2 * Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
12 * This file includes unit tests for NetEQ.
15 #include "webrtc/modules/audio_coding/neteq/interface/neteq.h"
19 #include <string.h> // memset
26 #include "gflags/gflags.h"
27 #include "gtest/gtest.h"
28 #include "webrtc/modules/audio_coding/neteq/test/NETEQTEST_RTPpacket.h"
29 #include "webrtc/modules/audio_coding/codecs/pcm16b/include/pcm16b.h"
30 #include "webrtc/test/testsupport/fileutils.h"
31 #include "webrtc/test/testsupport/gtest_disable.h"
32 #include "webrtc/typedefs.h"
34 DEFINE_bool(gen_ref, false, "Generate reference files.");
38 static bool IsAllZero(const int16_t* buf, int buf_length) {
40 for (int n = 0; n < buf_length && all_zero; ++n)
41 all_zero = buf[n] == 0;
45 static bool IsAllNonZero(const int16_t* buf, int buf_length) {
46 bool all_non_zero = true;
47 for (int n = 0; n < buf_length && all_non_zero; ++n)
48 all_non_zero = buf[n] != 0;
54 RefFiles(const std::string& input_file, const std::string& output_file);
56 template<class T> void ProcessReference(const T& test_results);
57 template<typename T, size_t n> void ProcessReference(
58 const T (&test_results)[n],
60 template<typename T, size_t n> void WriteToFile(
61 const T (&test_results)[n],
63 template<typename T, size_t n> void ReadFromFileAndCompare(
64 const T (&test_results)[n],
66 void WriteToFile(const NetEqNetworkStatistics& stats);
67 void ReadFromFileAndCompare(const NetEqNetworkStatistics& stats);
68 void WriteToFile(const RtcpStatistics& stats);
69 void ReadFromFileAndCompare(const RtcpStatistics& stats);
75 RefFiles::RefFiles(const std::string &input_file,
76 const std::string &output_file)
79 if (!input_file.empty()) {
80 input_fp_ = fopen(input_file.c_str(), "rb");
81 EXPECT_TRUE(input_fp_ != NULL);
83 if (!output_file.empty()) {
84 output_fp_ = fopen(output_file.c_str(), "wb");
85 EXPECT_TRUE(output_fp_ != NULL);
89 RefFiles::~RefFiles() {
91 EXPECT_EQ(EOF, fgetc(input_fp_)); // Make sure that we reached the end.
94 if (output_fp_) fclose(output_fp_);
98 void RefFiles::ProcessReference(const T& test_results) {
99 WriteToFile(test_results);
100 ReadFromFileAndCompare(test_results);
103 template<typename T, size_t n>
104 void RefFiles::ProcessReference(const T (&test_results)[n], size_t length) {
105 WriteToFile(test_results, length);
106 ReadFromFileAndCompare(test_results, length);
109 template<typename T, size_t n>
110 void RefFiles::WriteToFile(const T (&test_results)[n], size_t length) {
112 ASSERT_EQ(length, fwrite(&test_results, sizeof(T), length, output_fp_));
116 template<typename T, size_t n>
117 void RefFiles::ReadFromFileAndCompare(const T (&test_results)[n],
120 // Read from ref file.
121 T* ref = new T[length];
122 ASSERT_EQ(length, fread(ref, sizeof(T), length, input_fp_));
124 ASSERT_EQ(0, memcmp(&test_results, ref, sizeof(T) * length));
129 void RefFiles::WriteToFile(const NetEqNetworkStatistics& stats) {
131 ASSERT_EQ(1u, fwrite(&stats, sizeof(NetEqNetworkStatistics), 1,
136 void RefFiles::ReadFromFileAndCompare(
137 const NetEqNetworkStatistics& stats) {
139 // Read from ref file.
140 size_t stat_size = sizeof(NetEqNetworkStatistics);
141 NetEqNetworkStatistics ref_stats;
142 ASSERT_EQ(1u, fread(&ref_stats, stat_size, 1, input_fp_));
144 ASSERT_EQ(0, memcmp(&stats, &ref_stats, stat_size));
148 void RefFiles::WriteToFile(const RtcpStatistics& stats) {
150 ASSERT_EQ(1u, fwrite(&(stats.fraction_lost), sizeof(stats.fraction_lost), 1,
152 ASSERT_EQ(1u, fwrite(&(stats.cumulative_lost),
153 sizeof(stats.cumulative_lost), 1, output_fp_));
154 ASSERT_EQ(1u, fwrite(&(stats.extended_max_sequence_number),
155 sizeof(stats.extended_max_sequence_number), 1,
157 ASSERT_EQ(1u, fwrite(&(stats.jitter), sizeof(stats.jitter), 1,
162 void RefFiles::ReadFromFileAndCompare(
163 const RtcpStatistics& stats) {
165 // Read from ref file.
166 RtcpStatistics ref_stats;
167 ASSERT_EQ(1u, fread(&(ref_stats.fraction_lost),
168 sizeof(ref_stats.fraction_lost), 1, input_fp_));
169 ASSERT_EQ(1u, fread(&(ref_stats.cumulative_lost),
170 sizeof(ref_stats.cumulative_lost), 1, input_fp_));
171 ASSERT_EQ(1u, fread(&(ref_stats.extended_max_sequence_number),
172 sizeof(ref_stats.extended_max_sequence_number), 1,
174 ASSERT_EQ(1u, fread(&(ref_stats.jitter), sizeof(ref_stats.jitter), 1,
177 ASSERT_EQ(ref_stats.fraction_lost, stats.fraction_lost);
178 ASSERT_EQ(ref_stats.cumulative_lost, stats.cumulative_lost);
179 ASSERT_EQ(ref_stats.extended_max_sequence_number,
180 stats.extended_max_sequence_number);
181 ASSERT_EQ(ref_stats.jitter, stats.jitter);
185 class NetEqDecodingTest : public ::testing::Test {
187 // NetEQ must be polled for data once every 10 ms. Thus, neither of the
188 // constants below can be changed.
189 static const int kTimeStepMs = 10;
190 static const int kBlockSize8kHz = kTimeStepMs * 8;
191 static const int kBlockSize16kHz = kTimeStepMs * 16;
192 static const int kBlockSize32kHz = kTimeStepMs * 32;
193 static const int kMaxBlockSize = kBlockSize32kHz;
194 static const int kInitSampleRateHz = 8000;
197 virtual void SetUp();
198 virtual void TearDown();
199 void SelectDecoders(NetEqDecoder* used_codec);
201 void OpenInputFile(const std::string &rtp_file);
202 void Process(NETEQTEST_RTPpacket* rtp_ptr, int* out_len);
203 void DecodeAndCompare(const std::string& rtp_file,
204 const std::string& ref_file,
205 const std::string& stat_ref_file,
206 const std::string& rtcp_ref_file);
207 static void PopulateRtpInfo(int frame_index,
209 WebRtcRTPHeader* rtp_info);
210 static void PopulateCng(int frame_index,
212 WebRtcRTPHeader* rtp_info,
216 void WrapTest(uint16_t start_seq_no, uint32_t start_timestamp,
217 const std::set<uint16_t>& drop_seq_numbers,
218 bool expect_seq_no_wrap, bool expect_timestamp_wrap);
220 void LongCngWithClockDrift(double drift_factor,
221 double network_freeze_ms,
222 bool pull_audio_during_freeze,
223 int delay_tolerance_ms,
224 int max_time_to_speech_ms);
228 uint32_t PlayoutTimestamp();
231 NetEq::Config config_;
233 unsigned int sim_clock_;
234 int16_t out_data_[kMaxBlockSize];
235 int output_sample_rate_;
236 int algorithmic_delay_ms_;
239 // Allocating the static const so that it can be passed by reference.
