1 /*
2 * Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
3 *
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.
9 */
10
11 #include "webrtc/modules/audio_coding/neteq/neteq_impl.h"
12
13 #include <assert.h>
14 #include <memory.h> // memset
15
16 #include <algorithm>
17
18 #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
19 #include "webrtc/modules/audio_coding/neteq/accelerate.h"
20 #include "webrtc/modules/audio_coding/neteq/background_noise.h"
21 #include "webrtc/modules/audio_coding/neteq/buffer_level_filter.h"
22 #include "webrtc/modules/audio_coding/neteq/comfort_noise.h"
23 #include "webrtc/modules/audio_coding/neteq/decision_logic.h"
24 #include "webrtc/modules/audio_coding/neteq/decoder_database.h"
25 #include "webrtc/modules/audio_coding/neteq/defines.h"
26 #include "webrtc/modules/audio_coding/neteq/delay_manager.h"
27 #include "webrtc/modules/audio_coding/neteq/delay_peak_detector.h"
28 #include "webrtc/modules/audio_coding/neteq/dtmf_buffer.h"
29 #include "webrtc/modules/audio_coding/neteq/dtmf_tone_generator.h"
30 #include "webrtc/modules/audio_coding/neteq/expand.h"
31 #include "webrtc/modules/audio_coding/neteq/interface/audio_decoder.h"
32 #include "webrtc/modules/audio_coding/neteq/merge.h"
33 #include "webrtc/modules/audio_coding/neteq/normal.h"
34 #include "webrtc/modules/audio_coding/neteq/packet_buffer.h"
35 #include "webrtc/modules/audio_coding/neteq/packet.h"
36 #include "webrtc/modules/audio_coding/neteq/payload_splitter.h"
37 #include "webrtc/modules/audio_coding/neteq/post_decode_vad.h"
38 #include "webrtc/modules/audio_coding/neteq/preemptive_expand.h"
39 #include "webrtc/modules/audio_coding/neteq/sync_buffer.h"
40 #include "webrtc/modules/audio_coding/neteq/timestamp_scaler.h"
41 #include "webrtc/modules/interface/module_common_types.h"
42 #include "webrtc/system_wrappers/interface/critical_section_wrapper.h"
43 #include "webrtc/system_wrappers/interface/logging.h"
44
45 // Modify the code to obtain backwards bit-exactness. Once bit-exactness is no
46 // longer required, this #define should be removed (and the code that it
47 // enables).
48 #define LEGACY_BITEXACT
49
50 namespace webrtc {
51
NetEqImpl(const NetEq::Config & config,BufferLevelFilter * buffer_level_filter,DecoderDatabase * decoder_database,DelayManager * delay_manager,DelayPeakDetector * delay_peak_detector,DtmfBuffer * dtmf_buffer,DtmfToneGenerator * dtmf_tone_generator,PacketBuffer * packet_buffer,PayloadSplitter * payload_splitter,TimestampScaler * timestamp_scaler,AccelerateFactory * accelerate_factory,ExpandFactory * expand_factory,PreemptiveExpandFactory * preemptive_expand_factory,bool create_components)52 NetEqImpl::NetEqImpl(const NetEq::Config& config,
53 BufferLevelFilter* buffer_level_filter,
54 DecoderDatabase* decoder_database,
55 DelayManager* delay_manager,
56 DelayPeakDetector* delay_peak_detector,
57 DtmfBuffer* dtmf_buffer,
58 DtmfToneGenerator* dtmf_tone_generator,
59 PacketBuffer* packet_buffer,
60 PayloadSplitter* payload_splitter,
61 TimestampScaler* timestamp_scaler,
62 AccelerateFactory* accelerate_factory,
63 ExpandFactory* expand_factory,
64 PreemptiveExpandFactory* preemptive_expand_factory,
65 bool create_components)
66 : crit_sect_(CriticalSectionWrapper::CreateCriticalSection()),
67 buffer_level_filter_(buffer_level_filter),
68 decoder_database_(decoder_database),
69 delay_manager_(delay_manager),
70 delay_peak_detector_(delay_peak_detector),
71 dtmf_buffer_(dtmf_buffer),
72 dtmf_tone_generator_(dtmf_tone_generator),
73 packet_buffer_(packet_buffer),
74 payload_splitter_(payload_splitter),
75 timestamp_scaler_(timestamp_scaler),
76 vad_(new PostDecodeVad()),
77 expand_factory_(expand_factory),
78 accelerate_factory_(accelerate_factory),
79 preemptive_expand_factory_(preemptive_expand_factory),
80 last_mode_(kModeNormal),
81 decoded_buffer_length_(kMaxFrameSize),
82 decoded_buffer_(new int16_t[decoded_buffer_length_]),
83 playout_timestamp_(0),
84 new_codec_(false),
85 timestamp_(0),
86 reset_decoder_(false),
87 current_rtp_payload_type_(0xFF), // Invalid RTP payload type.
88 current_cng_rtp_payload_type_(0xFF), // Invalid RTP payload type.
89 ssrc_(0),
90 first_packet_(true),
91 error_code_(0),
92 decoder_error_code_(0),
93 background_noise_mode_(config.background_noise_mode),
94 decoded_packet_sequence_number_(-1),
95 decoded_packet_timestamp_(0) {
96 int fs = config.sample_rate_hz;
97 if (fs != 8000 && fs != 16000 && fs != 32000 && fs != 48000) {
98 LOG(LS_ERROR) << "Sample rate " << fs << " Hz not supported. " <<
99 "Changing to 8000 Hz.";
100 fs = 8000;
101 }
102 LOG(LS_VERBOSE) << "Create NetEqImpl object with fs = " << fs << ".";
103 fs_hz_ = fs;
104 fs_mult_ = fs / 8000;
105 output_size_samples_ = kOutputSizeMs * 8 * fs_mult_;
106 decoder_frame_length_ = 3 * output_size_samples_;
107 WebRtcSpl_Init();
108 if (create_components) {
109 SetSampleRateAndChannels(fs, 1); // Default is 1 channel.
110 }
111 }
112
~NetEqImpl()113 NetEqImpl::~NetEqImpl() {
114 LOG(LS_INFO) << "Deleting NetEqImpl object.";
115 }
116
InsertPacket(const WebRtcRTPHeader & rtp_header,const uint8_t * payload,int length_bytes,uint32_t receive_timestamp)117 int NetEqImpl::InsertPacket(const WebRtcRTPHeader& rtp_header,
118 const uint8_t* payload,
119 int length_bytes,
120 uint32_t receive_timestamp) {
121 CriticalSectionScoped lock(crit_sect_.get());
122 LOG(LS_VERBOSE) << "InsertPacket: ts=" << rtp_header.header.timestamp <<
123 ", sn=" << rtp_header.header.sequenceNumber <<
124 ", pt=" << static_cast<int>(rtp_header.header.payloadType) <<
125 ", ssrc=" << rtp_header.header.ssrc <<
126 ", len=" << length_bytes;
127 int error = InsertPacketInternal(rtp_header, payload, length_bytes,
128 receive_timestamp, false);
129 if (error != 0) {
130 LOG_FERR1(LS_WARNING, InsertPacketInternal, error);
131 error_code_ = error;
132 return kFail;
133 }
134 return kOK;
135 }
136
InsertSyncPacket(const WebRtcRTPHeader & rtp_header,uint32_t receive_timestamp)137 int NetEqImpl::InsertSyncPacket(const WebRtcRTPHeader& rtp_header,
138 uint32_t receive_timestamp) {
139 CriticalSectionScoped lock(crit_sect_.get());
140 LOG(LS_VERBOSE) << "InsertPacket-Sync: ts="
141 << rtp_header.header.timestamp <<
142 ", sn=" << rtp_header.header.sequenceNumber <<
143 ", pt=" << static_cast<int>(rtp_header.header.payloadType) <<
144 ", ssrc=" << rtp_header.header.ssrc;
145
146 const uint8_t kSyncPayload[] = { 's', 'y', 'n', 'c' };
147 int error = InsertPacketInternal(
148 rtp_header, kSyncPayload, sizeof(kSyncPayload), receive_timestamp, true);
149
150 if (error != 0) {
151 LOG_FERR1(LS_WARNING, InsertPacketInternal, error);
152 error_code_ = error;
153 return kFail;
154 }
155 return kOK;
156 }
157
GetAudio(size_t max_length,int16_t * output_audio,int * samples_per_channel,int * num_channels,NetEqOutputType * type)158 int NetEqImpl::GetAudio(size_t max_length, int16_t* output_audio,
159 int* samples_per_channel, int* num_channels,
160 NetEqOutputType* type) {
161 CriticalSectionScoped lock(crit_sect_.get());
162 LOG(LS_VERBOSE) << "GetAudio";
163 int error = GetAudioInternal(max_length, output_audio, samples_per_channel,
164 num_channels);
165 LOG(LS_VERBOSE) << "Produced " << *samples_per_channel <<
166 " samples/channel for " << *num_channels << " channel(s)";
167 if (error != 0) {
168 LOG_FERR1(LS_WARNING, GetAudioInternal, error);
169 error_code_ = error;
170 return kFail;
171 }
172 if (type) {
173 *type = LastOutputType();
174 }
175 return kOK;
176 }
177
RegisterPayloadType(enum NetEqDecoder codec,uint8_t rtp_payload_type)178 int NetEqImpl::RegisterPayloadType(enum NetEqDecoder codec,
179 uint8_t rtp_payload_type) {
180 CriticalSectionScoped lock(crit_sect_.get());
181 LOG_API2(static_cast<int>(rtp_payload_type), codec);
182 int ret = decoder_database_->RegisterPayload(rtp_payload_type, codec);
183 if (ret != DecoderDatabase::kOK) {
184 LOG_FERR2(LS_WARNING, RegisterPayload, rtp_payload_type, codec);
185 switch (ret) {
186 case DecoderDatabase::kInvalidRtpPayloadType:
187 error_code_ = kInvalidRtpPayloadType;
188 break;
189 case DecoderDatabase::kCodecNotSupported:
190 error_code_ = kCodecNotSupported;
191 break;
192 case DecoderDatabase::kDecoderExists:
193 error_code_ = kDecoderExists;
194 break;
195 default:
196 error_code_ = kOtherError;
197 }
198 return kFail;
199 }
200 return kOK;
201 }
202
RegisterExternalDecoder(AudioDecoder * decoder,enum NetEqDecoder codec,uint8_t rtp_payload_type)203 int NetEqImpl::RegisterExternalDecoder(AudioDecoder* decoder,
204 enum NetEqDecoder codec,
205 uint8_t rtp_payload_type) {
206 CriticalSectionScoped lock(crit_sect_.get());
207 LOG_API2(static_cast<int>(rtp_payload_type), codec);
208 if (!decoder) {
209 LOG(LS_ERROR) << "Cannot register external decoder with NULL pointer";
210 assert(false);
211 return kFail;
212 }
213 const int sample_rate_hz = AudioDecoder::CodecSampleRateHz(codec);
214 int ret = decoder_database_->InsertExternal(rtp_payload_type, codec,
215 sample_rate_hz, decoder);
216 if (ret != DecoderDatabase::kOK) {
217 LOG_FERR2(LS_WARNING, InsertExternal, rtp_payload_type, codec);
218 switch (ret) {
219 case DecoderDatabase::kInvalidRtpPayloadType:
220 error_code_ = kInvalidRtpPayloadType;
221 break;
222 case DecoderDatabase::kCodecNotSupported:
223 error_code_ = kCodecNotSupported;
224 break;
225 case DecoderDatabase::kDecoderExists:
226 error_code_ = kDecoderExists;
227 break;
228 case DecoderDatabase::kInvalidSampleRate:
229 error_code_ = kInvalidSampleRate;
230 break;
231 case DecoderDatabase::kInvalidPointer:
232 error_code_ = kInvalidPointer;
233 break;
234 default:
235 error_code_ = kOtherError;
236 }
237 return kFail;
238 }
239 return kOK;
240 }
241
RemovePayloadType(uint8_t rtp_payload_type)242 int NetEqImpl::RemovePayloadType(uint8_t rtp_payload_type) {
243 CriticalSectionScoped lock(crit_sect_.get());
244 LOG_API1(static_cast<int>(rtp_payload_type));
245 int ret = decoder_database_->Remove(rtp_payload_type);
246 if (ret == DecoderDatabase::kOK) {
247 return kOK;
248 } else if (ret == DecoderDatabase::kDecoderNotFound) {
249 error_code_ = kDecoderNotFound;
250 } else {
251 error_code_ = kOtherError;
252 }
253 LOG_FERR1(LS_WARNING, Remove, rtp_payload_type);
254 return kFail;
255 }
256
SetMinimumDelay(int delay_ms)257 bool NetEqImpl::SetMinimumDelay(int delay_ms) {
258 CriticalSectionScoped lock(crit_sect_.get());
259 if (delay_ms >= 0 && delay_ms < 10000) {
260 assert(delay_manager_.get());
261 return delay_manager_->SetMinimumDelay(delay_ms);
262 }
263 return false;
264 }
265
SetMaximumDelay(int delay_ms)266 bool NetEqImpl::SetMaximumDelay(int delay_ms) {
267 CriticalSectionScoped lock(crit_sect_.get());
268 if (delay_ms >= 0 && delay_ms < 10000) {
269 assert(delay_manager_.get());
270 return delay_manager_->SetMaximumDelay(delay_ms);
271 }
272 return false;
273 }
274
LeastRequiredDelayMs() const275 int NetEqImpl::LeastRequiredDelayMs() const {
276 CriticalSectionScoped lock(crit_sect_.get());
277 assert(delay_manager_.get());
278 return delay_manager_->least_required_delay_ms();
279 }
280
SetPlayoutMode(NetEqPlayoutMode mode)281 void NetEqImpl::SetPlayoutMode(NetEqPlayoutMode mode) {
282 CriticalSectionScoped lock(crit_sect_.get());
283 if (!decision_logic_.get() || mode != decision_logic_->playout_mode()) {
284 // The reset() method calls delete for the old object.
