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(int fs,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(int fs,
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 decoded_packet_sequence_number_(-1),
94 decoded_packet_timestamp_(0) {
95 if (fs != 8000 && fs != 16000 && fs != 32000 && fs != 48000) {
96 LOG(LS_ERROR) << "Sample rate " << fs << " Hz not supported. " <<
97 "Changing to 8000 Hz.";
98 fs = 8000;
99 }
100 LOG(LS_VERBOSE) << "Create NetEqImpl object with fs = " << fs << ".";
101 fs_hz_ = fs;
102 fs_mult_ = fs / 8000;
103 output_size_samples_ = kOutputSizeMs * 8 * fs_mult_;
104 decoder_frame_length_ = 3 * output_size_samples_;
105 WebRtcSpl_Init();
106 if (create_components) {
107 SetSampleRateAndChannels(fs, 1); // Default is 1 channel.
108 }
109 }
110
~NetEqImpl()111 NetEqImpl::~NetEqImpl() {
112 LOG(LS_INFO) << "Deleting NetEqImpl object.";
113 }
114
InsertPacket(const WebRtcRTPHeader & rtp_header,const uint8_t * payload,int length_bytes,uint32_t receive_timestamp)115 int NetEqImpl::InsertPacket(const WebRtcRTPHeader& rtp_header,
116 const uint8_t* payload,
117 int length_bytes,
118 uint32_t receive_timestamp) {
119 CriticalSectionScoped lock(crit_sect_.get());
120 LOG(LS_VERBOSE) << "InsertPacket: ts=" << rtp_header.header.timestamp <<
121 ", sn=" << rtp_header.header.sequenceNumber <<
122 ", pt=" << static_cast<int>(rtp_header.header.payloadType) <<
123 ", ssrc=" << rtp_header.header.ssrc <<
124 ", len=" << length_bytes;
125 int error = InsertPacketInternal(rtp_header, payload, length_bytes,
126 receive_timestamp, false);
127 if (error != 0) {
128 LOG_FERR1(LS_WARNING, InsertPacketInternal, error);
129 error_code_ = error;
130 return kFail;
131 }
132 return kOK;
133 }
134
InsertSyncPacket(const WebRtcRTPHeader & rtp_header,uint32_t receive_timestamp)135 int NetEqImpl::InsertSyncPacket(const WebRtcRTPHeader& rtp_header,
136 uint32_t receive_timestamp) {
137 CriticalSectionScoped lock(crit_sect_.get());
138 LOG(LS_VERBOSE) << "InsertPacket-Sync: ts="
139 << rtp_header.header.timestamp <<
140 ", sn=" << rtp_header.header.sequenceNumber <<
141 ", pt=" << static_cast<int>(rtp_header.header.payloadType) <<
142 ", ssrc=" << rtp_header.header.ssrc;
143
144 const uint8_t kSyncPayload[] = { 's', 'y', 'n', 'c' };
145 int error = InsertPacketInternal(
146 rtp_header, kSyncPayload, sizeof(kSyncPayload), receive_timestamp, true);
147
148 if (error != 0) {
149 LOG_FERR1(LS_WARNING, InsertPacketInternal, error);
150 error_code_ = error;
151 return kFail;
152 }
153 return kOK;
154 }
155
GetAudio(size_t max_length,int16_t * output_audio,int * samples_per_channel,int * num_channels,NetEqOutputType * type)156 int NetEqImpl::GetAudio(size_t max_length, int16_t* output_audio,
157 int* samples_per_channel, int* num_channels,
158 NetEqOutputType* type) {
159 CriticalSectionScoped lock(crit_sect_.get());
160 LOG(LS_VERBOSE) << "GetAudio";
161 int error = GetAudioInternal(max_length, output_audio, samples_per_channel,
162 num_channels);
163 LOG(LS_VERBOSE) << "Produced " << *samples_per_channel <<
164 " samples/channel for " << *num_channels << " channel(s)";
165 if (error != 0) {
166 LOG_FERR1(LS_WARNING, GetAudioInternal, error);
167 error_code_ = error;
168 return kFail;
169 }
170 if (type) {
171 *type = LastOutputType();
172 }
173 return kOK;
174 }
175
RegisterPayloadType(enum NetEqDecoder codec,uint8_t rtp_payload_type)176 int NetEqImpl::RegisterPayloadType(enum NetEqDecoder codec,
177 uint8_t rtp_payload_type) {
178 CriticalSectionScoped lock(crit_sect_.get());
179 LOG_API2(static_cast<int>(rtp_payload_type), codec);