240 const int NetEqDecodingTest::kTimeStepMs;
241 const int NetEqDecodingTest::kBlockSize8kHz;
242 const int NetEqDecodingTest::kBlockSize16kHz;
243 const int NetEqDecodingTest::kBlockSize32kHz;
244 const int NetEqDecodingTest::kMaxBlockSize;
245 const int NetEqDecodingTest::kInitSampleRateHz;
247 NetEqDecodingTest::NetEqDecodingTest()
252 output_sample_rate_(kInitSampleRateHz),
253 algorithmic_delay_ms_(0) {
254 config_.sample_rate_hz = kInitSampleRateHz;
255 memset(out_data_, 0, sizeof(out_data_));
258 void NetEqDecodingTest::SetUp() {
259 neteq_ = NetEq::Create(config_);
260 NetEqNetworkStatistics stat;
261 ASSERT_EQ(0, neteq_->NetworkStatistics(&stat));
262 algorithmic_delay_ms_ = stat.current_buffer_size_ms;
267 void NetEqDecodingTest::TearDown() {
273 void NetEqDecodingTest::LoadDecoders() {
275 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderPCMu, 0));
277 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderPCMa, 8));
278 #ifndef WEBRTC_ANDROID
280 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderILBC, 102));
281 #endif // WEBRTC_ANDROID
283 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderISAC, 103));
284 #ifndef WEBRTC_ANDROID
286 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderISACswb, 104));
288 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderISACfb, 105));
289 #endif // WEBRTC_ANDROID
291 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderPCM16B, 93));
293 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderPCM16Bwb, 94));
294 // Load PCM16B swb32.
295 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderPCM16Bswb32kHz, 95));
297 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderCNGnb, 13));
299 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderCNGwb, 98));
302 void NetEqDecodingTest::OpenInputFile(const std::string &rtp_file) {
303 rtp_fp_ = fopen(rtp_file.c_str(), "rb");
304 ASSERT_TRUE(rtp_fp_ != NULL);
305 ASSERT_EQ(0, NETEQTEST_RTPpacket::skipFileHeader(rtp_fp_));
308 void NetEqDecodingTest::Process(NETEQTEST_RTPpacket* rtp, int* out_len) {
309 // Check if time to receive.
310 while ((sim_clock_ >= rtp->time()) &&
311 (rtp->dataLen() >= 0)) {
312 if (rtp->dataLen() > 0) {
313 WebRtcRTPHeader rtpInfo;
314 rtp->parseHeader(&rtpInfo);
315 ASSERT_EQ(0, neteq_->InsertPacket(
319 rtp->time() * (output_sample_rate_ / 1000)));
322 ASSERT_NE(-1, rtp->readFromFile(rtp_fp_));
325 // Get audio from NetEq.
326 NetEqOutputType type;
328 ASSERT_EQ(0, neteq_->GetAudio(kMaxBlockSize, out_data_, out_len,
329 &num_channels, &type));
330 ASSERT_TRUE((*out_len == kBlockSize8kHz) ||
331 (*out_len == kBlockSize16kHz) ||
332 (*out_len == kBlockSize32kHz));
333 output_sample_rate_ = *out_len / 10 * 1000;
336 sim_clock_ += kTimeStepMs;
339 void NetEqDecodingTest::DecodeAndCompare(const std::string& rtp_file,
340 const std::string& ref_file,
341 const std::string& stat_ref_file,
342 const std::string& rtcp_ref_file) {
343 OpenInputFile(rtp_file);
345 std::string ref_out_file = "";
346 if (ref_file.empty()) {
347 ref_out_file = webrtc::test::OutputPath() + "neteq_universal_ref.pcm";
349 RefFiles ref_files(ref_file, ref_out_file);
351 std::string stat_out_file = "";
352 if (stat_ref_file.empty()) {
353 stat_out_file = webrtc::test::OutputPath() + "neteq_network_stats.dat";
355 RefFiles network_stat_files(stat_ref_file, stat_out_file);
357 std::string rtcp_out_file = "";
358 if (rtcp_ref_file.empty()) {
359 rtcp_out_file = webrtc::test::OutputPath() + "neteq_rtcp_stats.dat";
361 RefFiles rtcp_stat_files(rtcp_ref_file, rtcp_out_file);
363 NETEQTEST_RTPpacket rtp;
364 ASSERT_GT(rtp.readFromFile(rtp_fp_), 0);
366 while (rtp.dataLen() >= 0) {
367 std::ostringstream ss;
368 ss << "Lap number " << i++ << " in DecodeAndCompare while loop";
369 SCOPED_TRACE(ss.str()); // Print out the parameter values on failure.
371 ASSERT_NO_FATAL_FAILURE(Process(&rtp, &out_len));
372 ASSERT_NO_FATAL_FAILURE(ref_files.ProcessReference(out_data_, out_len));
374 // Query the network statistics API once per second
375 if (sim_clock_ % 1000 == 0) {
376 // Process NetworkStatistics.
377 NetEqNetworkStatistics network_stats;
378 ASSERT_EQ(0, neteq_->NetworkStatistics(&network_stats));
379 ASSERT_NO_FATAL_FAILURE(
380 network_stat_files.ProcessReference(network_stats));
383 RtcpStatistics rtcp_stats;
384 neteq_->GetRtcpStatistics(&rtcp_stats);
385 ASSERT_NO_FATAL_FAILURE(rtcp_stat_files.ProcessReference(rtcp_stats));
390 void NetEqDecodingTest::PopulateRtpInfo(int frame_index,
392 WebRtcRTPHeader* rtp_info) {
393 rtp_info->header.sequenceNumber = frame_index;
394 rtp_info->header.timestamp = timestamp;
395 rtp_info->header.ssrc = 0x1234; // Just an arbitrary SSRC.
396 rtp_info->header.payloadType = 94; // PCM16b WB codec.
397 rtp_info->header.markerBit = 0;
400 void NetEqDecodingTest::PopulateCng(int frame_index,
402 WebRtcRTPHeader* rtp_info,
405 rtp_info->header.sequenceNumber = frame_index;
406 rtp_info->header.timestamp = timestamp;
407 rtp_info->header.ssrc = 0x1234; // Just an arbitrary SSRC.
408 rtp_info->header.payloadType = 98; // WB CNG.
409 rtp_info->header.markerBit = 0;
410 payload[0] = 64; // Noise level -64 dBov, quite arbitrarily chosen.
411 *payload_len = 1; // Only noise level, no spectral parameters.
414 TEST_F(NetEqDecodingTest, DISABLED_ON_ANDROID(TestBitExactness)) {
415 const std::string input_rtp_file = webrtc::test::ProjectRootPath() +
416 "resources/audio_coding/neteq_universal_new.rtp";
417 // Note that neteq4_universal_ref.pcm and neteq4_universal_ref_win_32.pcm
418 // are identical. The latter could have been removed, but if clients still
419 // have a copy of the file, the test will fail.