285 CreateDecisionLogic(mode);
286 }
287 }
288
PlayoutMode() const289 NetEqPlayoutMode NetEqImpl::PlayoutMode() const {
290 CriticalSectionScoped lock(crit_sect_.get());
291 assert(decision_logic_.get());
292 return decision_logic_->playout_mode();
293 }
294
NetworkStatistics(NetEqNetworkStatistics * stats)295 int NetEqImpl::NetworkStatistics(NetEqNetworkStatistics* stats) {
296 CriticalSectionScoped lock(crit_sect_.get());
297 assert(decoder_database_.get());
298 const int total_samples_in_buffers = packet_buffer_->NumSamplesInBuffer(
299 decoder_database_.get(), decoder_frame_length_) +
300 static_cast<int>(sync_buffer_->FutureLength());
301 assert(delay_manager_.get());
302 assert(decision_logic_.get());
303 stats_.GetNetworkStatistics(fs_hz_, total_samples_in_buffers,
304 decoder_frame_length_, *delay_manager_.get(),
305 *decision_logic_.get(), stats);
306 return 0;
307 }
308
WaitingTimes(std::vector<int> * waiting_times)309 void NetEqImpl::WaitingTimes(std::vector<int>* waiting_times) {
310 CriticalSectionScoped lock(crit_sect_.get());
311 stats_.WaitingTimes(waiting_times);
312 }
313
GetRtcpStatistics(RtcpStatistics * stats)314 void NetEqImpl::GetRtcpStatistics(RtcpStatistics* stats) {
315 CriticalSectionScoped lock(crit_sect_.get());
316 if (stats) {
317 rtcp_.GetStatistics(false, stats);
318 }
319 }
320
GetRtcpStatisticsNoReset(RtcpStatistics * stats)321 void NetEqImpl::GetRtcpStatisticsNoReset(RtcpStatistics* stats) {
322 CriticalSectionScoped lock(crit_sect_.get());
323 if (stats) {
324 rtcp_.GetStatistics(true, stats);
325 }
326 }
327
EnableVad()328 void NetEqImpl::EnableVad() {
329 CriticalSectionScoped lock(crit_sect_.get());
330 assert(vad_.get());
331 vad_->Enable();
332 }
333
DisableVad()334 void NetEqImpl::DisableVad() {
335 CriticalSectionScoped lock(crit_sect_.get());
336 assert(vad_.get());
337 vad_->Disable();
338 }
339
GetPlayoutTimestamp(uint32_t * timestamp)340 bool NetEqImpl::GetPlayoutTimestamp(uint32_t* timestamp) {
341 CriticalSectionScoped lock(crit_sect_.get());
342 if (first_packet_) {
343 // We don't have a valid RTP timestamp until we have decoded our first
344 // RTP packet.
345 return false;
346 }
347 *timestamp = timestamp_scaler_->ToExternal(playout_timestamp_);
348 return true;
349 }
350
LastError()351 int NetEqImpl::LastError() {
352 CriticalSectionScoped lock(crit_sect_.get());
353 return error_code_;
354 }
355
LastDecoderError()356 int NetEqImpl::LastDecoderError() {
357 CriticalSectionScoped lock(crit_sect_.get());
358 return decoder_error_code_;
359 }
360
FlushBuffers()361 void NetEqImpl::FlushBuffers() {
362 CriticalSectionScoped lock(crit_sect_.get());
363 LOG_API0();
364 packet_buffer_->Flush();
365 assert(sync_buffer_.get());
366 assert(expand_.get());
367 sync_buffer_->Flush();
368 sync_buffer_->set_next_index(sync_buffer_->next_index() -
369 expand_->overlap_length());
370 // Set to wait for new codec.
371 first_packet_ = true;
372 }
373
PacketBufferStatistics(int * current_num_packets,int * max_num_packets) const374 void NetEqImpl::PacketBufferStatistics(int* current_num_packets,
375 int* max_num_packets) const {
376 CriticalSectionScoped lock(crit_sect_.get());
377 packet_buffer_->BufferStat(current_num_packets, max_num_packets);
378 }
379
DecodedRtpInfo(int * sequence_number,uint32_t * timestamp) const380 int NetEqImpl::DecodedRtpInfo(int* sequence_number, uint32_t* timestamp) const {
381 CriticalSectionScoped lock(crit_sect_.get());
382 if (decoded_packet_sequence_number_ < 0)
383 return -1;
384 *sequence_number = decoded_packet_sequence_number_;
385 *timestamp = decoded_packet_timestamp_;
386 return 0;
387 }
388
sync_buffer_for_test() const389 const SyncBuffer* NetEqImpl::sync_buffer_for_test() const {
390 CriticalSectionScoped lock(crit_sect_.get());
391 return sync_buffer_.get();
392 }
393
394 // Methods below this line are private.
395
InsertPacketInternal(const WebRtcRTPHeader & rtp_header,const uint8_t * payload,int length_bytes,uint32_t receive_timestamp,bool is_sync_packet)396 int NetEqImpl::InsertPacketInternal(const WebRtcRTPHeader& rtp_header,
397 const uint8_t* payload,
398 int length_bytes,
399 uint32_t receive_timestamp,
400 bool is_sync_packet) {
401 if (!payload) {
402 LOG_F(LS_ERROR) << "payload == NULL";
403 return kInvalidPointer;
404 }
405 // Sanity checks for sync-packets.
406 if (is_sync_packet) {
407 if (decoder_database_->IsDtmf(rtp_header.header.payloadType) ||
408 decoder_database_->IsRed(rtp_header.header.payloadType) ||
409 decoder_database_->IsComfortNoise(rtp_header.header.payloadType)) {
410 LOG_F(LS_ERROR) << "Sync-packet with an unacceptable payload type "
411 << rtp_header.header.payloadType;
412 return kSyncPacketNotAccepted;
413 }
414 if (first_packet_ ||
415 rtp_header.header.payloadType != current_rtp_payload_type_ ||
416 rtp_header.header.ssrc != ssrc_) {
417 // Even if |current_rtp_payload_type_| is 0xFF, sync-packet isn't
418 // accepted.
419 LOG_F(LS_ERROR) << "Changing codec, SSRC or first packet "
420 "with sync-packet.";
421 return kSyncPacketNotAccepted;
422 }
423 }
424 PacketList packet_list;
425 RTPHeader main_header;
426 {
427 // Convert to Packet.
428 // Create |packet| within this separate scope, since it should not be used
429 // directly once it's been inserted in the packet list. This way, |packet|
430 // is not defined outside of this block.
431 Packet* packet = new Packet;
432 packet->header.markerBit = false;
433 packet->header.payloadType = rtp_header.header.payloadType;
434 packet->header.sequenceNumber = rtp_header.header.sequenceNumber;
435 packet->header.timestamp = rtp_header.header.timestamp;
436 packet->header.ssrc = rtp_header.header.ssrc;
437 packet->header.numCSRCs = 0;
438 packet->payload_length = length_bytes;
439 packet->primary = true;
440 packet->waiting_time = 0;
441 packet->payload = new uint8_t[packet->payload_length];
442 packet->sync_packet = is_sync_packet;
443 if (!packet->payload) {
444 LOG_F(LS_ERROR) << "Payload pointer is NULL.";
445 }
446 assert(payload); // Already checked above.
447 memcpy(packet->payload, payload, packet->payload_length);
448 // Insert packet in a packet list.
449 packet_list.push_back(packet);
450 // Save main payloads header for later.
451 memcpy(&main_header, &packet->header, sizeof(main_header));
452 }
453
454 bool update_sample_rate_and_channels = false;
455 // Reinitialize NetEq if it's needed (changed SSRC or first call).
456 if ((main_header.ssrc != ssrc_) || first_packet_) {
457 rtcp_.Init(main_header.sequenceNumber);
458 first_packet_ = false;
459
460 // Flush the packet buffer and DTMF buffer.
461 packet_buffer_->Flush();
462 dtmf_buffer_->Flush();
463
464 // Store new SSRC.
465 ssrc_ = main_header.ssrc;
466
467 // Update audio buffer timestamp.
468 sync_buffer_->IncreaseEndTimestamp(main_header.timestamp - timestamp_);
469
470 // Update codecs.
