red: make red encoder more generic

potentially allowing distances of more than 2.

BUG=webrtc:11640

Change-Id: I0d8c831218285d57cf07f0a8e5829810afd4ab3f
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/188383
Reviewed-by: Jesus de Vicente Pena <devicentepena@webrtc.org>
Reviewed-by: Henrik Lundin <henrik.lundin@webrtc.org>
Commit-Queue: Jesus de Vicente Pena <devicentepena@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#33913}
This commit is contained in:
Philipp Hancke
2021-04-23 10:15:42 +02:00
committed by WebRTC LUCI CQ
parent 3371638229
commit b3bb13c572
3 changed files with 100 additions and 102 deletions

View File

@ -19,10 +19,18 @@
#include "rtc_base/checks.h"
namespace webrtc {
// RED packets must be less than 1024 bytes to fit the 10 bit block length.
static constexpr const int kRedMaxPacketSize = 1 << 10;
// The typical MTU is 1200 bytes.
static constexpr const size_t kAudioMaxRtpPacketLen = 1200;
static constexpr const int kRedMaxPacketSize =
1 << 10; // RED packets must be less than 1024 bytes to fit the 10 bit
// block length.
static constexpr const size_t kAudioMaxRtpPacketLen =
1200; // The typical MTU is 1200 bytes.
static constexpr size_t kRedHeaderLength = 4; // 4 bytes RED header.
static constexpr size_t kRedLastHeaderLength =
1; // reduced size for last RED header.
static constexpr size_t kRedNumberOfRedundantEncodings =
2; // The level of redundancy we support.
AudioEncoderCopyRed::Config::Config() = default;
AudioEncoderCopyRed::Config::Config(Config&&) = default;
@ -30,9 +38,16 @@ AudioEncoderCopyRed::Config::~Config() = default;
AudioEncoderCopyRed::AudioEncoderCopyRed(Config&& config)
: speech_encoder_(std::move(config.speech_encoder)),
primary_encoded_(0, kAudioMaxRtpPacketLen),
max_packet_length_(kAudioMaxRtpPacketLen),
red_payload_type_(config.payload_type) {
RTC_CHECK(speech_encoder_) << "Speech encoder not provided.";
for (size_t i = 0; i < kRedNumberOfRedundantEncodings; i++) {
std::pair<EncodedInfo, rtc::Buffer> redundant;
redundant.second.EnsureCapacity(kAudioMaxRtpPacketLen);
redundant_encodings_.push_front(std::move(redundant));
}
}
AudioEncoderCopyRed::~AudioEncoderCopyRed() = default;
@ -61,104 +76,86 @@ int AudioEncoderCopyRed::GetTargetBitrate() const {
return speech_encoder_->GetTargetBitrate();
}
size_t AudioEncoderCopyRed::CalculateHeaderLength(size_t encoded_bytes) const {
size_t header_size = 1;
size_t bytes_available = max_packet_length_ - encoded_bytes;
if (secondary_info_.encoded_bytes > 0 &&
secondary_info_.encoded_bytes < bytes_available) {
header_size += 4;
bytes_available -= secondary_info_.encoded_bytes;
}
if (tertiary_info_.encoded_bytes > 0 &&
tertiary_info_.encoded_bytes < bytes_available) {
header_size += 4;
}
return header_size > 1 ? header_size : 0;
}
AudioEncoder::EncodedInfo AudioEncoderCopyRed::EncodeImpl(
uint32_t rtp_timestamp,
rtc::ArrayView<const int16_t> audio,
rtc::Buffer* encoded) {
rtc::Buffer primary_encoded;
primary_encoded_.Clear();
EncodedInfo info =
speech_encoder_->Encode(rtp_timestamp, audio, &primary_encoded);
speech_encoder_->Encode(rtp_timestamp, audio, &primary_encoded_);
RTC_CHECK(info.redundant.empty()) << "Cannot use nested redundant encoders.";
RTC_DCHECK_EQ(primary_encoded.size(), info.encoded_bytes);
RTC_DCHECK_EQ(primary_encoded_.size(), info.encoded_bytes);
if (info.encoded_bytes == 0 || info.encoded_bytes > kRedMaxPacketSize) {
return info;
}
RTC_DCHECK_GT(max_packet_length_, info.encoded_bytes);
size_t header_length_bytes = kRedLastHeaderLength;
size_t bytes_available = max_packet_length_ - info.encoded_bytes;
auto it = redundant_encodings_.begin();
// Determine how much redundancy we can fit into our packet by
// iterating forward.
for (; it != redundant_encodings_.end(); it++) {
if (bytes_available < kRedHeaderLength + it->first.encoded_bytes) {
break;
}
if (it->first.encoded_bytes == 0) {
break;
}
bytes_available -= kRedHeaderLength + it->first.encoded_bytes;
header_length_bytes += kRedHeaderLength;
}
// Allocate room for RFC 2198 header if there is redundant data.
// Otherwise this will send the primary payload type without