420 const std::string input_ref_file =
421 webrtc::test::ResourcePath("audio_coding/neteq4_universal_ref", "pcm");
422 #if defined(_MSC_VER) && (_MSC_VER >= 1700)
423 // For Visual Studio 2012 and later, we will have to use the generic reference
424 // file, rather than the windows-specific one.
425 const std::string network_stat_ref_file = webrtc::test::ProjectRootPath() +
426 "resources/audio_coding/neteq4_network_stats.dat";
428 const std::string network_stat_ref_file =
429 webrtc::test::ResourcePath("audio_coding/neteq4_network_stats", "dat");
431 const std::string rtcp_stat_ref_file =
432 webrtc::test::ResourcePath("audio_coding/neteq4_rtcp_stats", "dat");
435 DecodeAndCompare(input_rtp_file, "", "", "");
437 DecodeAndCompare(input_rtp_file,
439 network_stat_ref_file,
444 // TODO(hlundin): Re-enable test once the statistics interface is up and again.
445 TEST_F(NetEqDecodingTest, TestFrameWaitingTimeStatistics) {
446 // Use fax mode to avoid time-scaling. This is to simplify the testing of
447 // packet waiting times in the packet buffer.
448 neteq_->SetPlayoutMode(kPlayoutFax);
449 ASSERT_EQ(kPlayoutFax, neteq_->PlayoutMode());
450 // Insert 30 dummy packets at once. Each packet contains 10 ms 16 kHz audio.
451 size_t num_frames = 30;
452 const int kSamples = 10 * 16;
453 const int kPayloadBytes = kSamples * 2;
454 for (size_t i = 0; i < num_frames; ++i) {
455 uint16_t payload[kSamples] = {0};
456 WebRtcRTPHeader rtp_info;
457 rtp_info.header.sequenceNumber = i;
458 rtp_info.header.timestamp = i * kSamples;
459 rtp_info.header.ssrc = 0x1234; // Just an arbitrary SSRC.
460 rtp_info.header.payloadType = 94; // PCM16b WB codec.
461 rtp_info.header.markerBit = 0;
462 ASSERT_EQ(0, neteq_->InsertPacket(
464 reinterpret_cast<uint8_t*>(payload),
467 // Pull out all data.
468 for (size_t i = 0; i < num_frames; ++i) {
471 NetEqOutputType type;
472 ASSERT_EQ(0, neteq_->GetAudio(kMaxBlockSize, out_data_, &out_len,
473 &num_channels, &type));
474 ASSERT_EQ(kBlockSize16kHz, out_len);
477 std::vector<int> waiting_times;
478 neteq_->WaitingTimes(&waiting_times);
479 EXPECT_EQ(num_frames, waiting_times.size());
480 // Since all frames are dumped into NetEQ at once, but pulled out with 10 ms
481 // spacing (per definition), we expect the delay to increase with 10 ms for
483 for (size_t i = 0; i < waiting_times.size(); ++i) {
484 EXPECT_EQ(static_cast<int>(i + 1) * 10, waiting_times[i]);
487 // Check statistics again and make sure it's been reset.
488 neteq_->WaitingTimes(&waiting_times);
489 int len = waiting_times.size();
492 // Process > 100 frames, and make sure that that we get statistics
493 // only for 100 frames. Note the new SSRC, causing NetEQ to reset.
495 for (size_t i = 0; i < num_frames; ++i) {
496 uint16_t payload[kSamples] = {0};
497 WebRtcRTPHeader rtp_info;
498 rtp_info.header.sequenceNumber = i;
499 rtp_info.header.timestamp = i * kSamples;
500 rtp_info.header.ssrc = 0x1235; // Just an arbitrary SSRC.
501 rtp_info.header.payloadType = 94; // PCM16b WB codec.
502 rtp_info.header.markerBit = 0;
503 ASSERT_EQ(0, neteq_->InsertPacket(
505 reinterpret_cast<uint8_t*>(payload),
509 NetEqOutputType type;
510 ASSERT_EQ(0, neteq_->GetAudio(kMaxBlockSize, out_data_, &out_len,
511 &num_channels, &type));
512 ASSERT_EQ(kBlockSize16kHz, out_len);
515 neteq_->WaitingTimes(&waiting_times);
516 EXPECT_EQ(100u, waiting_times.size());
519 TEST_F(NetEqDecodingTest, TestAverageInterArrivalTimeNegative) {
520 const int kNumFrames = 3000; // Needed for convergence.
522 const int kSamples = 10 * 16;
523 const int kPayloadBytes = kSamples * 2;
524 while (frame_index < kNumFrames) {
525 // Insert one packet each time, except every 10th time where we insert two
526 // packets at once. This will create a negative clock-drift of approx. 10%.
527 int num_packets = (frame_index % 10 == 0 ? 2 : 1);
528 for (int n = 0; n < num_packets; ++n) {
529 uint8_t payload[kPayloadBytes] = {0};
530 WebRtcRTPHeader rtp_info;
531 PopulateRtpInfo(frame_index, frame_index * kSamples, &rtp_info);
532 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes, 0));
536 // Pull out data once.
539 NetEqOutputType type;
540 ASSERT_EQ(0, neteq_->GetAudio(kMaxBlockSize, out_data_, &out_len,
541 &num_channels, &type));
542 ASSERT_EQ(kBlockSize16kHz, out_len);
545 NetEqNetworkStatistics network_stats;
546 ASSERT_EQ(0, neteq_->NetworkStatistics(&network_stats));
547 EXPECT_EQ(-103196, network_stats.clockdrift_ppm);
550 TEST_F(NetEqDecodingTest, TestAverageInterArrivalTimePositive) {
551 const int kNumFrames = 5000; // Needed for convergence.
553 const int kSamples = 10 * 16;
554 const int kPayloadBytes = kSamples * 2;
555 for (int i = 0; i < kNumFrames; ++i) {
556 // Insert one packet each time, except every 10th time where we don't insert
557 // any packet. This will create a positive clock-drift of approx. 11%.
558 int num_packets = (i % 10 == 9 ? 0 : 1);
559 for (int n = 0; n < num_packets; ++n) {
560 uint8_t payload[kPayloadBytes] = {0};
561 WebRtcRTPHeader rtp_info;
562 PopulateRtpInfo(frame_index, frame_index * kSamples, &rtp_info);
563 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes, 0));
567 // Pull out data once.
570 NetEqOutputType type;
571 ASSERT_EQ(0, neteq_->GetAudio(kMaxBlockSize, out_data_, &out_len,
572 &num_channels, &type));
573 ASSERT_EQ(kBlockSize16kHz, out_len);
576 NetEqNetworkStatistics network_stats;
577 ASSERT_EQ(0, neteq_->NetworkStatistics(&network_stats));
578 EXPECT_EQ(110946, network_stats.clockdrift_ppm);
581 void NetEqDecodingTest::LongCngWithClockDrift(double drift_factor,
582 double network_freeze_ms,
583 bool pull_audio_during_freeze,
584 int delay_tolerance_ms,
585 int max_time_to_speech_ms) {
587 uint32_t timestamp = 0;
588 const int kFrameSizeMs = 30;
589 const int kSamples = kFrameSizeMs * 16;
590 const int kPayloadBytes = kSamples * 2;
591 double next_input_time_ms = 0.0;
595 NetEqOutputType type;
597 // Insert speech for 5 seconds.