471 timestamp_ = main_header.timestamp;
472 current_rtp_payload_type_ = main_header.payloadType;
473
474 // Set MCU to update codec on next SignalMCU call.
475 new_codec_ = true;
476
477 // Reset timestamp scaling.
478 timestamp_scaler_->Reset();
479
480 // Triger an update of sampling rate and the number of channels.
481 update_sample_rate_and_channels = true;
482 }
483
484 // Update RTCP statistics, only for regular packets.
485 if (!is_sync_packet)
486 rtcp_.Update(main_header, receive_timestamp);
487
488 // Check for RED payload type, and separate payloads into several packets.
489 if (decoder_database_->IsRed(main_header.payloadType)) {
490 assert(!is_sync_packet); // We had a sanity check for this.
491 if (payload_splitter_->SplitRed(&packet_list) != PayloadSplitter::kOK) {
492 LOG_FERR1(LS_WARNING, SplitRed, packet_list.size());
493 PacketBuffer::DeleteAllPackets(&packet_list);
494 return kRedundancySplitError;
495 }
496 // Only accept a few RED payloads of the same type as the main data,
497 // DTMF events and CNG.
498 payload_splitter_->CheckRedPayloads(&packet_list, *decoder_database_);
499 // Update the stored main payload header since the main payload has now
500 // changed.
501 memcpy(&main_header, &packet_list.front()->header, sizeof(main_header));
502 }
503
504 // Check payload types.
505 if (decoder_database_->CheckPayloadTypes(packet_list) ==
506 DecoderDatabase::kDecoderNotFound) {
507 LOG_FERR1(LS_WARNING, CheckPayloadTypes, packet_list.size());
508 PacketBuffer::DeleteAllPackets(&packet_list);
509 return kUnknownRtpPayloadType;
510 }
511
512 // Scale timestamp to internal domain (only for some codecs).
513 timestamp_scaler_->ToInternal(&packet_list);
514
515 // Process DTMF payloads. Cycle through the list of packets, and pick out any
516 // DTMF payloads found.
517 PacketList::iterator it = packet_list.begin();
518 while (it != packet_list.end()) {
519 Packet* current_packet = (*it);
520 assert(current_packet);
521 assert(current_packet->payload);
522 if (decoder_database_->IsDtmf(current_packet->header.payloadType)) {
523 assert(!current_packet->sync_packet); // We had a sanity check for this.
524 DtmfEvent event;
525 int ret = DtmfBuffer::ParseEvent(
526 current_packet->header.timestamp,
527 current_packet->payload,
528 current_packet->payload_length,
529 &event);
530 if (ret != DtmfBuffer::kOK) {
531 LOG_FERR2(LS_WARNING, ParseEvent, ret,
532 current_packet->payload_length);
533 PacketBuffer::DeleteAllPackets(&packet_list);
534 return kDtmfParsingError;
535 }
536 if (dtmf_buffer_->InsertEvent(event) != DtmfBuffer::kOK) {
537 LOG_FERR0(LS_WARNING, InsertEvent);
538 PacketBuffer::DeleteAllPackets(&packet_list);
539 return kDtmfInsertError;
540 }
541 // TODO(hlundin): Let the destructor of Packet handle the payload.
542 delete [] current_packet->payload;
543 delete current_packet;
544 it = packet_list.erase(it);
545 } else {
546 ++it;
547 }
548 }
549
550 // Check for FEC in packets, and separate payloads into several packets.
551 int ret = payload_splitter_->SplitFec(&packet_list, decoder_database_.get());
552 if (ret != PayloadSplitter::kOK) {
553 LOG_FERR1(LS_WARNING, SplitFec, packet_list.size());
554 PacketBuffer::DeleteAllPackets(&packet_list);
555 switch (ret) {
556 case PayloadSplitter::kUnknownPayloadType:
557 return kUnknownRtpPayloadType;
558 default:
559 return kOtherError;
560 }
561 }
562
563 // Split payloads into smaller chunks. This also verifies that all payloads
564 // are of a known payload type. SplitAudio() method is protected against
565 // sync-packets.
566 ret = payload_splitter_->SplitAudio(&packet_list, *decoder_database_);
567 if (ret != PayloadSplitter::kOK) {
568 LOG_FERR1(LS_WARNING, SplitAudio, packet_list.size());
569 PacketBuffer::DeleteAllPackets(&packet_list);
570 switch (ret) {
571 case PayloadSplitter::kUnknownPayloadType:
572 return kUnknownRtpPayloadType;
573 case PayloadSplitter::kFrameSplitError:
574 return kFrameSplitError;
575 default:
576 return kOtherError;
577 }
578 }
579
580 // Update bandwidth estimate, if the packet is not sync-packet.
581 if (!packet_list.empty() && !packet_list.front()->sync_packet) {
582 // The list can be empty here if we got nothing but DTMF payloads.
583 AudioDecoder* decoder =
584 decoder_database_->GetDecoder(main_header.payloadType);
585 assert(decoder); // Should always get a valid object, since we have
586 // already checked that the payload types are known.
587 decoder->IncomingPacket(packet_list.front()->payload,
588 packet_list.front()->payload_length,
589 packet_list.front()->header.sequenceNumber,
590 packet_list.front()->header.timestamp,
591 receive_timestamp);
592 }
593
594 // Insert packets in buffer.
595 int temp_bufsize = packet_buffer_->NumPacketsInBuffer();
596 ret = packet_buffer_->InsertPacketList(
597 &packet_list,
598 *decoder_database_,
599 ¤t_rtp_payload_type_,
600 ¤t_cng_rtp_payload_type_);
601 if (ret == PacketBuffer::kFlushed) {
602 // Reset DSP timestamp etc. if packet buffer flushed.
603 new_codec_ = true;
604 update_sample_rate_and_channels = true;
605 LOG_F(LS_WARNING) << "Packet buffer flushed";
606 } else if (ret != PacketBuffer::kOK) {
607 LOG_FERR1(LS_WARNING, InsertPacketList, packet_list.size());
608 PacketBuffer::DeleteAllPackets(&packet_list);
609 return kOtherError;
610 }
611 if (current_rtp_payload_type_ != 0xFF) {
612 const DecoderDatabase::DecoderInfo* dec_info =
613 decoder_database_->GetDecoderInfo(current_rtp_payload_type_);
614 if (!dec_info) {
615 assert(false); // Already checked that the payload type is known.
616 }
617 }
618
619 if (update_sample_rate_and_channels && !packet_buffer_->Empty()) {
620 // We do not use |current_rtp_payload_type_| to |set payload_type|, but
621 // get the next RTP header from |packet_buffer_| to obtain the payload type.
622 // The reason for it is the following corner case. If NetEq receives a
623 // CNG packet with a sample rate different than the current CNG then it
624 // flushes its buffer, assuming send codec must have been changed. However,
625 // payload type of the hypothetically new send codec is not known.
626 const RTPHeader* rtp_header = packet_buffer_->NextRtpHeader();
627 assert(rtp_header);
628 int payload_type = rtp_header->payloadType;
629 AudioDecoder* decoder = decoder_database_->GetDecoder(payload_type);
630 assert(decoder); // Payloads are already checked to be valid.
631 const DecoderDatabase::DecoderInfo* decoder_info =
632 decoder_database_->GetDecoderInfo(payload_type);
633 assert(decoder_info);
634 if (decoder_info->fs_hz != fs_hz_ ||
635 decoder->channels() != algorithm_buffer_->Channels())
636 SetSampleRateAndChannels(decoder_info->fs_hz, decoder->channels());
637 }
638
639 // TODO(hlundin): Move this code to DelayManager class.
640 const DecoderDatabase::DecoderInfo* dec_info =
641 decoder_database_->GetDecoderInfo(main_header.payloadType);
642 assert(dec_info); // Already checked that the payload type is known.
643 delay_manager_->LastDecoderType(dec_info->codec_type);
644 if (delay_manager_->last_pack_cng_or_dtmf() == 0) {
645 // Calculate the total speech length carried in each packet.
646 temp_bufsize = packet_buffer_->NumPacketsInBuffer() - temp_bufsize;
647 temp_bufsize *= decoder_frame_length_;
648
649 if ((temp_bufsize > 0) &&
650 (temp_bufsize != decision_logic_->packet_length_samples())) {
651 decision_logic_->set_packet_length_samples(temp_bufsize);
652 delay_manager_->SetPacketAudioLength((1000 * temp_bufsize) / fs_hz_);
653 }
654
655 // Update statistics.
656 if ((int32_t) (main_header.timestamp - timestamp_) >= 0 &&
657 !new_codec_) {
658 // Only update statistics if incoming packet is not older than last played
659 // out packet, and if new codec flag is not set.
660 delay_manager_->Update(main_header.sequenceNumber, main_header.timestamp,
661 fs_hz_);
662 }
663 } else if (delay_manager_->last_pack_cng_or_dtmf() == -1) {
664 // This is first "normal" packet after CNG or DTMF.
665 // Reset packet time counter and measure time until next packet,
666 // but don't update statistics.
667 delay_manager_->set_last_pack_cng_or_dtmf(0);
668 delay_manager_->ResetPacketIatCount();
669 }
670 return 0;
671 }
672
GetAudioInternal(size_t max_length,int16_t * output,int * samples_per_channel,int * num_channels)673 int NetEqImpl::GetAudioInternal(size_t max_length, int16_t* output,
674 int* samples_per_channel, int* num_channels) {
675 PacketList packet_list;
676 DtmfEvent dtmf_event;
677 Operations operation;
678 bool play_dtmf;
679 int return_value = GetDecision(&operation, &packet_list, &dtmf_event,
680 &play_dtmf);
681 if (return_value != 0) {
682 LOG_FERR1(LS_WARNING, GetDecision, return_value);
683 assert(false);
684 last_mode_ = kModeError;
685 return return_value;
686 }
687 LOG(LS_VERBOSE) << "GetDecision returned operation=" << operation <<
688 " and " << packet_list.size() << " packet(s)";
689
690 AudioDecoder::SpeechType speech_type;
691 int length = 0;
692 int decode_return_value = Decode(&packet_list, &operation,
693 &length, &speech_type);
694
695 assert(vad_.get());
696 bool sid_frame_available =
697 (operation == kRfc3389Cng && !packet_list.empty());
698 vad_->Update(decoded_buffer_.get(), length, speech_type,
699 sid_frame_available, fs_hz_);
700
701 algorithm_buffer_->Clear();
702 switch (operation) {
703 case kNormal: {
704 DoNormal(decoded_buffer_.get(), length, speech_type, play_dtmf);
705 break;
706 }
707 case kMerge: {
708 DoMerge(decoded_buffer_.get(), length, speech_type, play_dtmf);
709 break;
710 }
711 case kExpand: {
712 return_value = DoExpand(play_dtmf);
713 break;
714 }
715 case kAccelerate: {
716 return_value = DoAccelerate(decoded_buffer_.get(), length, speech_type,
717 play_dtmf);
718 break;
719 }
720 case kPreemptiveExpand: {
721 return_value = DoPreemptiveExpand(decoded_buffer_.get(), length,
722 speech_type, play_dtmf);
723 break;
724 }
725 case kRfc3389Cng:
726 case kRfc3389CngNoPacket: {
727 return_value = DoRfc3389Cng(&packet_list, play_dtmf);
728 break;
729 }
730 case kCodecInternalCng: {
731 // This handles the case when there is no transmission and the decoder
732 // should produce internal comfort noise.