// wrapping in RED.
const size_t header_length_bytes = CalculateHeaderLength(info.encoded_bytes);
if (header_length_bytes == kRedLastHeaderLength) {
header_length_bytes = 0;
}
encoded->SetSize(header_length_bytes);
// Iterate backwards and append the data.
size_t header_offset = 0;
size_t bytes_available = max_packet_length_ - info.encoded_bytes;
if (tertiary_info_.encoded_bytes > 0 &&
tertiary_info_.encoded_bytes + secondary_info_.encoded_bytes <
bytes_available) {
encoded->AppendData(tertiary_encoded_);
while (it-- != redundant_encodings_.begin()) {
encoded->AppendData(it->second);
const uint32_t timestamp_delta =
info.encoded_timestamp - tertiary_info_.encoded_timestamp;
encoded->data()[header_offset] = tertiary_info_.payload_type | 0x80;
info.encoded_timestamp - it->first.encoded_timestamp;
encoded->data()[header_offset] = it->first.payload_type | 0x80;
rtc::SetBE16(static_cast<uint8_t*>(encoded->data()) + header_offset + 1,
(timestamp_delta << 2) | (tertiary_info_.encoded_bytes >> 8));
encoded->data()[header_offset + 3] = tertiary_info_.encoded_bytes & 0xff;
header_offset += 4;
bytes_available -= tertiary_info_.encoded_bytes;
}
if (secondary_info_.encoded_bytes > 0 &&
secondary_info_.encoded_bytes < bytes_available) {
encoded->AppendData(secondary_encoded_);
const uint32_t timestamp_delta =
info.encoded_timestamp - secondary_info_.encoded_timestamp;
encoded->data()[header_offset] = secondary_info_.payload_type | 0x80;
rtc::SetBE16(static_cast<uint8_t*>(encoded->data()) + header_offset + 1,
(timestamp_delta << 2) | (secondary_info_.encoded_bytes >> 8));
encoded->data()[header_offset + 3] = secondary_info_.encoded_bytes & 0xff;
header_offset += 4;
bytes_available -= secondary_info_.encoded_bytes;
}
encoded->AppendData(primary_encoded);
if (header_length_bytes > 0) {
RTC_DCHECK_EQ(header_offset, header_length_bytes - 1);
encoded->data()[header_offset] = info.payload_type;
(timestamp_delta << 2) | (it->first.encoded_bytes >> 8));
encoded->data()[header_offset + 3] = it->first.encoded_bytes & 0xff;
header_offset += kRedHeaderLength;
info.redundant.push_back(it->first);
}
// |info| will be implicitly cast to an EncodedInfoLeaf struct, effectively
// discarding the (empty) vector of redundant information. This is
// intentional.
if (header_length_bytes > 0) {
info.redundant.push_back(info);
RTC_DCHECK_EQ(info.redundant.size(), 1);
RTC_DCHECK_EQ(info.speech, info.redundant[0].speech);
if (secondary_info_.encoded_bytes > 0) {
info.redundant.push_back(secondary_info_);
RTC_DCHECK_EQ(info.redundant.size(), 2);
}
if (tertiary_info_.encoded_bytes > 0) {
info.redundant.push_back(tertiary_info_);
RTC_DCHECK_EQ(info.redundant.size(),
2 + (secondary_info_.encoded_bytes > 0 ? 1 : 0));
RTC_DCHECK_EQ(info.speech,
info.redundant[info.redundant.size() - 1].speech);
}
// Save secondary to tertiary.
tertiary_encoded_.SetData(secondary_encoded_);
tertiary_info_ = secondary_info_;
encoded->AppendData(primary_encoded_);
if (header_length_bytes > 0) {
RTC_DCHECK_EQ(header_offset, header_length_bytes - 1);
encoded->data()[header_offset] = info.payload_type;
}
// Save primary to secondary.
secondary_encoded_.SetData(primary_encoded);
secondary_info_ = info;
// Shift the redundant encodings.
it = redundant_encodings_.begin();
for (auto next = std::next(it); next != redundant_encodings_.end();
it++, next = std::next(it)) {
next->first = it->first;
next->second.SetData(it->second);
}
it = redundant_encodings_.begin();
it->first = info;
it->second.SetData(primary_encoded_);
// Update main EncodedInfo.
if (header_length_bytes > 0) {
@ -170,8 +167,12 @@ AudioEncoder::EncodedInfo AudioEncoderCopyRed::EncodeImpl(
void AudioEncoderCopyRed::Reset() {
speech_encoder_->Reset();
secondary_encoded_.Clear();
secondary_info_.encoded_bytes = 0;
redundant_encodings_.clear();
for (size_t i = 0; i < kRedNumberOfRedundantEncodings; i++) {
std::pair<EncodedInfo, rtc::Buffer> redundant;
redundant.second.EnsureCapacity(kAudioMaxRtpPacketLen);
redundant_encodings_.push_front(std::move(redundant));
}
}
bool AudioEncoderCopyRed::SetFec(bool enable) {

