQueueDelete(commands_);
vSemaphoreDelete(is_idle_);
vStreamBufferDelete(source_);
- heap_caps_free(quantisation_buffer_.data());
- if (resampler_ != nullptr) {
- src_delete(resampler_);
- }
}
auto SinkMixer::MixAndSend(InputStream& input, const StreamInfo::Pcm& target)
@@ -109,10 +92,12 @@ auto SinkMixer::Main() -> void {
case Command::kSetSourceFormat:
ESP_LOGI(kTag, "setting source format");
input_receiver.prepare(args.format, {});
+ resampler_.reset();
break;
case Command::kSetTargetFormat:
ESP_LOGI(kTag, "setting target format");
target_format_ = args.format;
+ resampler_.reset();
break;
case Command::kReadBytes:
xSemaphoreTake(is_idle_, 0);
@@ -150,152 +135,84 @@ auto SinkMixer::HandleBytes() -> void {
return;
}
- // Work out the resampling ratio using floating point arithmetic, since
- // relying on the FPU for this will be much faster, and the difference in
- // accuracy is unlikely to be noticeable.
- float src_ratio = static_cast<float>(target_format_.sample_rate) /
- static_cast<float>(pcm->sample_rate);
-
- // Loop until we don't have any complete frames left in the input stream,
- // where a 'frame' is one complete sample per channel.
while (!input_stream_->empty()) {
- // The first step of both resampling and requantising is to convert the
- // fixed point pcm input data into 32 bit floating point samples.
- OutputStream floating_writer{floating_point_stream_.get()};
- if (pcm->bits_per_sample == 16) {
- ConvertFixedToFloating<short>(input, floating_writer);
+ RawStream* output_source;
+ if (pcm->sample_rate != target_format_.sample_rate) {
+ OutputStream resampled_writer{resampled_stream_.get()};
+ if (Resample(input, resampled_writer)) {
+ // Zero samples used or written. We need more input.
+ break;
+ }
+ output_source = resampled_stream_.get();
} else {
- // FIXME: We should consider treating 24 bit and 32 bit samples
- // differently.
- ConvertFixedToFloating<int>(input, floating_writer);
+ output_source = input_stream_.get();
}
- InputStream floating_reader{floating_point_stream_.get()};
+ if (target_format_.bits_per_sample == 16) {
+ // This is slightly scary; we're basically reaching into the internals of
+ // the stream buffer to do in-place conversion of samples. Saving an
+ // extra buffer + copy into that buffer is certainly worth it however.
+ cpp::span<sample::Sample> src =
+ output_source->data_as<sample::Sample>().first(
+ output_source->info().bytes_in_stream() / sizeof(sample::Sample));
+ cpp::span<int16_t> dest = output_source->data_as<int16_t>().first(
+ output_source->info().bytes_in_stream() / sizeof(int16_t));
- while (!floating_point_stream_->empty()) {
- RawStream* quantisation_source;
- if (pcm->sample_rate != target_format_.sample_rate) {
- // The input data needs to be resampled before being sent to the sink.
- OutputStream resample_writer{resampled_stream_.get()};
- Resample(src_ratio, pcm->channels, floating_reader, resample_writer);
- quantisation_source = resampled_stream_.get();
- } else {
- // The input data already has an acceptable sample rate. All we need to
- // do is quantise it.
- quantisation_source = floating_point_stream_.get();
- }
+ ApplyDither(src, 16);
+ Downscale(src, dest);
- InputStream quantise_reader{quantisation_source};
- while (!quantisation_source->empty()) {
- std::size_t samples_available;
- if (target_format_.bits_per_sample == 16) {
- samples_available = Quantise<short>(quantise_reader);
- } else {
- samples_available = Quantise<int>(quantise_reader);
- }
+ output_source->info().bytes_in_stream() /= 2;
+ }
- assert(samples_available * target_format_.real_bytes_per_sample() <=
- quantisation_buffer_.size_bytes());
+ InputStream output{output_source};
+ cpp::span<const std::byte> buf = output.data();
- std::size_t bytes_sent = xStreamBufferSend(
- sink_, quantisation_buffer_.data(),
- samples_available * target_format_.real_bytes_per_sample(),
- portMAX_DELAY);
- assert(bytes_sent ==
- samples_available * target_format_.real_bytes_per_sample());
- }
+ size_t bytes_sent = 0;
+ while (bytes_sent < buf.size_bytes()) {
+ auto cropped = buf.subspan(bytes_sent);