733 // TODO(hlundin): Write test for codec-internal CNG.
734 DoCodecInternalCng();
735 break;
736 }
737 case kDtmf: {
738 // TODO(hlundin): Write test for this.
739 return_value = DoDtmf(dtmf_event, &play_dtmf);
740 break;
741 }
742 case kAlternativePlc: {
743 // TODO(hlundin): Write test for this.
744 DoAlternativePlc(false);
745 break;
746 }
747 case kAlternativePlcIncreaseTimestamp: {
748 // TODO(hlundin): Write test for this.
749 DoAlternativePlc(true);
750 break;
751 }
752 case kAudioRepetitionIncreaseTimestamp: {
753 // TODO(hlundin): Write test for this.
754 sync_buffer_->IncreaseEndTimestamp(output_size_samples_);
755 // Skipping break on purpose. Execution should move on into the
756 // next case.
757 }
758 case kAudioRepetition: {
759 // TODO(hlundin): Write test for this.
760 // Copy last |output_size_samples_| from |sync_buffer_| to
761 // |algorithm_buffer|.
762 algorithm_buffer_->PushBackFromIndex(
763 *sync_buffer_, sync_buffer_->Size() - output_size_samples_);
764 expand_->Reset();
765 break;
766 }
767 case kUndefined: {
768 LOG_F(LS_ERROR) << "Invalid operation kUndefined.";
769 assert(false); // This should not happen.
770 last_mode_ = kModeError;
771 return kInvalidOperation;
772 }
773 } // End of switch.
774 if (return_value < 0) {
775 return return_value;
776 }
777
778 if (last_mode_ != kModeRfc3389Cng) {
779 comfort_noise_->Reset();
780 }
781
782 // Copy from |algorithm_buffer| to |sync_buffer_|.
783 sync_buffer_->PushBack(*algorithm_buffer_);
784
785 // Extract data from |sync_buffer_| to |output|.
786 size_t num_output_samples_per_channel = output_size_samples_;
787 size_t num_output_samples = output_size_samples_ * sync_buffer_->Channels();
788 if (num_output_samples > max_length) {
789 LOG(LS_WARNING) << "Output array is too short. " << max_length << " < " <<
790 output_size_samples_ << " * " << sync_buffer_->Channels();
791 num_output_samples = max_length;
792 num_output_samples_per_channel = static_cast<int>(
793 max_length / sync_buffer_->Channels());
794 }
795 int samples_from_sync = static_cast<int>(
796 sync_buffer_->GetNextAudioInterleaved(num_output_samples_per_channel,
797 output));
798 *num_channels = static_cast<int>(sync_buffer_->Channels());
799 LOG(LS_VERBOSE) << "Sync buffer (" << *num_channels << " channel(s)):" <<
800 " insert " << algorithm_buffer_->Size() << " samples, extract " <<
801 samples_from_sync << " samples";
802 if (samples_from_sync != output_size_samples_) {
803 LOG_F(LS_ERROR) << "samples_from_sync != output_size_samples_";
804 // TODO(minyue): treatment of under-run, filling zeros
805 memset(output, 0, num_output_samples * sizeof(int16_t));
806 *samples_per_channel = output_size_samples_;
807 return kSampleUnderrun;
808 }
809 *samples_per_channel = output_size_samples_;
810
811 // Should always have overlap samples left in the |sync_buffer_|.
812 assert(sync_buffer_->FutureLength() >= expand_->overlap_length());
813
814 if (play_dtmf) {
815 return_value = DtmfOverdub(dtmf_event, sync_buffer_->Channels(), output);
816 }
817
818 // Update the background noise parameters if last operation wrote data
819 // straight from the decoder to the |sync_buffer_|. That is, none of the
820 // operations that modify the signal can be followed by a parameter update.
821 if ((last_mode_ == kModeNormal) ||
822 (last_mode_ == kModeAccelerateFail) ||
823 (last_mode_ == kModePreemptiveExpandFail) ||
824 (last_mode_ == kModeRfc3389Cng) ||
825 (last_mode_ == kModeCodecInternalCng)) {
826 background_noise_->Update(*sync_buffer_, *vad_.get());
827 }
828
829 if (operation == kDtmf) {
830 // DTMF data was written the end of |sync_buffer_|.
831 // Update index to end of DTMF data in |sync_buffer_|.
832 sync_buffer_->set_dtmf_index(sync_buffer_->Size());
833 }
834
835 if (last_mode_ != kModeExpand) {
836 // If last operation was not expand, calculate the |playout_timestamp_| from
837 // the |sync_buffer_|. However, do not update the |playout_timestamp_| if it
838 // would be moved "backwards".
839 uint32_t temp_timestamp = sync_buffer_->end_timestamp() -
840 static_cast<uint32_t>(sync_buffer_->FutureLength());
841 if (static_cast<int32_t>(temp_timestamp - playout_timestamp_) > 0) {
842 playout_timestamp_ = temp_timestamp;
843 }
844 } else {
845 // Use dead reckoning to estimate the |playout_timestamp_|.
846 playout_timestamp_ += output_size_samples_;
847 }
848
849 if (decode_return_value) return decode_return_value;
850 return return_value;
851 }
852
GetDecision(Operations * operation,PacketList * packet_list,DtmfEvent * dtmf_event,bool * play_dtmf)853 int NetEqImpl::GetDecision(Operations* operation,
854 PacketList* packet_list,
855 DtmfEvent* dtmf_event,
856 bool* play_dtmf) {
857 // Initialize output variables.
858 *play_dtmf = false;
859 *operation = kUndefined;
860
861 // Increment time counters.
862 packet_buffer_->IncrementWaitingTimes();
863 stats_.IncreaseCounter(output_size_samples_, fs_hz_);
864
865 assert(sync_buffer_.get());
866 uint32_t end_timestamp = sync_buffer_->end_timestamp();
867 const RTPHeader* header = packet_buffer_->NextRtpHeader();
868
869 if (decision_logic_->CngRfc3389On() || last_mode_ == kModeRfc3389Cng) {
870 // Because of timestamp peculiarities, we have to "manually" disallow using
871 // a CNG packet with the same timestamp as the one that was last played.
872 // This can happen when using redundancy and will cause the timing to shift.
873 while (header && decoder_database_->IsComfortNoise(header->payloadType) &&
874 (end_timestamp >= header->timestamp ||
875 end_timestamp + decision_logic_->generated_noise_samples() >
876 header->timestamp)) {
877 // Don't use this packet, discard it.
878 if (packet_buffer_->DiscardNextPacket() != PacketBuffer::kOK) {
879 assert(false); // Must be ok by design.
880 }
881 // Check buffer again.
882 if (!new_codec_) {
883 packet_buffer_->DiscardOldPackets(end_timestamp);
884 }
885 header = packet_buffer_->NextRtpHeader();
886 }
887 }
888
889 assert(expand_.get());
890 const int samples_left = static_cast<int>(sync_buffer_->FutureLength() -
891 expand_->overlap_length());
892 if (last_mode_ == kModeAccelerateSuccess ||
893 last_mode_ == kModeAccelerateLowEnergy ||
894 last_mode_ == kModePreemptiveExpandSuccess ||
895 last_mode_ == kModePreemptiveExpandLowEnergy) {
896 // Subtract (samples_left + output_size_samples_) from sampleMemory.
897 decision_logic_->AddSampleMemory(-(samples_left + output_size_samples_));
898 }
899
900 // Check if it is time to play a DTMF event.
901 if (dtmf_buffer_->GetEvent(end_timestamp +
902 decision_logic_->generated_noise_samples(),
903 dtmf_event)) {
904 *play_dtmf = true;
905 }
906
907 // Get instruction.
908 assert(sync_buffer_.get());
909 assert(expand_.get());
910 *operation = decision_logic_->GetDecision(*sync_buffer_,
911 *expand_,
912 decoder_frame_length_,
913 header,
914 last_mode_,
915 *play_dtmf,
916 &reset_decoder_);
917
918 // Check if we already have enough samples in the |sync_buffer_|. If so,
919 // change decision to normal, unless the decision was merge, accelerate, or
920 // preemptive expand.
921 if (samples_left >= output_size_samples_ &&
922 *operation != kMerge &&
923 *operation != kAccelerate &&
924 *operation != kPreemptiveExpand) {
925 *operation = kNormal;
926 return 0;
927 }
928
929 decision_logic_->ExpandDecision(*operation);
930
931 // Check conditions for reset.
932 if (new_codec_ || *operation == kUndefined) {
933 // The only valid reason to get kUndefined is that new_codec_ is set.
934 assert(new_codec_);
935 if (*play_dtmf && !header) {
936 timestamp_ = dtmf_event->timestamp;
937 } else {
938 assert(header);
939 if (!header) {
940 LOG_F(LS_ERROR) << "Packet missing where it shouldn't.";
941 return -1;
942 }
943 timestamp_ = header->timestamp;
944 if (*operation == kRfc3389CngNoPacket
945 #ifndef LEGACY_BITEXACT
946 // Without this check, it can happen that a non-CNG packet is sent to
947 // the CNG decoder as if it was a SID frame. This is clearly a bug,
948 // but is kept for now to maintain bit-exactness with the test
949 // vectors.
950 && decoder_database_->IsComfortNoise(header->payloadType)
951 #endif
952 ) {
953 // Change decision to CNG packet, since we do have a CNG packet, but it
954 // was considered too early to use. Now, use it anyway.
955 *operation = kRfc3389Cng;
956 } else if (*operation != kRfc3389Cng) {
957 *operation = kNormal;
958 }
959 }
960 // Adjust |sync_buffer_| timestamp before setting |end_timestamp| to the
961 // new value.
962 sync_buffer_->IncreaseEndTimestamp(timestamp_ - end_timestamp);
963 end_timestamp = timestamp_;
964 new_codec_ = false;
965 decision_logic_->SoftReset();
966 buffer_level_filter_->Reset();
967 delay_manager_->Reset();
968 stats_.ResetMcu();
969 }
970
971 int required_samples = output_size_samples_;
972 const int samples_10_ms = 80 * fs_mult_;
973 const int samples_20_ms = 2 * samples_10_ms;
974 const int samples_30_ms = 3 * samples_10_ms;
975
976 switch (*operation) {
977 case kExpand: {
978 timestamp_ = end_timestamp;
979 return 0;
980 }
981 case kRfc3389CngNoPacket:
982 case kCodecInternalCng: {
983 return 0;
984 }
985 case kDtmf: {
986 // TODO(hlundin): Write test for this.
987 // Update timestamp.
988 timestamp_ = end_timestamp;
989 if (decision_logic_->generated_noise_samples() > 0 &&
990 last_mode_ != kModeDtmf) {
991 // Make a jump in timestamp due to the recently played comfort noise.