598 const int kSpeechDurationMs = 5000;
599 for (t_ms = 0; t_ms < kSpeechDurationMs; t_ms += 10) {
600 // Each turn in this for loop is 10 ms.
601 while (next_input_time_ms <= t_ms) {
602 // Insert one 30 ms speech frame.
603 uint8_t payload[kPayloadBytes] = {0};
604 WebRtcRTPHeader rtp_info;
605 PopulateRtpInfo(seq_no, timestamp, &rtp_info);
606 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes, 0));
608 timestamp += kSamples;
609 next_input_time_ms += static_cast<double>(kFrameSizeMs) * drift_factor;
611 // Pull out data once.
612 ASSERT_EQ(0, neteq_->GetAudio(kMaxBlockSize, out_data_, &out_len,
613 &num_channels, &type));
614 ASSERT_EQ(kBlockSize16kHz, out_len);
617 EXPECT_EQ(kOutputNormal, type);
618 int32_t delay_before = timestamp - PlayoutTimestamp();
620 // Insert CNG for 1 minute (= 60000 ms).
621 const int kCngPeriodMs = 100;
622 const int kCngPeriodSamples = kCngPeriodMs * 16; // Period in 16 kHz samples.
623 const int kCngDurationMs = 60000;
624 for (; t_ms < kSpeechDurationMs + kCngDurationMs; t_ms += 10) {
625 // Each turn in this for loop is 10 ms.
626 while (next_input_time_ms <= t_ms) {
627 // Insert one CNG frame each 100 ms.
628 uint8_t payload[kPayloadBytes];
630 WebRtcRTPHeader rtp_info;
631 PopulateCng(seq_no, timestamp, &rtp_info, payload, &payload_len);
632 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, payload_len, 0));
634 timestamp += kCngPeriodSamples;
635 next_input_time_ms += static_cast<double>(kCngPeriodMs) * drift_factor;
637 // Pull out data once.
638 ASSERT_EQ(0, neteq_->GetAudio(kMaxBlockSize, out_data_, &out_len,
639 &num_channels, &type));
640 ASSERT_EQ(kBlockSize16kHz, out_len);
643 EXPECT_EQ(kOutputCNG, type);
645 if (network_freeze_ms > 0) {
646 // First keep pulling audio for |network_freeze_ms| without inserting
647 // any data, then insert CNG data corresponding to |network_freeze_ms|
648 // without pulling any output audio.
649 const double loop_end_time = t_ms + network_freeze_ms;
650 for (; t_ms < loop_end_time; t_ms += 10) {
651 // Pull out data once.
654 kMaxBlockSize, out_data_, &out_len, &num_channels, &type));
655 ASSERT_EQ(kBlockSize16kHz, out_len);
656 EXPECT_EQ(kOutputCNG, type);
658 bool pull_once = pull_audio_during_freeze;
659 // If |pull_once| is true, GetAudio will be called once half-way through
660 // the network recovery period.
661 double pull_time_ms = (t_ms + next_input_time_ms) / 2;
662 while (next_input_time_ms <= t_ms) {
663 if (pull_once && next_input_time_ms >= pull_time_ms) {
665 // Pull out data once.
669 kMaxBlockSize, out_data_, &out_len, &num_channels, &type));
670 ASSERT_EQ(kBlockSize16kHz, out_len);
671 EXPECT_EQ(kOutputCNG, type);
674 // Insert one CNG frame each 100 ms.
675 uint8_t payload[kPayloadBytes];
677 WebRtcRTPHeader rtp_info;
678 PopulateCng(seq_no, timestamp, &rtp_info, payload, &payload_len);
679 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, payload_len, 0));
681 timestamp += kCngPeriodSamples;
682 next_input_time_ms += kCngPeriodMs * drift_factor;
686 // Insert speech again until output type is speech.
687 double speech_restart_time_ms = t_ms;
688 while (type != kOutputNormal) {
689 // Each turn in this for loop is 10 ms.
690 while (next_input_time_ms <= t_ms) {
691 // Insert one 30 ms speech frame.
692 uint8_t payload[kPayloadBytes] = {0};
693 WebRtcRTPHeader rtp_info;
694 PopulateRtpInfo(seq_no, timestamp, &rtp_info);
695 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes, 0));
697 timestamp += kSamples;
698 next_input_time_ms += kFrameSizeMs * drift_factor;
700 // Pull out data once.
701 ASSERT_EQ(0, neteq_->GetAudio(kMaxBlockSize, out_data_, &out_len,
702 &num_channels, &type));
703 ASSERT_EQ(kBlockSize16kHz, out_len);
708 // Check that the speech starts again within reasonable time.
709 double time_until_speech_returns_ms = t_ms - speech_restart_time_ms;
710 EXPECT_LT(time_until_speech_returns_ms, max_time_to_speech_ms);
711 int32_t delay_after = timestamp - PlayoutTimestamp();
712 // Compare delay before and after, and make sure it differs less than 20 ms.
713 EXPECT_LE(delay_after, delay_before + delay_tolerance_ms * 16);
714 EXPECT_GE(delay_after, delay_before - delay_tolerance_ms * 16);
717 TEST_F(NetEqDecodingTest, LongCngWithNegativeClockDrift) {
718 // Apply a clock drift of -25 ms / s (sender faster than receiver).
719 const double kDriftFactor = 1000.0 / (1000.0 + 25.0);
720 const double kNetworkFreezeTimeMs = 0.0;
721 const bool kGetAudioDuringFreezeRecovery = false;
722 const int kDelayToleranceMs = 20;
723 const int kMaxTimeToSpeechMs = 100;
724 LongCngWithClockDrift(kDriftFactor,
725 kNetworkFreezeTimeMs,
726 kGetAudioDuringFreezeRecovery,
731 TEST_F(NetEqDecodingTest, LongCngWithPositiveClockDrift) {
732 // Apply a clock drift of +25 ms / s (sender slower than receiver).
733 const double kDriftFactor = 1000.0 / (1000.0 - 25.0);
734 const double kNetworkFreezeTimeMs = 0.0;
735 const bool kGetAudioDuringFreezeRecovery = false;
736 const int kDelayToleranceMs = 20;
737 const int kMaxTimeToSpeechMs = 100;
738 LongCngWithClockDrift(kDriftFactor,
739 kNetworkFreezeTimeMs,
740 kGetAudioDuringFreezeRecovery,
745 TEST_F(NetEqDecodingTest, LongCngWithNegativeClockDriftNetworkFreeze) {
746 // Apply a clock drift of -25 ms / s (sender faster than receiver).
747 const double kDriftFactor = 1000.0 / (1000.0 + 25.0);
748 const double kNetworkFreezeTimeMs = 5000.0;
749 const bool kGetAudioDuringFreezeRecovery = false;
750 const int kDelayToleranceMs = 50;
751 const int kMaxTimeToSpeechMs = 200;
752 LongCngWithClockDrift(kDriftFactor,
753 kNetworkFreezeTimeMs,
754 kGetAudioDuringFreezeRecovery,
759 TEST_F(NetEqDecodingTest, LongCngWithPositiveClockDriftNetworkFreeze) {
760 // Apply a clock drift of +25 ms / s (sender slower than receiver).