180 int ret = decoder_database_->RegisterPayload(rtp_payload_type, codec);
181 if (ret != DecoderDatabase::kOK) {
182 LOG_FERR2(LS_WARNING, RegisterPayload, rtp_payload_type, codec);
183 switch (ret) {
184 case DecoderDatabase::kInvalidRtpPayloadType:
185 error_code_ = kInvalidRtpPayloadType;
186 break;
187 case DecoderDatabase::kCodecNotSupported:
188 error_code_ = kCodecNotSupported;
189 break;
190 case DecoderDatabase::kDecoderExists:
191 error_code_ = kDecoderExists;
192 break;
193 default:
194 error_code_ = kOtherError;
195 }
196 return kFail;
197 }
198 return kOK;
199 }
200
RegisterExternalDecoder(AudioDecoder * decoder,enum NetEqDecoder codec,uint8_t rtp_payload_type)201 int NetEqImpl::RegisterExternalDecoder(AudioDecoder* decoder,
202 enum NetEqDecoder codec,
203 uint8_t rtp_payload_type) {
204 CriticalSectionScoped lock(crit_sect_.get());
205 LOG_API2(static_cast<int>(rtp_payload_type), codec);
206 if (!decoder) {
207 LOG(LS_ERROR) << "Cannot register external decoder with NULL pointer";
208 assert(false);
209 return kFail;
210 }
211 const int sample_rate_hz = AudioDecoder::CodecSampleRateHz(codec);
212 int ret = decoder_database_->InsertExternal(rtp_payload_type, codec,
213 sample_rate_hz, decoder);
214 if (ret != DecoderDatabase::kOK) {
215 LOG_FERR2(LS_WARNING, InsertExternal, rtp_payload_type, codec);
216 switch (ret) {
217 case DecoderDatabase::kInvalidRtpPayloadType:
218 error_code_ = kInvalidRtpPayloadType;
219 break;
220 case DecoderDatabase::kCodecNotSupported:
221 error_code_ = kCodecNotSupported;
222 break;
223 case DecoderDatabase::kDecoderExists:
224 error_code_ = kDecoderExists;
225 break;
226 case DecoderDatabase::kInvalidSampleRate:
227 error_code_ = kInvalidSampleRate;
228 break;
229 case DecoderDatabase::kInvalidPointer:
230 error_code_ = kInvalidPointer;
231 break;
232 default:
233 error_code_ = kOtherError;
234 }
235 return kFail;
236 }
237 return kOK;
238 }
239
RemovePayloadType(uint8_t rtp_payload_type)240 int NetEqImpl::RemovePayloadType(uint8_t rtp_payload_type) {
241 CriticalSectionScoped lock(crit_sect_.get());
242 LOG_API1(static_cast<int>(rtp_payload_type));
243 int ret = decoder_database_->Remove(rtp_payload_type);
244 if (ret == DecoderDatabase::kOK) {
245 return kOK;
246 } else if (ret == DecoderDatabase::kDecoderNotFound) {
247 error_code_ = kDecoderNotFound;
248 } else {
249 error_code_ = kOtherError;
250 }
251 LOG_FERR1(LS_WARNING, Remove, rtp_payload_type);
252 return kFail;
253 }
254
SetMinimumDelay(int delay_ms)255 bool NetEqImpl::SetMinimumDelay(int delay_ms) {
256 CriticalSectionScoped lock(crit_sect_.get());
257 if (delay_ms >= 0 && delay_ms < 10000) {
258 assert(delay_manager_.get());
259 return delay_manager_->SetMinimumDelay(delay_ms);
260 }
261 return false;
262 }
263
SetMaximumDelay(int delay_ms)264 bool NetEqImpl::SetMaximumDelay(int delay_ms) {
265 CriticalSectionScoped lock(crit_sect_.get());
266 if (delay_ms >= 0 && delay_ms < 10000) {
267 assert(delay_manager_.get());
268 return delay_manager_->SetMaximumDelay(delay_ms);
269 }
270 return false;
271 }
272
LeastRequiredDelayMs() const273 int NetEqImpl::LeastRequiredDelayMs() const {
274 CriticalSectionScoped lock(crit_sect_.get());
275 assert(delay_manager_.get());
276 return delay_manager_->least_required_delay_ms();
277 }
278
SetPlayoutMode(NetEqPlayoutMode mode)279 void NetEqImpl::SetPlayoutMode(NetEqPlayoutMode mode) {
280 CriticalSectionScoped lock(crit_sect_.get());
281 if (!decision_logic_.get() || mode != decision_logic_->playout_mode()) {
282 // The reset() method calls delete for the old object.