992 uint32_t timestamp_jump = decision_logic_->generated_noise_samples();
993 sync_buffer_->IncreaseEndTimestamp(timestamp_jump);
994 timestamp_ += timestamp_jump;
995 }
996 decision_logic_->set_generated_noise_samples(0);
997 return 0;
998 }
999 case kAccelerate: {
1000 // In order to do a accelerate we need at least 30 ms of audio data.
1001 if (samples_left >= samples_30_ms) {
1002 // Already have enough data, so we do not need to extract any more.
1003 decision_logic_->set_sample_memory(samples_left);
1004 decision_logic_->set_prev_time_scale(true);
1005 return 0;
1006 } else if (samples_left >= samples_10_ms &&
1007 decoder_frame_length_ >= samples_30_ms) {
1008 // Avoid decoding more data as it might overflow the playout buffer.
1009 *operation = kNormal;
1010 return 0;
1011 } else if (samples_left < samples_20_ms &&
1012 decoder_frame_length_ < samples_30_ms) {
1013 // Build up decoded data by decoding at least 20 ms of audio data. Do
1014 // not perform accelerate yet, but wait until we only need to do one
1015 // decoding.
1016 required_samples = 2 * output_size_samples_;
1017 *operation = kNormal;
1018 }
1019 // If none of the above is true, we have one of two possible situations:
1020 // (1) 20 ms <= samples_left < 30 ms and decoder_frame_length_ < 30 ms; or
1021 // (2) samples_left < 10 ms and decoder_frame_length_ >= 30 ms.
1022 // In either case, we move on with the accelerate decision, and decode one
1023 // frame now.
1024 break;
1025 }
1026 case kPreemptiveExpand: {
1027 // In order to do a preemptive expand we need at least 30 ms of decoded
1028 // audio data.
1029 if ((samples_left >= samples_30_ms) ||
1030 (samples_left >= samples_10_ms &&
1031 decoder_frame_length_ >= samples_30_ms)) {
1032 // Already have enough data, so we do not need to extract any more.
1033 // Or, avoid decoding more data as it might overflow the playout buffer.
1034 // Still try preemptive expand, though.
1035 decision_logic_->set_sample_memory(samples_left);
1036 decision_logic_->set_prev_time_scale(true);
1037 return 0;
1038 }
1039 if (samples_left < samples_20_ms &&
1040 decoder_frame_length_ < samples_30_ms) {
1041 // Build up decoded data by decoding at least 20 ms of audio data.
1042 // Still try to perform preemptive expand.
1043 required_samples = 2 * output_size_samples_;
1044 }
1045 // Move on with the preemptive expand decision.
1046 break;
1047 }
1048 case kMerge: {
1049 required_samples =
1050 std::max(merge_->RequiredFutureSamples(), required_samples);
1051 break;
1052 }
1053 default: {
1054 // Do nothing.
1055 }
1056 }
1057
1058 // Get packets from buffer.
1059 int extracted_samples = 0;
1060 if (header &&
1061 *operation != kAlternativePlc &&
1062 *operation != kAlternativePlcIncreaseTimestamp &&
1063 *operation != kAudioRepetition &&
1064 *operation != kAudioRepetitionIncreaseTimestamp) {
1065 sync_buffer_->IncreaseEndTimestamp(header->timestamp - end_timestamp);
1066 if (decision_logic_->CngOff()) {
1067 // Adjustment of timestamp only corresponds to an actual packet loss
1068 // if comfort noise is not played. If comfort noise was just played,
1069 // this adjustment of timestamp is only done to get back in sync with the
1070 // stream timestamp; no loss to report.
1071 stats_.LostSamples(header->timestamp - end_timestamp);
1072 }
1073
1074 if (*operation != kRfc3389Cng) {
1075 // We are about to decode and use a non-CNG packet.
1076 decision_logic_->SetCngOff();
1077 }
1078 // Reset CNG timestamp as a new packet will be delivered.
1079 // (Also if this is a CNG packet, since playedOutTS is updated.)
1080 decision_logic_->set_generated_noise_samples(0);
1081
1082 extracted_samples = ExtractPackets(required_samples, packet_list);
1083 if (extracted_samples < 0) {
1084 LOG_F(LS_WARNING) << "Failed to extract packets from buffer.";
1085 return kPacketBufferCorruption;
1086 }
1087 }
1088
1089 if (*operation == kAccelerate ||
1090 *operation == kPreemptiveExpand) {
1091 decision_logic_->set_sample_memory(samples_left + extracted_samples);
1092 decision_logic_->set_prev_time_scale(true);
1093 }
1094
1095 if (*operation == kAccelerate) {
1096 // Check that we have enough data (30ms) to do accelerate.
1097 if (extracted_samples + samples_left < samples_30_ms) {
1098 // TODO(hlundin): Write test for this.
1099 // Not enough, do normal operation instead.
1100 *operation = kNormal;
1101 }
1102 }
1103
1104 timestamp_ = end_timestamp;
1105 return 0;
1106 }
1107
Decode(PacketList * packet_list,Operations * operation,int * decoded_length,AudioDecoder::SpeechType * speech_type)1108 int NetEqImpl::Decode(PacketList* packet_list, Operations* operation,
1109 int* decoded_length,
1110 AudioDecoder::SpeechType* speech_type) {
1111 *speech_type = AudioDecoder::kSpeech;
1112 AudioDecoder* decoder = NULL;
1113 if (!packet_list->empty()) {
1114 const Packet* packet = packet_list->front();
1115 int payload_type = packet->header.payloadType;
1116 if (!decoder_database_->IsComfortNoise(payload_type)) {
1117 decoder = decoder_database_->GetDecoder(payload_type);
1118 assert(decoder);
1119 if (!decoder) {
1120 LOG_FERR1(LS_WARNING, GetDecoder, payload_type);
1121 PacketBuffer::DeleteAllPackets(packet_list);
1122 return kDecoderNotFound;
1123 }
1124 bool decoder_changed;
1125 decoder_database_->SetActiveDecoder(payload_type, &decoder_changed);
1126 if (decoder_changed) {
1127 // We have a new decoder. Re-init some values.
1128 const DecoderDatabase::DecoderInfo* decoder_info = decoder_database_
1129 ->GetDecoderInfo(payload_type);
1130 assert(decoder_info);
1131 if (!decoder_info) {
1132 LOG_FERR1(LS_WARNING, GetDecoderInfo, payload_type);
1133 PacketBuffer::DeleteAllPackets(packet_list);
1134 return kDecoderNotFound;
1135 }
1136 // If sampling rate or number of channels has changed, we need to make
1137 // a reset.
1138 if (decoder_info->fs_hz != fs_hz_ ||
1139 decoder->channels() != algorithm_buffer_->Channels()) {
1140 // TODO(tlegrand): Add unittest to cover this event.
1141 SetSampleRateAndChannels(decoder_info->fs_hz, decoder->channels());
1142 }
1143 sync_buffer_->set_end_timestamp(timestamp_);
1144 playout_timestamp_ = timestamp_;
1145 }
1146 }
1147 }
1148
1149 if (reset_decoder_) {
1150 // TODO(hlundin): Write test for this.
1151 // Reset decoder.
1152 if (decoder) {
1153 decoder->Init();
1154 }
1155 // Reset comfort noise decoder.
1156 AudioDecoder* cng_decoder = decoder_database_->GetActiveCngDecoder();
1157 if (cng_decoder) {
1158 cng_decoder->Init();
1159 }
1160 reset_decoder_ = false;
1161 }
1162
1163 #ifdef LEGACY_BITEXACT
1164 // Due to a bug in old SignalMCU, it could happen that CNG operation was
1165 // decided, but a speech packet was provided. The speech packet will be used
1166 // to update the comfort noise decoder, as if it was a SID frame, which is
1167 // clearly wrong.
1168 if (*operation == kRfc3389Cng) {
1169 return 0;
1170 }
1171 #endif
1172
1173 *decoded_length = 0;
1174 // Update codec-internal PLC state.
1175 if ((*operation == kMerge) && decoder && decoder->HasDecodePlc()) {
1176 decoder->DecodePlc(1, &decoded_buffer_[*decoded_length]);
1177 }
1178
1179 int return_value = DecodeLoop(packet_list, operation, decoder,
1180 decoded_length, speech_type);
1181
1182 if (*decoded_length < 0) {
1183 // Error returned from the decoder.
1184 *decoded_length = 0;
1185 sync_buffer_->IncreaseEndTimestamp(decoder_frame_length_);
1186 int error_code = 0;
1187 if (decoder)
1188 error_code = decoder->ErrorCode();
1189 if (error_code != 0) {
1190 // Got some error code from the decoder.
1191 decoder_error_code_ = error_code;
1192 return_value = kDecoderErrorCode;
1193 } else {
1194 // Decoder does not implement error codes. Return generic error.
1195 return_value = kOtherDecoderError;
1196 }
1197 LOG_FERR2(LS_WARNING, DecodeLoop, error_code, packet_list->size());
1198 *operation = kExpand; // Do expansion to get data instead.
1199 }
1200 if (*speech_type != AudioDecoder::kComfortNoise) {
1201 // Don't increment timestamp if codec returned CNG speech type
1202 // since in this case, the we will increment the CNGplayedTS counter.
1203 // Increase with number of samples per channel.
1204 assert(*decoded_length == 0 ||
1205 (decoder && decoder->channels() == sync_buffer_->Channels()));
1206 sync_buffer_->IncreaseEndTimestamp(
1207 *decoded_length / static_cast<int>(sync_buffer_->Channels()));
1208 }
1209 return return_value;
1210 }
1211
DecodeLoop(PacketList * packet_list,Operations * operation,AudioDecoder * decoder,int * decoded_length,AudioDecoder::SpeechType * speech_type)1212 int NetEqImpl::DecodeLoop(PacketList* packet_list, Operations* operation,
1213 AudioDecoder* decoder, int* decoded_length,
1214 AudioDecoder::SpeechType* speech_type) {
1215 Packet* packet = NULL;
1216 if (!packet_list->empty()) {
1217 packet = packet_list->front();
1218 }
1219 // Do decoding.
1220 while (packet &&
1221 !decoder_database_->IsComfortNoise(packet->header.payloadType)) {
1222 assert(decoder); // At this point, we must have a decoder object.
1223 // The number of channels in the |sync_buffer_| should be the same as the
1224 // number decoder channels.
1225 assert(sync_buffer_->Channels() == decoder->channels());
1226 assert(decoded_buffer_length_ >= kMaxFrameSize * decoder->channels());
1227 assert(*operation == kNormal || *operation == kAccelerate ||
1228 *operation == kMerge || *operation == kPreemptiveExpand);
1229 packet_list->pop_front();
1230 int payload_length = packet->payload_length;
1231 int16_t decode_length;
1232 if (packet->sync_packet) {
1233 // Decode to silence with the same frame size as the last decode.