761 const double kDriftFactor = 1000.0 / (1000.0 - 25.0);
762 const double kNetworkFreezeTimeMs = 5000.0;
763 const bool kGetAudioDuringFreezeRecovery = false;
764 const int kDelayToleranceMs = 20;
765 const int kMaxTimeToSpeechMs = 100;
766 LongCngWithClockDrift(kDriftFactor,
767 kNetworkFreezeTimeMs,
768 kGetAudioDuringFreezeRecovery,
773 TEST_F(NetEqDecodingTest, LongCngWithPositiveClockDriftNetworkFreezeExtraPull) {
774 // Apply a clock drift of +25 ms / s (sender slower than receiver).
775 const double kDriftFactor = 1000.0 / (1000.0 - 25.0);
776 const double kNetworkFreezeTimeMs = 5000.0;
777 const bool kGetAudioDuringFreezeRecovery = true;
778 const int kDelayToleranceMs = 20;
779 const int kMaxTimeToSpeechMs = 100;
780 LongCngWithClockDrift(kDriftFactor,
781 kNetworkFreezeTimeMs,
782 kGetAudioDuringFreezeRecovery,
787 TEST_F(NetEqDecodingTest, LongCngWithoutClockDrift) {
788 const double kDriftFactor = 1.0; // No drift.
789 const double kNetworkFreezeTimeMs = 0.0;
790 const bool kGetAudioDuringFreezeRecovery = false;
791 const int kDelayToleranceMs = 10;
792 const int kMaxTimeToSpeechMs = 50;
793 LongCngWithClockDrift(kDriftFactor,
794 kNetworkFreezeTimeMs,
795 kGetAudioDuringFreezeRecovery,
800 TEST_F(NetEqDecodingTest, UnknownPayloadType) {
801 const int kPayloadBytes = 100;
802 uint8_t payload[kPayloadBytes] = {0};
803 WebRtcRTPHeader rtp_info;
804 PopulateRtpInfo(0, 0, &rtp_info);
805 rtp_info.header.payloadType = 1; // Not registered as a decoder.
806 EXPECT_EQ(NetEq::kFail,
807 neteq_->InsertPacket(rtp_info, payload, kPayloadBytes, 0));
808 EXPECT_EQ(NetEq::kUnknownRtpPayloadType, neteq_->LastError());
811 TEST_F(NetEqDecodingTest, DISABLED_ON_ANDROID(DecoderError)) {
812 const int kPayloadBytes = 100;
813 uint8_t payload[kPayloadBytes] = {0};
814 WebRtcRTPHeader rtp_info;
815 PopulateRtpInfo(0, 0, &rtp_info);
816 rtp_info.header.payloadType = 103; // iSAC, but the payload is invalid.
817 EXPECT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes, 0));
818 NetEqOutputType type;
819 // Set all of |out_data_| to 1, and verify that it was set to 0 by the call
821 for (int i = 0; i < kMaxBlockSize; ++i) {
825 int samples_per_channel;
826 EXPECT_EQ(NetEq::kFail,
827 neteq_->GetAudio(kMaxBlockSize, out_data_,
828 &samples_per_channel, &num_channels, &type));
829 // Verify that there is a decoder error to check.
830 EXPECT_EQ(NetEq::kDecoderErrorCode, neteq_->LastError());
831 // Code 6730 is an iSAC error code.
832 EXPECT_EQ(6730, neteq_->LastDecoderError());
833 // Verify that the first 160 samples are set to 0, and that the remaining
834 // samples are left unmodified.
835 static const int kExpectedOutputLength = 160; // 10 ms at 16 kHz sample rate.
836 for (int i = 0; i < kExpectedOutputLength; ++i) {
837 std::ostringstream ss;
839 SCOPED_TRACE(ss.str()); // Print out the parameter values on failure.
840 EXPECT_EQ(0, out_data_[i]);
842 for (int i = kExpectedOutputLength; i < kMaxBlockSize; ++i) {
843 std::ostringstream ss;
845 SCOPED_TRACE(ss.str()); // Print out the parameter values on failure.
846 EXPECT_EQ(1, out_data_[i]);
850 TEST_F(NetEqDecodingTest, GetAudioBeforeInsertPacket) {
851 NetEqOutputType type;
852 // Set all of |out_data_| to 1, and verify that it was set to 0 by the call
854 for (int i = 0; i < kMaxBlockSize; ++i) {
858 int samples_per_channel;
859 EXPECT_EQ(0, neteq_->GetAudio(kMaxBlockSize, out_data_,
860 &samples_per_channel,
861 &num_channels, &type));
862 // Verify that the first block of samples is set to 0.
863 static const int kExpectedOutputLength =
864 kInitSampleRateHz / 100; // 10 ms at initial sample rate.
865 for (int i = 0; i < kExpectedOutputLength; ++i) {
866 std::ostringstream ss;
868 SCOPED_TRACE(ss.str()); // Print out the parameter values on failure.
869 EXPECT_EQ(0, out_data_[i]);
874 : public NetEqDecodingTest,
875 public ::testing::WithParamInterface<NetEq::BackgroundNoiseMode> {
877 NetEqBgnTest() : NetEqDecodingTest() {
878 config_.background_noise_mode = GetParam();
881 void CheckBgnOff(int sampling_rate_hz) {
882 int expected_samples_per_channel = 0;
883 uint8_t payload_type = 0xFF; // Invalid.
884 if (sampling_rate_hz == 8000) {
885 expected_samples_per_channel = kBlockSize8kHz;
886 payload_type = 93; // PCM 16, 8 kHz.
887 } else if (sampling_rate_hz == 16000) {
888 expected_samples_per_channel = kBlockSize16kHz;
889 payload_type = 94; // PCM 16, 16 kHZ.
890 } else if (sampling_rate_hz == 32000) {
891 expected_samples_per_channel = kBlockSize32kHz;
892 payload_type = 95; // PCM 16, 32 kHz.
894 ASSERT_TRUE(false); // Unsupported test case.
897 NetEqOutputType type;
898 int16_t output[kBlockSize32kHz]; // Maximum size is chosen.
899 int16_t input[kBlockSize32kHz]; // Maximum size is chosen.
901 // Payload of 10 ms of PCM16 32 kHz.
902 uint8_t payload[kBlockSize32kHz * sizeof(int16_t)];
905 for (int n = 0; n < expected_samples_per_channel; ++n) {
906 input[n] = (rand() & ((1 << 10) - 1)) - ((1 << 5) - 1);
909 WebRtcPcm16b_EncodeW16(input,
910 expected_samples_per_channel,
911 reinterpret_cast<int16_t*>(payload));
912 ASSERT_EQ(enc_len_bytes, expected_samples_per_channel * 2);
914 WebRtcRTPHeader rtp_info;
915 PopulateRtpInfo(0, 0, &rtp_info);
916 rtp_info.header.payloadType = payload_type;
918 int number_channels = 0;
919 int samples_per_channel = 0;
921 uint32_t receive_timestamp = 0;
922 for (int n = 0; n < 10; ++n) { // Insert few packets and get audio.
924 samples_per_channel = 0;
926 neteq_->InsertPacket(
927 rtp_info, payload, enc_len_bytes, receive_timestamp));
929 neteq_->GetAudio(kBlockSize32kHz,
931 &samples_per_channel,
934 ASSERT_EQ(1, number_channels);
935 ASSERT_EQ(expected_samples_per_channel, samples_per_channel);
936 ASSERT_EQ(kOutputNormal, type);
939 rtp_info.header.timestamp += expected_samples_per_channel;
940 rtp_info.header.sequenceNumber++;
941 receive_timestamp += expected_samples_per_channel;
945 samples_per_channel = 0;
947 // Get audio without inserting packets, expecting PLC and PLC-to-CNG. Pull
948 // one frame without checking speech-type. This is the first frame pulled
949 // without inserting any packet, and might not be labeled as PLC.