283 CreateDecisionLogic(mode);
284 }
285 }
286
PlayoutMode() const287 NetEqPlayoutMode NetEqImpl::PlayoutMode() const {
288 CriticalSectionScoped lock(crit_sect_.get());
289 assert(decision_logic_.get());
290 return decision_logic_->playout_mode();
291 }
292
NetworkStatistics(NetEqNetworkStatistics * stats)293 int NetEqImpl::NetworkStatistics(NetEqNetworkStatistics* stats) {
294 CriticalSectionScoped lock(crit_sect_.get());
295 assert(decoder_database_.get());
296 const int total_samples_in_buffers = packet_buffer_->NumSamplesInBuffer(
297 decoder_database_.get(), decoder_frame_length_) +
298 static_cast<int>(sync_buffer_->FutureLength());
299 assert(delay_manager_.get());
300 assert(decision_logic_.get());
301 stats_.GetNetworkStatistics(fs_hz_, total_samples_in_buffers,
302 decoder_frame_length_, *delay_manager_.get(),
303 *decision_logic_.get(), stats);
304 return 0;
305 }
306
WaitingTimes(std::vector<int> * waiting_times)307 void NetEqImpl::WaitingTimes(std::vector<int>* waiting_times) {
308 CriticalSectionScoped lock(crit_sect_.get());
309 stats_.WaitingTimes(waiting_times);
310 }
311
GetRtcpStatistics(RtcpStatistics * stats)312 void NetEqImpl::GetRtcpStatistics(RtcpStatistics* stats) {
313 CriticalSectionScoped lock(crit_sect_.get());
314 if (stats) {
315 rtcp_.GetStatistics(false, stats);
316 }
317 }
318
GetRtcpStatisticsNoReset(RtcpStatistics * stats)319 void NetEqImpl::GetRtcpStatisticsNoReset(RtcpStatistics* stats) {
320 CriticalSectionScoped lock(crit_sect_.get());
321 if (stats) {
322 rtcp_.GetStatistics(true, stats);
323 }
324 }
325
EnableVad()326 void NetEqImpl::EnableVad() {
327 CriticalSectionScoped lock(crit_sect_.get());
328 assert(vad_.get());
329 vad_->Enable();
330 }
331
DisableVad()332 void NetEqImpl::DisableVad() {
333 CriticalSectionScoped lock(crit_sect_.get());
334 assert(vad_.get());
335 vad_->Disable();
336 }
337
GetPlayoutTimestamp(uint32_t * timestamp)338 bool NetEqImpl::GetPlayoutTimestamp(uint32_t* timestamp) {
339 CriticalSectionScoped lock(crit_sect_.get());
340 if (first_packet_) {
341 // We don't have a valid RTP timestamp until we have decoded our first
342 // RTP packet.
343 return false;
344 }
345 *timestamp = timestamp_scaler_->ToExternal(playout_timestamp_);
346 return true;
347 }
348
LastError()349 int NetEqImpl::LastError() {
350 CriticalSectionScoped lock(crit_sect_.get());
351 return error_code_;
352 }
353
LastDecoderError()354 int NetEqImpl::LastDecoderError() {
355 CriticalSectionScoped lock(crit_sect_.get());
356 return decoder_error_code_;
357 }
358
FlushBuffers()359 void NetEqImpl::FlushBuffers() {
360 CriticalSectionScoped lock(crit_sect_.get());
361 LOG_API0();
362 packet_buffer_->Flush();
363 assert(sync_buffer_.get());
364 assert(expand_.get());
365 sync_buffer_->Flush();
366 sync_buffer_->set_next_index(sync_buffer_->next_index() -
367 expand_->overlap_length());
368 // Set to wait for new codec.