+ bytes_sent += xStreamBufferSend(sink_, cropped.data(),
+ cropped.size_bytes(), portMAX_DELAY);
}
+ output.consume(bytes_sent);
}
}
-template <>
-auto SinkMixer::ConvertFixedToFloating<short>(InputStream& in_str,
- OutputStream& out_str) -> void {
- auto in = in_str.data_as<short>();
- auto out = out_str.data_as<float>();
- std::size_t samples_converted = std::min(in.size(), out.size());
+auto SinkMixer::Resample(InputStream& in, OutputStream& out) -> bool {
+ if (resampler_ == nullptr) {
+ ESP_LOGI(kTag, "creating new resampler");
+ auto format = in.info().format_as<StreamInfo::Pcm>();
+ resampler_.reset(new Resampler(
+ format->sample_rate, target_format_.sample_rate, format->channels));
+ }
- src_short_to_float_array(in.data(), out.data(), samples_converted);
+ auto res = resampler_->Process(in.data_as<sample::Sample>(),
+ out.data_as<sample::Sample>(), false);
- in_str.consume(samples_converted * sizeof(short));
- out_str.add(samples_converted * sizeof(float));
-}
+ ESP_LOGI(kTag, "resampler sent %u samples, consumed %u, produced %u",
+ in.data().size(), res.first, res.second);
-template <>
-auto SinkMixer::ConvertFixedToFloating<int>(InputStream& in_str,
- OutputStream& out_str) -> void {
- auto in = in_str.data_as<int>();
- auto out = out_str.data_as<float>();
- std::size_t samples_converted = std::min(in.size(), out.size());
+ in.consume(res.first * sizeof(sample::Sample));
+ out.add(res.first * sizeof(sample::Sample));
- src_int_to_float_array(in.data(), out.data(), samples_converted);
-
- in_str.consume(samples_converted * sizeof(int));
- out_str.add(samples_converted * sizeof(float));
+ return res.first == 0 && res.second == 0;
}
-auto SinkMixer::Resample(float src_ratio,
- int channels,
- InputStream& in,
- OutputStream& out) -> void {
- if (resampler_ == nullptr || src_get_channels(resampler_) != channels) {
- if (resampler_ != nullptr) {
- src_delete(resampler_);
- }
-
- ESP_LOGI(kTag, "creating new resampler with %u channels", channels);
-
- int err = 0;
- resampler_ = src_new(SRC_LINEAR, channels, &err);
- assert(resampler_ != NULL);
- assert(err == 0);
+auto SinkMixer::Downscale(cpp::span<sample::Sample> samples,
+ cpp::span<int16_t> output) -> void {
+ for (size_t i = 0; i < samples.size(); i++) {
+ output[i] = sample::ToSigned16Bit(samples[i]);
}
-
- auto in_buf = in.data_as<float>();
- auto out_buf = out.data_as<float>();
-
- src_set_ratio(resampler_, src_ratio);
- SRC_DATA args{
- .data_in = in_buf.data(),
- .data_out = out_buf.data(),
- .input_frames = static_cast<long>(in_buf.size()),
- .output_frames = static_cast<long>(out_buf.size()),
- .input_frames_used = 0,
- .output_frames_gen = 0,
- .end_of_input = 0,
- .src_ratio = src_ratio,
- };
- int err = src_process(resampler_, &args);
- if (err != 0) {
- ESP_LOGE(kTag, "resampler error: %s", src_strerror(err));
- }
-
- in.consume(args.input_frames_used * sizeof(float));
- out.add(args.output_frames_gen * sizeof(float));
}
-template <>
-auto SinkMixer::Quantise<short>(InputStream& in) -> std::size_t {
- auto src = in.data_as<float>();
- cpp::span<short> dest = quantisation_buffer_as_shorts_;
- dest = dest.first(std::min(src.size(), dest.size()));
-
- src_float_to_short_array(src.data(), dest.data(), dest.size());
-
- in.consume(dest.size() * sizeof(float));
- return dest.size();
-}
-
-template <>
-auto SinkMixer::Quantise<int>(InputStream& in) -> std::size_t {
- auto src = in.data_as<float>();
- cpp::span<int> dest = quantisation_buffer_as_ints_;
- dest = dest.first(std::min<int>(src.size(), dest.size()));
-
- src_float_to_int_array(src.data(), dest.data(), dest.size());
-
- in.consume(dest.size() * sizeof(float));
- return dest.size();
+auto SinkMixer::ApplyDither(cpp::span<sample::Sample> samples,
+ uint_fast8_t bits) -> void {
+ static uint32_t prnd;
+ for (auto& s : samples) {
+ prnd = (prnd * 0x19660dL + 0x3c6ef35fL) & 0xffffffffL;
+ s = sample::Clip(
+ static_cast<int64_t>(s) +
+ (static_cast<int>(prnd) >> (sizeof(sample::Sample) - bits)));
+ }
}
} // namespace audio
generated by cgit v1.2.3 (git 2.25.1) at 2026-04-29 22:49:07 +0000