1234 LOG(LS_VERBOSE) << "Decoding sync-packet: " <<
1235 " ts=" << packet->header.timestamp <<
1236 ", sn=" << packet->header.sequenceNumber <<
1237 ", pt=" << static_cast<int>(packet->header.payloadType) <<
1238 ", ssrc=" << packet->header.ssrc <<
1239 ", len=" << packet->payload_length;
1240 memset(&decoded_buffer_[*decoded_length], 0, decoder_frame_length_ *
1241 decoder->channels() * sizeof(decoded_buffer_[0]));
1242 decode_length = decoder_frame_length_;
1243 } else if (!packet->primary) {
1244 // This is a redundant payload; call the special decoder method.
1245 LOG(LS_VERBOSE) << "Decoding packet (redundant):" <<
1246 " ts=" << packet->header.timestamp <<
1247 ", sn=" << packet->header.sequenceNumber <<
1248 ", pt=" << static_cast<int>(packet->header.payloadType) <<
1249 ", ssrc=" << packet->header.ssrc <<
1250 ", len=" << packet->payload_length;
1251 decode_length = decoder->DecodeRedundant(
1252 packet->payload, packet->payload_length,
1253 &decoded_buffer_[*decoded_length], speech_type);
1254 } else {
1255 LOG(LS_VERBOSE) << "Decoding packet: ts=" << packet->header.timestamp <<
1256 ", sn=" << packet->header.sequenceNumber <<
1257 ", pt=" << static_cast<int>(packet->header.payloadType) <<
1258 ", ssrc=" << packet->header.ssrc <<
1259 ", len=" << packet->payload_length;
1260 decode_length = decoder->Decode(packet->payload,
1261 packet->payload_length,
1262 &decoded_buffer_[*decoded_length],
1263 speech_type);
1264 }
1265
1266 delete[] packet->payload;
1267 delete packet;
1268 packet = NULL;
1269 if (decode_length > 0) {
1270 *decoded_length += decode_length;
1271 // Update |decoder_frame_length_| with number of samples per channel.
1272 decoder_frame_length_ = decode_length /
1273 static_cast<int>(decoder->channels());
1274 LOG(LS_VERBOSE) << "Decoded " << decode_length << " samples (" <<
1275 decoder->channels() << " channel(s) -> " << decoder_frame_length_ <<
1276 " samples per channel)";
1277 } else if (decode_length < 0) {
1278 // Error.
1279 LOG_FERR2(LS_WARNING, Decode, decode_length, payload_length);
1280 *decoded_length = -1;
1281 PacketBuffer::DeleteAllPackets(packet_list);
1282 break;
1283 }
1284 if (*decoded_length > static_cast<int>(decoded_buffer_length_)) {
1285 // Guard against overflow.
1286 LOG_F(LS_WARNING) << "Decoded too much.";
1287 PacketBuffer::DeleteAllPackets(packet_list);
1288 return kDecodedTooMuch;
1289 }
1290 if (!packet_list->empty()) {
1291 packet = packet_list->front();
1292 } else {
1293 packet = NULL;
1294 }
1295 } // End of decode loop.
1296
1297 // If the list is not empty at this point, either a decoding error terminated
1298 // the while-loop, or list must hold exactly one CNG packet.
1299 assert(packet_list->empty() || *decoded_length < 0 ||
1300 (packet_list->size() == 1 && packet &&
1301 decoder_database_->IsComfortNoise(packet->header.payloadType)));
1302 return 0;
1303 }
1304
DoNormal(const int16_t * decoded_buffer,size_t decoded_length,AudioDecoder::SpeechType speech_type,bool play_dtmf)1305 void NetEqImpl::DoNormal(const int16_t* decoded_buffer, size_t decoded_length,
1306 AudioDecoder::SpeechType speech_type, bool play_dtmf) {
1307 assert(normal_.get());
1308 assert(mute_factor_array_.get());
1309 normal_->Process(decoded_buffer, decoded_length, last_mode_,
1310 mute_factor_array_.get(), algorithm_buffer_.get());
1311 if (decoded_length != 0) {
1312 last_mode_ = kModeNormal;
1313 }
1314
1315 // If last packet was decoded as an inband CNG, set mode to CNG instead.
1316 if ((speech_type == AudioDecoder::kComfortNoise)
1317 || ((last_mode_ == kModeCodecInternalCng)
1318 && (decoded_length == 0))) {
1319 // TODO(hlundin): Remove second part of || statement above.
1320 last_mode_ = kModeCodecInternalCng;
1321 }
1322
1323 if (!play_dtmf) {
1324 dtmf_tone_generator_->Reset();
1325 }
1326 }
1327
DoMerge(int16_t * decoded_buffer,size_t decoded_length,AudioDecoder::SpeechType speech_type,bool play_dtmf)1328 void NetEqImpl::DoMerge(int16_t* decoded_buffer, size_t decoded_length,
1329 AudioDecoder::SpeechType speech_type, bool play_dtmf) {
1330 assert(mute_factor_array_.get());
1331 assert(merge_.get());
1332 int new_length = merge_->Process(decoded_buffer, decoded_length,
1333 mute_factor_array_.get(),
1334 algorithm_buffer_.get());
1335
1336 // Update in-call and post-call statistics.
1337 if (expand_->MuteFactor(0) == 0) {
1338 // Expand generates only noise.
1339 stats_.ExpandedNoiseSamples(new_length - static_cast<int>(decoded_length));
1340 } else {
1341 // Expansion generates more than only noise.
1342 stats_.ExpandedVoiceSamples(new_length - static_cast<int>(decoded_length));
1343 }
1344
1345 last_mode_ = kModeMerge;
1346 // If last packet was decoded as an inband CNG, set mode to CNG instead.
1347 if (speech_type == AudioDecoder::kComfortNoise) {
1348 last_mode_ = kModeCodecInternalCng;
1349 }
1350 expand_->Reset();
1351 if (!play_dtmf) {
1352 dtmf_tone_generator_->Reset();
1353 }
1354 }
1355
DoExpand(bool play_dtmf)1356 int NetEqImpl::DoExpand(bool play_dtmf) {
1357 while ((sync_buffer_->FutureLength() - expand_->overlap_length()) <
1358 static_cast<size_t>(output_size_samples_)) {
1359 algorithm_buffer_->Clear();
1360 int return_value = expand_->Process(algorithm_buffer_.get());
1361 int length = static_cast<int>(algorithm_buffer_->Size());
1362
1363 // Update in-call and post-call statistics.
1364 if (expand_->MuteFactor(0) == 0) {
1365 // Expand operation generates only noise.
1366 stats_.ExpandedNoiseSamples(length);
1367 } else {
1368 // Expand operation generates more than only noise.
1369 stats_.ExpandedVoiceSamples(length);
1370 }
1371
1372 last_mode_ = kModeExpand;
1373
1374 if (return_value < 0) {
1375 return return_value;
1376 }
1377
1378 sync_buffer_->PushBack(*algorithm_buffer_);
1379 algorithm_buffer_->Clear();
1380 }
1381 if (!play_dtmf) {
1382 dtmf_tone_generator_->Reset();
1383 }
1384 return 0;
1385 }
1386
DoAccelerate(int16_t * decoded_buffer,size_t decoded_length,AudioDecoder::SpeechType speech_type,bool play_dtmf)1387 int NetEqImpl::DoAccelerate(int16_t* decoded_buffer, size_t decoded_length,
1388 AudioDecoder::SpeechType speech_type,
1389 bool play_dtmf) {
1390 const size_t required_samples = 240 * fs_mult_; // Must have 30 ms.
1391 size_t borrowed_samples_per_channel = 0;
1392 size_t num_channels = algorithm_buffer_->Channels();
1393 size_t decoded_length_per_channel = decoded_length / num_channels;
1394 if (decoded_length_per_channel < required_samples) {
1395 // Must move data from the |sync_buffer_| in order to get 30 ms.
1396 borrowed_samples_per_channel = static_cast<int>(required_samples -
1397 decoded_length_per_channel);
1398 memmove(&decoded_buffer[borrowed_samples_per_channel * num_channels],
1399 decoded_buffer,
1400 sizeof(int16_t) * decoded_length);
1401 sync_buffer_->ReadInterleavedFromEnd(borrowed_samples_per_channel,
1402 decoded_buffer);
1403 decoded_length = required_samples * num_channels;
1404 }
1405
1406 int16_t samples_removed;
1407 Accelerate::ReturnCodes return_code = accelerate_->Process(
1408 decoded_buffer, decoded_length, algorithm_buffer_.get(),
1409 &samples_removed);
1410 stats_.AcceleratedSamples(samples_removed);
1411 switch (return_code) {
1412 case Accelerate::kSuccess:
1413 last_mode_ = kModeAccelerateSuccess;
1414 break;
1415 case Accelerate::kSuccessLowEnergy:
1416 last_mode_ = kModeAccelerateLowEnergy;
1417 break;
1418 case Accelerate::kNoStretch:
1419 last_mode_ = kModeAccelerateFail;
1420 break;
1421 case Accelerate::kError:
1422 // TODO(hlundin): Map to kModeError instead?
1423 last_mode_ = kModeAccelerateFail;
1424 return kAccelerateError;
1425 }
1426
1427 if (borrowed_samples_per_channel > 0) {
1428 // Copy borrowed samples back to the |sync_buffer_|.
1429 size_t length = algorithm_buffer_->Size();
1430 if (length < borrowed_samples_per_channel) {
1431 // This destroys the beginning of the buffer, but will not cause any
1432 // problems.
1433 sync_buffer_->ReplaceAtIndex(*algorithm_buffer_,
1434 sync_buffer_->Size() -
1435 borrowed_samples_per_channel);
1436 sync_buffer_->PushFrontZeros(borrowed_samples_per_channel - length);
1437 algorithm_buffer_->PopFront(length);
1438 assert(algorithm_buffer_->Empty());
1439 } else {
1440 sync_buffer_->ReplaceAtIndex(*algorithm_buffer_,
1441 borrowed_samples_per_channel,
1442 sync_buffer_->Size() -
1443 borrowed_samples_per_channel);
1444 algorithm_buffer_->PopFront(borrowed_samples_per_channel);
1445 }
1446 }
1447
1448 // If last packet was decoded as an inband CNG, set mode to CNG instead.
1449 if (speech_type == AudioDecoder::kComfortNoise) {
1450 last_mode_ = kModeCodecInternalCng;
1451 }
1452 if (!play_dtmf) {
1453 dtmf_tone_generator_->Reset();
1454 }
1455 expand_->Reset();
1456 return 0;
1457 }
1458
DoPreemptiveExpand(int16_t * decoded_buffer,size_t decoded_length,AudioDecoder::SpeechType speech_type,bool play_dtmf)1459 int NetEqImpl::DoPreemptiveExpand(int16_t* decoded_buffer,
1460 size_t decoded_length,
1461 AudioDecoder::SpeechType speech_type,
1462 bool play_dtmf) {
1463 const size_t required_samples = 240 * fs_mult_; // Must have 30 ms.
1464 size_t num_channels = algorithm_buffer_->Channels();
1465 int borrowed_samples_per_channel = 0;
1466 int old_borrowed_samples_per_channel = 0;
1467 size_t decoded_length_per_channel = decoded_length / num_channels;
1468 if (decoded_length_per_channel < required_samples) {
1469 // Must move data from the |sync_buffer_| in order to get 30 ms.