951 neteq_->GetAudio(kBlockSize32kHz,
953 &samples_per_channel,
956 ASSERT_EQ(1, number_channels);
957 ASSERT_EQ(expected_samples_per_channel, samples_per_channel);
959 // To be able to test the fading of background noise we need at lease to
961 const int kFadingThreshold = 611;
963 // Test several CNG-to-PLC packet for the expected behavior. The number 20
964 // is arbitrary, but sufficiently large to test enough number of frames.
965 const int kNumPlcToCngTestFrames = 20;
966 bool plc_to_cng = false;
967 for (int n = 0; n < kFadingThreshold + kNumPlcToCngTestFrames; ++n) {
969 samples_per_channel = 0;
970 memset(output, 1, sizeof(output)); // Set to non-zero.
972 neteq_->GetAudio(kBlockSize32kHz,
974 &samples_per_channel,
977 ASSERT_EQ(1, number_channels);
978 ASSERT_EQ(expected_samples_per_channel, samples_per_channel);
979 if (type == kOutputPLCtoCNG) {
981 double sum_squared = 0;
982 for (int k = 0; k < number_channels * samples_per_channel; ++k)
983 sum_squared += output[k] * output[k];
984 if (config_.background_noise_mode == NetEq::kBgnOn) {
985 EXPECT_NE(0, sum_squared);
986 } else if (config_.background_noise_mode == NetEq::kBgnOff ||
987 n > kFadingThreshold) {
988 EXPECT_EQ(0, sum_squared);
991 EXPECT_EQ(kOutputPLC, type);
994 EXPECT_TRUE(plc_to_cng); // Just to be sure that PLC-to-CNG has occurred.
998 TEST_P(NetEqBgnTest, BackgroundNoise) {
1004 INSTANTIATE_TEST_CASE_P(BgnModes,
1006 ::testing::Values(NetEq::kBgnOn,
1010 TEST_F(NetEqDecodingTest, SyncPacketInsert) {
1011 WebRtcRTPHeader rtp_info;
1012 uint32_t receive_timestamp = 0;
1013 // For the readability use the following payloads instead of the defaults of
1015 uint8_t kPcm16WbPayloadType = 1;
1016 uint8_t kCngNbPayloadType = 2;
1017 uint8_t kCngWbPayloadType = 3;
1018 uint8_t kCngSwb32PayloadType = 4;
1019 uint8_t kCngSwb48PayloadType = 5;
1020 uint8_t kAvtPayloadType = 6;
1021 uint8_t kRedPayloadType = 7;
1022 uint8_t kIsacPayloadType = 9; // Payload type 8 is already registered.
1024 // Register decoders.
1025 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderPCM16Bwb,
1026 kPcm16WbPayloadType));
1027 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderCNGnb, kCngNbPayloadType));
1028 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderCNGwb, kCngWbPayloadType));
1029 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderCNGswb32kHz,
1030 kCngSwb32PayloadType));
1031 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderCNGswb48kHz,
1032 kCngSwb48PayloadType));
1033 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderAVT, kAvtPayloadType));
1034 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderRED, kRedPayloadType));
1035 ASSERT_EQ(0, neteq_->RegisterPayloadType(kDecoderISAC, kIsacPayloadType));
1037 PopulateRtpInfo(0, 0, &rtp_info);
1038 rtp_info.header.payloadType = kPcm16WbPayloadType;
1040 // The first packet injected cannot be sync-packet.
1041 EXPECT_EQ(-1, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1043 // Payload length of 10 ms PCM16 16 kHz.
1044 const int kPayloadBytes = kBlockSize16kHz * sizeof(int16_t);
1045 uint8_t payload[kPayloadBytes] = {0};
1046 ASSERT_EQ(0, neteq_->InsertPacket(
1047 rtp_info, payload, kPayloadBytes, receive_timestamp));
1049 // Next packet. Last packet contained 10 ms audio.
1050 rtp_info.header.sequenceNumber++;
1051 rtp_info.header.timestamp += kBlockSize16kHz;
1052 receive_timestamp += kBlockSize16kHz;
1054 // Unacceptable payload types CNG, AVT (DTMF), RED.
1055 rtp_info.header.payloadType = kCngNbPayloadType;
1056 EXPECT_EQ(-1, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1058 rtp_info.header.payloadType = kCngWbPayloadType;
1059 EXPECT_EQ(-1, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1061 rtp_info.header.payloadType = kCngSwb32PayloadType;
1062 EXPECT_EQ(-1, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1064 rtp_info.header.payloadType = kCngSwb48PayloadType;
1065 EXPECT_EQ(-1, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1067 rtp_info.header.payloadType = kAvtPayloadType;
1068 EXPECT_EQ(-1, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1070 rtp_info.header.payloadType = kRedPayloadType;
1071 EXPECT_EQ(-1, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1073 // Change of codec cannot be initiated with a sync packet.
1074 rtp_info.header.payloadType = kIsacPayloadType;
1075 EXPECT_EQ(-1, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1077 // Change of SSRC is not allowed with a sync packet.
1078 rtp_info.header.payloadType = kPcm16WbPayloadType;
1079 ++rtp_info.header.ssrc;
1080 EXPECT_EQ(-1, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1082 --rtp_info.header.ssrc;
1083 EXPECT_EQ(0, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1086 // First insert several noise like packets, then sync-packets. Decoding all
1087 // packets should not produce error, statistics should not show any packet loss
1088 // and sync-packets should decode to zero.
1089 // TODO(turajs) we will have a better test if we have a referece NetEq, and
1090 // when Sync packets are inserted in "test" NetEq we insert all-zero payload
1091 // in reference NetEq and compare the output of those two.
1092 TEST_F(NetEqDecodingTest, SyncPacketDecode) {
1093 WebRtcRTPHeader rtp_info;
1094 PopulateRtpInfo(0, 0, &rtp_info);
1095 const int kPayloadBytes = kBlockSize16kHz * sizeof(int16_t);
1096 uint8_t payload[kPayloadBytes];
1097 int16_t decoded[kBlockSize16kHz];
1098 int algorithmic_frame_delay = algorithmic_delay_ms_ / 10 + 1;
1099 for (int n = 0; n < kPayloadBytes; ++n) {
1100 payload[n] = (rand() & 0xF0) + 1; // Non-zero random sequence.
1102 // Insert some packets which decode to noise. We are not interested in
1103 // actual decoded values.
1104 NetEqOutputType output_type;
1106 int samples_per_channel;
1107 uint32_t receive_timestamp = 0;
1108 for (int n = 0; n < 100; ++n) {
1109 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes,
1110 receive_timestamp));
1111 ASSERT_EQ(0, neteq_->GetAudio(kBlockSize16kHz, decoded,
1112 &samples_per_channel, &num_channels,
1114 ASSERT_EQ(kBlockSize16kHz, samples_per_channel);
1115 ASSERT_EQ(1, num_channels);
1117 rtp_info.header.sequenceNumber++;
1118 rtp_info.header.timestamp += kBlockSize16kHz;
1119 receive_timestamp += kBlockSize16kHz;
1121 const int kNumSyncPackets = 10;
1123 // Make sure sufficient number of sync packets are inserted that we can
1125 ASSERT_GT(kNumSyncPackets, algorithmic_frame_delay);
1126 // Insert sync-packets, the decoded sequence should be all-zero.