369 first_packet_ = true;
370 }
371
PacketBufferStatistics(int * current_num_packets,int * max_num_packets) const372 void NetEqImpl::PacketBufferStatistics(int* current_num_packets,
373 int* max_num_packets) const {
374 CriticalSectionScoped lock(crit_sect_.get());
375 packet_buffer_->BufferStat(current_num_packets, max_num_packets);
376 }
377
DecodedRtpInfo(int * sequence_number,uint32_t * timestamp) const378 int NetEqImpl::DecodedRtpInfo(int* sequence_number, uint32_t* timestamp) const {
379 CriticalSectionScoped lock(crit_sect_.get());
380 if (decoded_packet_sequence_number_ < 0)
381 return -1;
382 *sequence_number = decoded_packet_sequence_number_;
383 *timestamp = decoded_packet_timestamp_;
384 return 0;
385 }
386
SetBackgroundNoiseMode(NetEqBackgroundNoiseMode mode)387 void NetEqImpl::SetBackgroundNoiseMode(NetEqBackgroundNoiseMode mode) {
388 CriticalSectionScoped lock(crit_sect_.get());
389 assert(background_noise_.get());
390 background_noise_->set_mode(mode);
391 }
392
BackgroundNoiseMode() const393 NetEqBackgroundNoiseMode NetEqImpl::BackgroundNoiseMode() const {
394 CriticalSectionScoped lock(crit_sect_.get());
395 assert(background_noise_.get());
396 return background_noise_->mode();
397 }
398
sync_buffer_for_test() const399 const SyncBuffer* NetEqImpl::sync_buffer_for_test() const {
400 CriticalSectionScoped lock(crit_sect_.get());
401 return sync_buffer_.get();
402 }
403
404 // Methods below this line are private.
405
InsertPacketInternal(const WebRtcRTPHeader & rtp_header,const uint8_t * payload,int length_bytes,uint32_t receive_timestamp,bool is_sync_packet)406 int NetEqImpl::InsertPacketInternal(const WebRtcRTPHeader& rtp_header,
407 const uint8_t* payload,
408 int length_bytes,
409 uint32_t receive_timestamp,
410 bool is_sync_packet) {
411 if (!payload) {
412 LOG_F(LS_ERROR) << "payload == NULL";
413 return kInvalidPointer;
414 }
415 // Sanity checks for sync-packets.
416 if (is_sync_packet) {
417 if (decoder_database_->IsDtmf(rtp_header.header.payloadType) ||
418 decoder_database_->IsRed(rtp_header.header.payloadType) ||
419 decoder_database_->IsComfortNoise(rtp_header.header.payloadType)) {
420 LOG_F(LS_ERROR) << "Sync-packet with an unacceptable payload type "
421 << rtp_header.header.payloadType;
422 return kSyncPacketNotAccepted;
423 }
424 if (first_packet_ ||
425 rtp_header.header.payloadType != current_rtp_payload_type_ ||
426 rtp_header.header.ssrc != ssrc_) {
427 // Even if |current_rtp_payload_type_| is 0xFF, sync-packet isn't
428 // accepted.
429 LOG_F(LS_ERROR) << "Changing codec, SSRC or first packet "
430 "with sync-packet.";
431 return kSyncPacketNotAccepted;
432 }
433 }
434 PacketList packet_list;
435 RTPHeader main_header;
436 {
437 // Convert to Packet.
438 // Create |packet| within this separate scope, since it should not be used
439 // directly once it's been inserted in the packet list. This way, |packet|
440 // is not defined outside of this block.