1470 borrowed_samples_per_channel = static_cast<int>(required_samples -
1471 decoded_length_per_channel);
1472 // Calculate how many of these were already played out.
1473 old_borrowed_samples_per_channel = static_cast<int>(
1474 borrowed_samples_per_channel - sync_buffer_->FutureLength());
1475 old_borrowed_samples_per_channel = std::max(
1476 0, old_borrowed_samples_per_channel);
1477 memmove(&decoded_buffer[borrowed_samples_per_channel * num_channels],
1478 decoded_buffer,
1479 sizeof(int16_t) * decoded_length);
1480 sync_buffer_->ReadInterleavedFromEnd(borrowed_samples_per_channel,
1481 decoded_buffer);
1482 decoded_length = required_samples * num_channels;
1483 }
1484
1485 int16_t samples_added;
1486 PreemptiveExpand::ReturnCodes return_code = preemptive_expand_->Process(
1487 decoded_buffer, static_cast<int>(decoded_length),
1488 old_borrowed_samples_per_channel,
1489 algorithm_buffer_.get(), &samples_added);
1490 stats_.PreemptiveExpandedSamples(samples_added);
1491 switch (return_code) {
1492 case PreemptiveExpand::kSuccess:
1493 last_mode_ = kModePreemptiveExpandSuccess;
1494 break;
1495 case PreemptiveExpand::kSuccessLowEnergy:
1496 last_mode_ = kModePreemptiveExpandLowEnergy;
1497 break;
1498 case PreemptiveExpand::kNoStretch:
1499 last_mode_ = kModePreemptiveExpandFail;
1500 break;
1501 case PreemptiveExpand::kError:
1502 // TODO(hlundin): Map to kModeError instead?
1503 last_mode_ = kModePreemptiveExpandFail;
1504 return kPreemptiveExpandError;
1505 }
1506
1507 if (borrowed_samples_per_channel > 0) {
1508 // Copy borrowed samples back to the |sync_buffer_|.
1509 sync_buffer_->ReplaceAtIndex(
1510 *algorithm_buffer_, borrowed_samples_per_channel,
1511 sync_buffer_->Size() - borrowed_samples_per_channel);
1512 algorithm_buffer_->PopFront(borrowed_samples_per_channel);
1513 }
1514
1515 // If last packet was decoded as an inband CNG, set mode to CNG instead.
1516 if (speech_type == AudioDecoder::kComfortNoise) {
1517 last_mode_ = kModeCodecInternalCng;
1518 }
1519 if (!play_dtmf) {
1520 dtmf_tone_generator_->Reset();
1521 }
1522 expand_->Reset();
1523 return 0;
1524 }
1525
DoRfc3389Cng(PacketList * packet_list,bool play_dtmf)1526 int NetEqImpl::DoRfc3389Cng(PacketList* packet_list, bool play_dtmf) {
1527 if (!packet_list->empty()) {
1528 // Must have exactly one SID frame at this point.
1529 assert(packet_list->size() == 1);
1530 Packet* packet = packet_list->front();
1531 packet_list->pop_front();
1532 if (!decoder_database_->IsComfortNoise(packet->header.payloadType)) {
1533 #ifdef LEGACY_BITEXACT
1534 // This can happen due to a bug in GetDecision. Change the payload type
1535 // to a CNG type, and move on. Note that this means that we are in fact
1536 // sending a non-CNG payload to the comfort noise decoder for decoding.
1537 // Clearly wrong, but will maintain bit-exactness with legacy.
1538 if (fs_hz_ == 8000) {
1539 packet->header.payloadType =
1540 decoder_database_->GetRtpPayloadType(kDecoderCNGnb);
1541 } else if (fs_hz_ == 16000) {
1542 packet->header.payloadType =
1543 decoder_database_->GetRtpPayloadType(kDecoderCNGwb);
1544 } else if (fs_hz_ == 32000) {
1545 packet->header.payloadType =
1546 decoder_database_->GetRtpPayloadType(kDecoderCNGswb32kHz);
1547 } else if (fs_hz_ == 48000) {
1548 packet->header.payloadType =
1549 decoder_database_->GetRtpPayloadType(kDecoderCNGswb48kHz);
1550 }
1551 assert(decoder_database_->IsComfortNoise(packet->header.payloadType));
1552 #else
1553 LOG(LS_ERROR) << "Trying to decode non-CNG payload as CNG.";
1554 return kOtherError;
1555 #endif
1556 }
1557 // UpdateParameters() deletes |packet|.
1558 if (comfort_noise_->UpdateParameters(packet) ==
1559 ComfortNoise::kInternalError) {
1560 LOG_FERR0(LS_WARNING, UpdateParameters);
1561 algorithm_buffer_->Zeros(output_size_samples_);
1562 return -comfort_noise_->internal_error_code();
1563 }
1564 }
1565 int cn_return = comfort_noise_->Generate(output_size_samples_,
1566 algorithm_buffer_.get());
1567 expand_->Reset();
1568 last_mode_ = kModeRfc3389Cng;
1569 if (!play_dtmf) {
1570 dtmf_tone_generator_->Reset();
1571 }
1572 if (cn_return == ComfortNoise::kInternalError) {
1573 LOG_FERR1(LS_WARNING, comfort_noise_->Generate, cn_return);
1574 decoder_error_code_ = comfort_noise_->internal_error_code();
1575 return kComfortNoiseErrorCode;
1576 } else if (cn_return == ComfortNoise::kUnknownPayloadType) {
1577 LOG_FERR1(LS_WARNING, comfort_noise_->Generate, cn_return);
1578 return kUnknownRtpPayloadType;
1579 }
1580 return 0;
1581 }
1582
DoCodecInternalCng()1583 void NetEqImpl::DoCodecInternalCng() {
1584 int length = 0;
1585 // TODO(hlundin): Will probably need a longer buffer for multi-channel.
1586 int16_t decoded_buffer[kMaxFrameSize];
1587 AudioDecoder* decoder = decoder_database_->GetActiveDecoder();
1588 if (decoder) {
1589 const uint8_t* dummy_payload = NULL;
1590 AudioDecoder::SpeechType speech_type;
1591 length = decoder->Decode(dummy_payload, 0, decoded_buffer, &speech_type);
1592 }
1593 assert(mute_factor_array_.get());
1594 normal_->Process(decoded_buffer, length, last_mode_, mute_factor_array_.get(),
1595 algorithm_buffer_.get());
1596 last_mode_ = kModeCodecInternalCng;
1597 expand_->Reset();
1598 }
1599
DoDtmf(const DtmfEvent & dtmf_event,bool * play_dtmf)1600 int NetEqImpl::DoDtmf(const DtmfEvent& dtmf_event, bool* play_dtmf) {
1601 // This block of the code and the block further down, handling |dtmf_switch|
1602 // are commented out. Otherwise playing out-of-band DTMF would fail in VoE
1603 // test, DtmfTest.ManualSuccessfullySendsOutOfBandTelephoneEvents. This is
1604 // equivalent to |dtmf_switch| always be false.
1605 //
1606 // See http://webrtc-codereview.appspot.com/1195004/ for discussion
1607 // On this issue. This change might cause some glitches at the point of
1608 // switch from audio to DTMF. Issue 1545 is filed to track this.
1609 //
1610 // bool dtmf_switch = false;
1611 // if ((last_mode_ != kModeDtmf) && dtmf_tone_generator_->initialized()) {
1612 // // Special case; see below.
1613 // // We must catch this before calling Generate, since |initialized| is
1614 // // modified in that call.
1615 // dtmf_switch = true;
1616 // }
1617
1618 int dtmf_return_value = 0;
1619 if (!dtmf_tone_generator_->initialized()) {
1620 // Initialize if not already done.
1621 dtmf_return_value = dtmf_tone_generator_->Init(fs_hz_, dtmf_event.event_no,
1622 dtmf_event.volume);
1623 }
1624
1625 if (dtmf_return_value == 0) {
1626 // Generate DTMF signal.
1627 dtmf_return_value = dtmf_tone_generator_->Generate(output_size_samples_,
1628 algorithm_buffer_.get());
1629 }
1630
1631 if (dtmf_return_value < 0) {
1632 algorithm_buffer_->Zeros(output_size_samples_);
1633 return dtmf_return_value;
1634 }
1635
1636 // if (dtmf_switch) {
1637 // // This is the special case where the previous operation was DTMF
1638 // // overdub, but the current instruction is "regular" DTMF. We must make
1639 // // sure that the DTMF does not have any discontinuities. The first DTMF
1640 // // sample that we generate now must be played out immediately, therefore
1641 // // it must be copied to the speech buffer.
1642 // // TODO(hlundin): This code seems incorrect. (Legacy.) Write test and
1643 // // verify correct operation.
1644 // assert(false);
1645 // // Must generate enough data to replace all of the |sync_buffer_|
1646 // // "future".
1647 // int required_length = sync_buffer_->FutureLength();
1648 // assert(dtmf_tone_generator_->initialized());
1649 // dtmf_return_value = dtmf_tone_generator_->Generate(required_length,
1650 // algorithm_buffer_);
1651 // assert((size_t) required_length == algorithm_buffer_->Size());
1652 // if (dtmf_return_value < 0) {
1653 // algorithm_buffer_->Zeros(output_size_samples_);
1654 // return dtmf_return_value;
1655 // }
1656 //
1657 // // Overwrite the "future" part of the speech buffer with the new DTMF
1658 // // data.
1659 // // TODO(hlundin): It seems that this overwriting has gone lost.
1660 // // Not adapted for multi-channel yet.
1661 // assert(algorithm_buffer_->Channels() == 1);
1662 // if (algorithm_buffer_->Channels() != 1) {
1663 // LOG(LS_WARNING) << "DTMF not supported for more than one channel";
1664 // return kStereoNotSupported;
1665 // }
1666 // // Shuffle the remaining data to the beginning of algorithm buffer.
1667 // algorithm_buffer_->PopFront(sync_buffer_->FutureLength());
1668 // }
1669
1670 sync_buffer_->IncreaseEndTimestamp(output_size_samples_);
1671 expand_->Reset();
1672 last_mode_ = kModeDtmf;
1673
1674 // Set to false because the DTMF is already in the algorithm buffer.
1675 *play_dtmf = false;
1676 return 0;
1677 }
1678
DoAlternativePlc(bool increase_timestamp)1679 void NetEqImpl::DoAlternativePlc(bool increase_timestamp) {
1680 AudioDecoder* decoder = decoder_database_->GetActiveDecoder();
1681 int length;
1682 if (decoder && decoder->HasDecodePlc()) {
1683 // Use the decoder's packet-loss concealment.
1684 // TODO(hlundin): Will probably need a longer buffer for multi-channel.
1685 int16_t decoded_buffer[kMaxFrameSize];
1686 length = decoder->DecodePlc(1, decoded_buffer);
1687 if (length > 0) {
1688 algorithm_buffer_->PushBackInterleaved(decoded_buffer, length);
1689 } else {
1690 length = 0;
1691 }
1692 } else {
1693 // Do simple zero-stuffing.
1694 length = output_size_samples_;
1695 algorithm_buffer_->Zeros(length);
1696 // By not advancing the timestamp, NetEq inserts samples.