1127 for (int n = 0; n < kNumSyncPackets; ++n) {
1128 ASSERT_EQ(0, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1129 ASSERT_EQ(0, neteq_->GetAudio(kBlockSize16kHz, decoded,
1130 &samples_per_channel, &num_channels,
1132 ASSERT_EQ(kBlockSize16kHz, samples_per_channel);
1133 ASSERT_EQ(1, num_channels);
1134 if (n > algorithmic_frame_delay) {
1135 EXPECT_TRUE(IsAllZero(decoded, samples_per_channel * num_channels));
1137 rtp_info.header.sequenceNumber++;
1138 rtp_info.header.timestamp += kBlockSize16kHz;
1139 receive_timestamp += kBlockSize16kHz;
1142 // We insert regular packets, if sync packet are not correctly buffered then
1143 // network statistics would show some packet loss.
1144 for (int n = 0; n <= algorithmic_frame_delay + 10; ++n) {
1145 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes,
1146 receive_timestamp));
1147 ASSERT_EQ(0, neteq_->GetAudio(kBlockSize16kHz, decoded,
1148 &samples_per_channel, &num_channels,
1150 if (n >= algorithmic_frame_delay + 1) {
1151 // Expect that this frame contain samples from regular RTP.
1152 EXPECT_TRUE(IsAllNonZero(decoded, samples_per_channel * num_channels));
1154 rtp_info.header.sequenceNumber++;
1155 rtp_info.header.timestamp += kBlockSize16kHz;
1156 receive_timestamp += kBlockSize16kHz;
1158 NetEqNetworkStatistics network_stats;
1159 ASSERT_EQ(0, neteq_->NetworkStatistics(&network_stats));
1160 // Expecting a "clean" network.
1161 EXPECT_EQ(0, network_stats.packet_loss_rate);
1162 EXPECT_EQ(0, network_stats.expand_rate);
1163 EXPECT_EQ(0, network_stats.accelerate_rate);
1164 EXPECT_LE(network_stats.preemptive_rate, 150);
1167 // Test if the size of the packet buffer reported correctly when containing
1168 // sync packets. Also, test if network packets override sync packets. That is to
1169 // prefer decoding a network packet to a sync packet, if both have same sequence
1170 // number and timestamp.
1171 TEST_F(NetEqDecodingTest, SyncPacketBufferSizeAndOverridenByNetworkPackets) {
1172 WebRtcRTPHeader rtp_info;
1173 PopulateRtpInfo(0, 0, &rtp_info);
1174 const int kPayloadBytes = kBlockSize16kHz * sizeof(int16_t);
1175 uint8_t payload[kPayloadBytes];
1176 int16_t decoded[kBlockSize16kHz];
1177 for (int n = 0; n < kPayloadBytes; ++n) {
1178 payload[n] = (rand() & 0xF0) + 1; // Non-zero random sequence.
1180 // Insert some packets which decode to noise. We are not interested in
1181 // actual decoded values.
1182 NetEqOutputType output_type;
1184 int samples_per_channel;
1185 uint32_t receive_timestamp = 0;
1186 int algorithmic_frame_delay = algorithmic_delay_ms_ / 10 + 1;
1187 for (int n = 0; n < algorithmic_frame_delay; ++n) {
1188 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes,
1189 receive_timestamp));
1190 ASSERT_EQ(0, neteq_->GetAudio(kBlockSize16kHz, decoded,
1191 &samples_per_channel, &num_channels,
1193 ASSERT_EQ(kBlockSize16kHz, samples_per_channel);
1194 ASSERT_EQ(1, num_channels);
1195 rtp_info.header.sequenceNumber++;
1196 rtp_info.header.timestamp += kBlockSize16kHz;
1197 receive_timestamp += kBlockSize16kHz;
1199 const int kNumSyncPackets = 10;
1201 WebRtcRTPHeader first_sync_packet_rtp_info;
1202 memcpy(&first_sync_packet_rtp_info, &rtp_info, sizeof(rtp_info));
1204 // Insert sync-packets, but no decoding.
1205 for (int n = 0; n < kNumSyncPackets; ++n) {
1206 ASSERT_EQ(0, neteq_->InsertSyncPacket(rtp_info, receive_timestamp));
1207 rtp_info.header.sequenceNumber++;
1208 rtp_info.header.timestamp += kBlockSize16kHz;
1209 receive_timestamp += kBlockSize16kHz;
1211 NetEqNetworkStatistics network_stats;
1212 ASSERT_EQ(0, neteq_->NetworkStatistics(&network_stats));
1213 EXPECT_EQ(kNumSyncPackets * 10 + algorithmic_delay_ms_,
1214 network_stats.current_buffer_size_ms);
1216 // Rewind |rtp_info| to that of the first sync packet.
1217 memcpy(&rtp_info, &first_sync_packet_rtp_info, sizeof(rtp_info));
1220 for (int n = 0; n < kNumSyncPackets; ++n) {
1221 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes,
1222 receive_timestamp));
1223 rtp_info.header.sequenceNumber++;
1224 rtp_info.header.timestamp += kBlockSize16kHz;
1225 receive_timestamp += kBlockSize16kHz;
1229 for (int n = 0; n < kNumSyncPackets; ++n) {
1230 ASSERT_EQ(0, neteq_->GetAudio(kBlockSize16kHz, decoded,
1231 &samples_per_channel, &num_channels,
1233 ASSERT_EQ(kBlockSize16kHz, samples_per_channel);
1234 ASSERT_EQ(1, num_channels);
1235 EXPECT_TRUE(IsAllNonZero(decoded, samples_per_channel * num_channels));
1239 void NetEqDecodingTest::WrapTest(uint16_t start_seq_no,
1240 uint32_t start_timestamp,
1241 const std::set<uint16_t>& drop_seq_numbers,
1242 bool expect_seq_no_wrap,
1243 bool expect_timestamp_wrap) {
1244 uint16_t seq_no = start_seq_no;
1245 uint32_t timestamp = start_timestamp;
1246 const int kBlocksPerFrame = 3; // Number of 10 ms blocks per frame.
1247 const int kFrameSizeMs = kBlocksPerFrame * kTimeStepMs;
1248 const int kSamples = kBlockSize16kHz * kBlocksPerFrame;
1249 const int kPayloadBytes = kSamples * sizeof(int16_t);
1250 double next_input_time_ms = 0.0;
1251 int16_t decoded[kBlockSize16kHz];
1253 int samples_per_channel;
1254 NetEqOutputType output_type;
1255 uint32_t receive_timestamp = 0;
1257 // Insert speech for 2 seconds.
1258 const int kSpeechDurationMs = 2000;
1259 int packets_inserted = 0;
1260 uint16_t last_seq_no;
1261 uint32_t last_timestamp;
1262 bool timestamp_wrapped = false;
1263 bool seq_no_wrapped = false;
1264 for (double t_ms = 0; t_ms < kSpeechDurationMs; t_ms += 10) {
1265 // Each turn in this for loop is 10 ms.
1266 while (next_input_time_ms <= t_ms) {
1267 // Insert one 30 ms speech frame.