View File

@ -14,6 +14,7 @@
#include <stddef.h>
#include <stdint.h>
#include <list>
#include <memory>
#include <utility>
@ -26,10 +27,12 @@
namespace webrtc {
// This class implements redundant audio coding. The class object will have an
// underlying AudioEncoder object that performs the actual encodings. The
// current class will gather the two latest encodings from the underlying codec
// into one packet.
// This class implements redundant audio coding as described in
// https://tools.ietf.org/html/rfc2198
// The class object will have an underlying AudioEncoder object that performs
// the actual encodings. The current class will gather the N latest encodings
// from the underlying codec into one packet. Currently N is hard-coded to 2.
class AudioEncoderCopyRed final : public AudioEncoder {
public:
struct Config {
@ -84,15 +87,11 @@ class AudioEncoderCopyRed final : public AudioEncoder {
rtc::Buffer* encoded) override;
private:
size_t CalculateHeaderLength(size_t encoded_bytes) const;
std::unique_ptr<AudioEncoder> speech_encoder_;
rtc::Buffer primary_encoded_;
size_t max_packet_length_;
int red_payload_type_;
rtc::Buffer secondary_encoded_;
EncodedInfoLeaf secondary_info_;
rtc::Buffer tertiary_encoded_;
EncodedInfoLeaf tertiary_info_;
std::list<std::pair<EncodedInfo, rtc::Buffer>> redundant_encodings_;
RTC_DISALLOW_COPY_AND_ASSIGN(AudioEncoderCopyRed);
};