1697 stats_.AddZeros(length);
1698 }
1699 if (increase_timestamp) {
1700 sync_buffer_->IncreaseEndTimestamp(length);
1701 }
1702 expand_->Reset();
1703 }
1704
DtmfOverdub(const DtmfEvent & dtmf_event,size_t num_channels,int16_t * output) const1705 int NetEqImpl::DtmfOverdub(const DtmfEvent& dtmf_event, size_t num_channels,
1706 int16_t* output) const {
1707 size_t out_index = 0;
1708 int overdub_length = output_size_samples_; // Default value.
1709
1710 if (sync_buffer_->dtmf_index() > sync_buffer_->next_index()) {
1711 // Special operation for transition from "DTMF only" to "DTMF overdub".
1712 out_index = std::min(
1713 sync_buffer_->dtmf_index() - sync_buffer_->next_index(),
1714 static_cast<size_t>(output_size_samples_));
1715 overdub_length = output_size_samples_ - static_cast<int>(out_index);
1716 }
1717
1718 AudioMultiVector dtmf_output(num_channels);
1719 int dtmf_return_value = 0;
1720 if (!dtmf_tone_generator_->initialized()) {
1721 dtmf_return_value = dtmf_tone_generator_->Init(fs_hz_, dtmf_event.event_no,
1722 dtmf_event.volume);
1723 }
1724 if (dtmf_return_value == 0) {
1725 dtmf_return_value = dtmf_tone_generator_->Generate(overdub_length,
1726 &dtmf_output);
1727 assert((size_t) overdub_length == dtmf_output.Size());
1728 }
1729 dtmf_output.ReadInterleaved(overdub_length, &output[out_index]);
1730 return dtmf_return_value < 0 ? dtmf_return_value : 0;
1731 }
1732
ExtractPackets(int required_samples,PacketList * packet_list)1733 int NetEqImpl::ExtractPackets(int required_samples, PacketList* packet_list) {
1734 bool first_packet = true;
1735 uint8_t prev_payload_type = 0;
1736 uint32_t prev_timestamp = 0;
1737 uint16_t prev_sequence_number = 0;
1738 bool next_packet_available = false;
1739
1740 const RTPHeader* header = packet_buffer_->NextRtpHeader();
1741 assert(header);
1742 if (!header) {
1743 return -1;
1744 }
1745 uint32_t first_timestamp = header->timestamp;
1746 int extracted_samples = 0;
1747
1748 // Packet extraction loop.
1749 do {
1750 timestamp_ = header->timestamp;
1751 int discard_count = 0;
1752 Packet* packet = packet_buffer_->GetNextPacket(&discard_count);
1753 // |header| may be invalid after the |packet_buffer_| operation.
1754 header = NULL;
1755 if (!packet) {
1756 LOG_FERR1(LS_ERROR, GetNextPacket, discard_count) <<
1757 "Should always be able to extract a packet here";
1758 assert(false); // Should always be able to extract a packet here.
1759 return -1;
1760 }
1761 stats_.PacketsDiscarded(discard_count);
1762 // Store waiting time in ms; packets->waiting_time is in "output blocks".
1763 stats_.StoreWaitingTime(packet->waiting_time * kOutputSizeMs);
1764 assert(packet->payload_length > 0);
1765 packet_list->push_back(packet); // Store packet in list.
1766
1767 if (first_packet) {
1768 first_packet = false;
1769 decoded_packet_sequence_number_ = prev_sequence_number =
1770 packet->header.sequenceNumber;
1771 decoded_packet_timestamp_ = prev_timestamp = packet->header.timestamp;
1772 prev_payload_type = packet->header.payloadType;
1773 }
1774
1775 // Store number of extracted samples.
1776 int packet_duration = 0;
1777 AudioDecoder* decoder = decoder_database_->GetDecoder(
1778 packet->header.payloadType);
1779 if (decoder) {
1780 if (packet->sync_packet) {
1781 packet_duration = decoder_frame_length_;
1782 } else {
1783 packet_duration = packet->primary ?
1784 decoder->PacketDuration(packet->payload, packet->payload_length) :
1785 decoder->PacketDurationRedundant(packet->payload,
1786 packet->payload_length);
1787 }
1788 } else {
1789 LOG_FERR1(LS_WARNING, GetDecoder, packet->header.payloadType) <<
1790 "Could not find a decoder for a packet about to be extracted.";
1791 assert(false);
1792 }
1793 if (packet_duration <= 0) {
1794 // Decoder did not return a packet duration. Assume that the packet
1795 // contains the same number of samples as the previous one.
1796 packet_duration = decoder_frame_length_;
1797 }
1798 extracted_samples = packet->header.timestamp - first_timestamp +
1799 packet_duration;
1800
1801 // Check what packet is available next.
1802 header = packet_buffer_->NextRtpHeader();
1803 next_packet_available = false;
1804 if (header && prev_payload_type == header->payloadType) {
1805 int16_t seq_no_diff = header->sequenceNumber - prev_sequence_number;
1806 int32_t ts_diff = header->timestamp - prev_timestamp;
1807 if (seq_no_diff == 1 ||
1808 (seq_no_diff == 0 && ts_diff == decoder_frame_length_)) {
1809 // The next sequence number is available, or the next part of a packet
1810 // that was split into pieces upon insertion.
1811 next_packet_available = true;
1812 }
1813 prev_sequence_number = header->sequenceNumber;
1814 }
1815 } while (extracted_samples < required_samples && next_packet_available);
1816
1817 if (extracted_samples > 0) {
1818 // Delete old packets only when we are going to decode something. Otherwise,
1819 // we could end up in the situation where we never decode anything, since
1820 // all incoming packets are considered too old but the buffer will also
1821 // never be flooded and flushed.
1822 packet_buffer_->DiscardOldPackets(timestamp_);
1823 }
1824
1825 return extracted_samples;
1826 }
1827
UpdatePlcComponents(int fs_hz,size_t channels)1828 void NetEqImpl::UpdatePlcComponents(int fs_hz, size_t channels) {
1829 // Delete objects and create new ones.
1830 expand_.reset(expand_factory_->Create(background_noise_.get(),
1831 sync_buffer_.get(), &random_vector_,
1832 fs_hz, channels));
1833 merge_.reset(new Merge(fs_hz, channels, expand_.get(), sync_buffer_.get()));
1834 }
1835
SetSampleRateAndChannels(int fs_hz,size_t channels)1836 void NetEqImpl::SetSampleRateAndChannels(int fs_hz, size_t channels) {
1837 LOG_API2(fs_hz, channels);
1838 // TODO(hlundin): Change to an enumerator and skip assert.
1839 assert(fs_hz == 8000 || fs_hz == 16000 || fs_hz == 32000 || fs_hz == 48000);
1840 assert(channels > 0);
1841
1842 fs_hz_ = fs_hz;
1843 fs_mult_ = fs_hz / 8000;
1844 output_size_samples_ = kOutputSizeMs * 8 * fs_mult_;
1845 decoder_frame_length_ = 3 * output_size_samples_; // Initialize to 30ms.
1846
1847 last_mode_ = kModeNormal;
1848
1849 // Create a new array of mute factors and set all to 1.
1850 mute_factor_array_.reset(new int16_t[channels]);
1851 for (size_t i = 0; i < channels; ++i) {
1852 mute_factor_array_[i] = 16384; // 1.0 in Q14.
1853 }
1854
1855 // Reset comfort noise decoder, if there is one active.
1856 AudioDecoder* cng_decoder = decoder_database_->GetActiveCngDecoder();
1857 if (cng_decoder) {
1858 cng_decoder->Init();
1859 }
1860
1861 // Reinit post-decode VAD with new sample rate.
1862 assert(vad_.get()); // Cannot be NULL here.
1863 vad_->Init();
1864
1865 // Delete algorithm buffer and create a new one.
1866 algorithm_buffer_.reset(new AudioMultiVector(channels));
1867
1868 // Delete sync buffer and create a new one.
1869 sync_buffer_.reset(new SyncBuffer(channels, kSyncBufferSize * fs_mult_));
1870
1871 // Delete BackgroundNoise object and create a new one.
1872 background_noise_.reset(new BackgroundNoise(channels));
1873 background_noise_->set_mode(background_noise_mode_);
1874
1875 // Reset random vector.
1876 random_vector_.Reset();
1877
1878 UpdatePlcComponents(fs_hz, channels);
1879
1880 // Move index so that we create a small set of future samples (all 0).
1881 sync_buffer_->set_next_index(sync_buffer_->next_index() -
1882 expand_->overlap_length());
1883
1884 normal_.reset(new Normal(fs_hz, decoder_database_.get(), *background_noise_,
1885 expand_.get()));
1886 accelerate_.reset(
1887 accelerate_factory_->Create(fs_hz, channels, *background_noise_));
1888 preemptive_expand_.reset(preemptive_expand_factory_->Create(
1889 fs_hz, channels,
1890 *background_noise_,
1891 static_cast<int>(expand_->overlap_length())));
1892
1893 // Delete ComfortNoise object and create a new one.
1894 comfort_noise_.reset(new ComfortNoise(fs_hz, decoder_database_.get(),
1895 sync_buffer_.get()));
1896
1897 // Verify that |decoded_buffer_| is long enough.
1898 if (decoded_buffer_length_ < kMaxFrameSize * channels) {
1899 // Reallocate to larger size.
1900 decoded_buffer_length_ = kMaxFrameSize * channels;
1901 decoded_buffer_.reset(new int16_t[decoded_buffer_length_]);
1902 }
1903
1904 // Create DecisionLogic if it is not created yet, then communicate new sample
1905 // rate and output size to DecisionLogic object.
1906 if (!decision_logic_.get()) {
1907 CreateDecisionLogic(kPlayoutOn);
1908 }
1909 decision_logic_->SetSampleRate(fs_hz_, output_size_samples_);
1910 }
1911
LastOutputType()1912 NetEqOutputType NetEqImpl::LastOutputType() {
1913 assert(vad_.get());
1914 assert(expand_.get());
1915 if (last_mode_ == kModeCodecInternalCng || last_mode_ == kModeRfc3389Cng) {
1916 return kOutputCNG;
1917 } else if (last_mode_ == kModeExpand && expand_->MuteFactor(0) == 0) {
1918 // Expand mode has faded down to background noise only (very long expand).
1919 return kOutputPLCtoCNG;
1920 } else if (last_mode_ == kModeExpand) {
1921 return kOutputPLC;
1922 } else if (vad_->running() && !vad_->active_speech()) {
1923 return kOutputVADPassive;
1924 } else {
1925 return kOutputNormal;
1926 }
1927 }
1928
CreateDecisionLogic(NetEqPlayoutMode mode)1929 void NetEqImpl::CreateDecisionLogic(NetEqPlayoutMode mode) {
1930 decision_logic_.reset(DecisionLogic::Create(fs_hz_, output_size_samples_,
1931 mode,
1932 decoder_database_.get(),
1933 *packet_buffer_.get(),
1934 delay_manager_.get(),
1935 buffer_level_filter_.get()));
1936 }
1937 } // namespace webrtc
1938