1268 uint8_t payload[kPayloadBytes] = {0};
1269 WebRtcRTPHeader rtp_info;
1270 PopulateRtpInfo(seq_no, timestamp, &rtp_info);
1271 if (drop_seq_numbers.find(seq_no) == drop_seq_numbers.end()) {
1272 // This sequence number was not in the set to drop. Insert it.
1274 neteq_->InsertPacket(rtp_info, payload, kPayloadBytes,
1275 receive_timestamp));
1278 NetEqNetworkStatistics network_stats;
1279 ASSERT_EQ(0, neteq_->NetworkStatistics(&network_stats));
1281 // Due to internal NetEq logic, preferred buffer-size is about 4 times the
1282 // packet size for first few packets. Therefore we refrain from checking
1284 if (packets_inserted > 4) {
1285 // Expect preferred and actual buffer size to be no more than 2 frames.
1286 EXPECT_LE(network_stats.preferred_buffer_size_ms, kFrameSizeMs * 2);
1287 EXPECT_LE(network_stats.current_buffer_size_ms, kFrameSizeMs * 2 +
1288 algorithmic_delay_ms_);
1290 last_seq_no = seq_no;
1291 last_timestamp = timestamp;
1294 timestamp += kSamples;
1295 receive_timestamp += kSamples;
1296 next_input_time_ms += static_cast<double>(kFrameSizeMs);
1298 seq_no_wrapped |= seq_no < last_seq_no;
1299 timestamp_wrapped |= timestamp < last_timestamp;
1301 // Pull out data once.
1302 ASSERT_EQ(0, neteq_->GetAudio(kBlockSize16kHz, decoded,
1303 &samples_per_channel, &num_channels,
1305 ASSERT_EQ(kBlockSize16kHz, samples_per_channel);
1306 ASSERT_EQ(1, num_channels);
1308 // Expect delay (in samples) to be less than 2 packets.
1309 EXPECT_LE(timestamp - PlayoutTimestamp(),
1310 static_cast<uint32_t>(kSamples * 2));
1312 // Make sure we have actually tested wrap-around.
1313 ASSERT_EQ(expect_seq_no_wrap, seq_no_wrapped);
1314 ASSERT_EQ(expect_timestamp_wrap, timestamp_wrapped);
1317 TEST_F(NetEqDecodingTest, SequenceNumberWrap) {
1318 // Start with a sequence number that will soon wrap.
1319 std::set<uint16_t> drop_seq_numbers; // Don't drop any packets.
1320 WrapTest(0xFFFF - 10, 0, drop_seq_numbers, true, false);
1323 TEST_F(NetEqDecodingTest, SequenceNumberWrapAndDrop) {
1324 // Start with a sequence number that will soon wrap.
1325 std::set<uint16_t> drop_seq_numbers;
1326 drop_seq_numbers.insert(0xFFFF);
1327 drop_seq_numbers.insert(0x0);
1328 WrapTest(0xFFFF - 10, 0, drop_seq_numbers, true, false);
1331 TEST_F(NetEqDecodingTest, TimestampWrap) {
1332 // Start with a timestamp that will soon wrap.
1333 std::set<uint16_t> drop_seq_numbers;
1334 WrapTest(0, 0xFFFFFFFF - 3000, drop_seq_numbers, false, true);
1337 TEST_F(NetEqDecodingTest, TimestampAndSequenceNumberWrap) {
1338 // Start with a timestamp and a sequence number that will wrap at the same
1340 std::set<uint16_t> drop_seq_numbers;
1341 WrapTest(0xFFFF - 10, 0xFFFFFFFF - 5000, drop_seq_numbers, true, true);
1344 void NetEqDecodingTest::DuplicateCng() {
1345 uint16_t seq_no = 0;
1346 uint32_t timestamp = 0;
1347 const int kFrameSizeMs = 10;
1348 const int kSampleRateKhz = 16;
1349 const int kSamples = kFrameSizeMs * kSampleRateKhz;
1350 const int kPayloadBytes = kSamples * 2;
1352 const int algorithmic_delay_samples = std::max(
1353 algorithmic_delay_ms_ * kSampleRateKhz, 5 * kSampleRateKhz / 8);
1354 // Insert three speech packet. Three are needed to get the frame length
1358 NetEqOutputType type;
1359 uint8_t payload[kPayloadBytes] = {0};
1360 WebRtcRTPHeader rtp_info;
1361 for (int i = 0; i < 3; ++i) {
1362 PopulateRtpInfo(seq_no, timestamp, &rtp_info);
1363 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes, 0));
1365 timestamp += kSamples;
1370 kMaxBlockSize, out_data_, &out_len, &num_channels, &type));
1371 ASSERT_EQ(kBlockSize16kHz, out_len);
1373 // Verify speech output.
1374 EXPECT_EQ(kOutputNormal, type);
1376 // Insert same CNG packet twice.
1377 const int kCngPeriodMs = 100;
1378 const int kCngPeriodSamples = kCngPeriodMs * kSampleRateKhz;
1380 PopulateCng(seq_no, timestamp, &rtp_info, payload, &payload_len);
1381 // This is the first time this CNG packet is inserted.
1382 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, payload_len, 0));
1384 // Pull audio once and make sure CNG is played.
1387 kMaxBlockSize, out_data_, &out_len, &num_channels, &type));
1388 ASSERT_EQ(kBlockSize16kHz, out_len);
1389 EXPECT_EQ(kOutputCNG, type);
1390 EXPECT_EQ(timestamp - algorithmic_delay_samples, PlayoutTimestamp());
1392 // Insert the same CNG packet again. Note that at this point it is old, since
1393 // we have already decoded the first copy of it.
1394 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, payload_len, 0));
1396 // Pull audio until we have played |kCngPeriodMs| of CNG. Start at 10 ms since
1397 // we have already pulled out CNG once.
1398 for (int cng_time_ms = 10; cng_time_ms < kCngPeriodMs; cng_time_ms += 10) {
1401 kMaxBlockSize, out_data_, &out_len, &num_channels, &type));
1402 ASSERT_EQ(kBlockSize16kHz, out_len);
1403 EXPECT_EQ(kOutputCNG, type);
1404 EXPECT_EQ(timestamp - algorithmic_delay_samples,
1405 PlayoutTimestamp());
1408 // Insert speech again.
1410 timestamp += kCngPeriodSamples;
1411 PopulateRtpInfo(seq_no, timestamp, &rtp_info);
1412 ASSERT_EQ(0, neteq_->InsertPacket(rtp_info, payload, kPayloadBytes, 0));
1414 // Pull audio once and verify that the output is speech again.
1417 kMaxBlockSize, out_data_, &out_len, &num_channels, &type));
1418 ASSERT_EQ(kBlockSize16kHz, out_len);
1419 EXPECT_EQ(kOutputNormal, type);
1420 EXPECT_EQ(timestamp + kSamples - algorithmic_delay_samples,
1421 PlayoutTimestamp());
1424 uint32_t NetEqDecodingTest::PlayoutTimestamp() {
1425 uint32_t playout_timestamp = 0;
1426 EXPECT_TRUE(neteq_->GetPlayoutTimestamp(&playout_timestamp));
1427 return playout_timestamp;
1430 TEST_F(NetEqDecodingTest, DiscardDuplicateCng) { DuplicateCng(); }
1431 } // namespace webrtc