View File

@ -152,7 +152,7 @@ TEST_F(AudioEncoderCopyRedTest, CheckNoOutput) {
Encode();
// First call is a special case, since it does not include a secondary
// payload.
EXPECT_EQ(1u, encoded_info_.redundant.size());
EXPECT_EQ(0u, encoded_info_.redundant.size());
EXPECT_EQ(kEncodedSize, encoded_info_.encoded_bytes);
// Next call to the speech encoder will not produce any output.
@ -180,7 +180,7 @@ TEST_F(AudioEncoderCopyRedTest, CheckPayloadSizes) {
// First call is a special case, since it does not include a secondary
// payload.
Encode();
EXPECT_EQ(1u, encoded_info_.redundant.size());
EXPECT_EQ(0u, encoded_info_.redundant.size());
EXPECT_EQ(1u, encoded_info_.encoded_bytes);
// Second call is also special since it does not include a ternary
@ -192,9 +192,9 @@ TEST_F(AudioEncoderCopyRedTest, CheckPayloadSizes) {
for (size_t i = 3; i <= kNumPackets; ++i) {
Encode();
ASSERT_EQ(3u, encoded_info_.redundant.size());
EXPECT_EQ(i, encoded_info_.redundant[0].encoded_bytes);
EXPECT_EQ(i, encoded_info_.redundant[2].encoded_bytes);
EXPECT_EQ(i - 1, encoded_info_.redundant[1].encoded_bytes);
EXPECT_EQ(i - 2, encoded_info_.redundant[2].encoded_bytes);
EXPECT_EQ(i - 2, encoded_info_.redundant[0].encoded_bytes);
EXPECT_EQ(9 + i + (i - 1) + (i - 2), encoded_info_.encoded_bytes);
}
}
@ -222,8 +222,8 @@ TEST_F(AudioEncoderCopyRedTest, CheckTimestamps) {
Encode();
ASSERT_EQ(2u, encoded_info_.redundant.size());
EXPECT_EQ(primary_timestamp, encoded_info_.redundant[0].encoded_timestamp);
EXPECT_EQ(secondary_timestamp, encoded_info_.redundant[1].encoded_timestamp);
EXPECT_EQ(primary_timestamp, encoded_info_.redundant[1].encoded_timestamp);
EXPECT_EQ(secondary_timestamp, encoded_info_.redundant[0].encoded_timestamp);
EXPECT_EQ(primary_timestamp, encoded_info_.encoded_timestamp);
}
@ -280,9 +280,7 @@ TEST_F(AudioEncoderCopyRedTest, CheckPayloadType) {
// First call is a special case, since it does not include a secondary
// payload.
Encode();
ASSERT_EQ(1u, encoded_info_.redundant.size());
EXPECT_EQ(primary_payload_type, encoded_info_.redundant[0].payload_type);
EXPECT_EQ(primary_payload_type, encoded_info_.payload_type);
ASSERT_EQ(0u, encoded_info_.redundant.size());
const int secondary_payload_type = red_payload_type_ + 2;
info.payload_type = secondary_payload_type;
@ -291,8 +289,8 @@ TEST_F(AudioEncoderCopyRedTest, CheckPayloadType) {
Encode();
ASSERT_EQ(2u, encoded_info_.redundant.size());
EXPECT_EQ(secondary_payload_type, encoded_info_.redundant[0].payload_type);
EXPECT_EQ(primary_payload_type, encoded_info_.redundant[1].payload_type);
EXPECT_EQ(secondary_payload_type, encoded_info_.redundant[1].payload_type);
EXPECT_EQ(primary_payload_type, encoded_info_.redundant[0].payload_type);
EXPECT_EQ(red_payload_type_, encoded_info_.payload_type);
}
@ -316,7 +314,7 @@ TEST_F(AudioEncoderCopyRedTest, CheckRFC2198Header) {
EXPECT_EQ(encoded_[0], primary_payload_type | 0x80);
uint32_t timestamp_delta = encoded_info_.encoded_timestamp -
encoded_info_.redundant[1].encoded_timestamp;
encoded_info_.redundant[0].encoded_timestamp;
// Timestamp delta is encoded as a 14 bit value.
EXPECT_EQ(encoded_[1], timestamp_delta >> 6);
EXPECT_EQ(static_cast<uint8_t>(encoded_[2] >> 2), timestamp_delta & 0x3f);
@ -335,13 +333,13 @@ TEST_F(AudioEncoderCopyRedTest, CheckRFC2198Header) {
EXPECT_EQ(encoded_[0], primary_payload_type | 0x80);
timestamp_delta = encoded_info_.encoded_timestamp -
encoded_info_.redundant[2].encoded_timestamp;
encoded_info_.redundant[0].encoded_timestamp;
// Timestamp delta is encoded as a 14 bit value.
EXPECT_EQ(encoded_[1], timestamp_delta >> 6);
EXPECT_EQ(static_cast<uint8_t>(encoded_[2] >> 2), timestamp_delta & 0x3f);
// Redundant length is encoded as 10 bit value.
EXPECT_EQ(encoded_[2] & 0x3u, encoded_info_.redundant[2].encoded_bytes >> 8);
EXPECT_EQ(encoded_[3], encoded_info_.redundant[2].encoded_bytes & 0xff);
EXPECT_EQ(encoded_[2] & 0x3u, encoded_info_.redundant[1].encoded_bytes >> 8);
EXPECT_EQ(encoded_[3], encoded_info_.redundant[1].encoded_bytes & 0xff);
EXPECT_EQ(encoded_[4], primary_payload_type | 0x80);
timestamp_delta = encoded_info_.encoded_timestamp -
@ -350,8 +348,8 @@ TEST_F(AudioEncoderCopyRedTest, CheckRFC2198Header) {
EXPECT_EQ(encoded_[5], timestamp_delta >> 6);
EXPECT_EQ(static_cast<uint8_t>(encoded_[6] >> 2), timestamp_delta & 0x3f);
// Redundant length is encoded as 10 bit value.
EXPECT_EQ(encoded_[6] & 0x3u, encoded_info_.redundant[2].encoded_bytes >> 8);
EXPECT_EQ(encoded_[7], encoded_info_.redundant[2].encoded_bytes & 0xff);
EXPECT_EQ(encoded_[6] & 0x3u, encoded_info_.redundant[1].encoded_bytes >> 8);
EXPECT_EQ(encoded_[7], encoded_info_.redundant[1].encoded_bytes & 0xff);
EXPECT_EQ(encoded_[8], primary_payload_type);
}