441 Packet* packet = new Packet;
442 packet->header.markerBit = false;
443 packet->header.payloadType = rtp_header.header.payloadType;
444 packet->header.sequenceNumber = rtp_header.header.sequenceNumber;
445 packet->header.timestamp = rtp_header.header.timestamp;
446 packet->header.ssrc = rtp_header.header.ssrc;
447 packet->header.numCSRCs = 0;
448 packet->payload_length = length_bytes;
449 packet->primary = true;
450 packet->waiting_time = 0;
451 packet->payload = new uint8_t[packet->payload_length];
452 packet->sync_packet = is_sync_packet;
453 if (!packet->payload) {
454 LOG_F(LS_ERROR) << "Payload pointer is NULL.";
455 }
456 assert(payload); // Already checked above.
457 memcpy(packet->payload, payload, packet->payload_length);
458 // Insert packet in a packet list.
459 packet_list.push_back(packet);
460 // Save main payloads header for later.
461 memcpy(&main_header, &packet->header, sizeof(main_header));
462 }
463
464 bool update_sample_rate_and_channels = false;
465 // Reinitialize NetEq if it's needed (changed SSRC or first call).
466 if ((main_header.ssrc != ssrc_) || first_packet_) {
467 rtcp_.Init(main_header.sequenceNumber);
468 first_packet_ = false;
469
470 // Flush the packet buffer and DTMF buffer.
471 packet_buffer_->Flush();
472 dtmf_buffer_->Flush();
473
474 // Store new SSRC.
475 ssrc_ = main_header.ssrc;
476
477 // Update audio buffer timestamp.
478 sync_buffer_->IncreaseEndTimestamp(main_header.timestamp - timestamp_);
479
480 // Update codecs.
481 timestamp_ = main_header.timestamp;
482 current_rtp_payload_type_ = main_header.payloadType;
483
484 // Set MCU to update codec on next SignalMCU call.
485 new_codec_ = true;
486
487 // Reset timestamp scaling.
488 timestamp_scaler_->Reset();
489
490 // Triger an update of sampling rate and the number of channels.
491 update_sample_rate_and_channels = true;
492 }
493
494 // Update RTCP statistics, only for regular packets.
495 if (!is_sync_packet)
496 rtcp_.Update(main_header, receive_timestamp);
497
498 // Check for RED payload type, and separate payloads into several packets.
499 if (decoder_database_->IsRed(main_header.payloadType)) {
500 assert(!is_sync_packet); // We had a sanity check for this.
501 if (payload_splitter_->SplitRed(&packet_list) != PayloadSplitter::kOK) {
502 LOG_FERR1(LS_WARNING, SplitRed, packet_list.size());
503 PacketBuffer::DeleteAllPackets(&packet_list);
504 return kRedundancySplitError;
505 }
506 // Only accept a few RED payloads of the same type as the main data,
507 // DTMF events and CNG.
508 payload_splitter_->CheckRedPayloads(&packet_list, *decoder_database_);
509 // Update the stored main payload header since the main payload has now
510 // changed.
511 memcpy(&main_header, &packet_list.front()->header, sizeof(main_header));
512 }
513
514 // Check payload types.
515 if (decoder_database_->CheckPayloadTypes(packet_list) ==
516 DecoderDatabase::kDecoderNotFound) {
517 LOG_FERR1(LS_WARNING, CheckPayloadTypes, packet_list.size());
518 PacketBuffer::DeleteAllPackets(&packet_list);
519 return kUnknownRtpPayloadType;
520 }
521
522 // Scale timestamp to internal domain (only for some codecs).
523 timestamp_scaler_->ToInternal(&packet_list);
524
525 // Process DTMF payloads. Cycle through the list of packets, and pick out any
526 // DTMF payloads found.
527 PacketList::iterator it = packet_list.begin();
528 while (it != packet_list.end()) {
529 Packet* current_packet = (*it);
530 assert(current_packet);
531 assert(current_packet->payload);
532 if (decoder_database_->IsDtmf(current_packet->header.payloadType)) {
533 assert(!current_packet->sync_packet); // We had a sanity check for this.
534 DtmfEvent event;
535 int ret = DtmfBuffer::ParseEvent(
536 current_packet->header.timestamp,
537 current_packet->payload,
538 current_packet->payload_length,
539 &event);
540 if (ret != DtmfBuffer::kOK) {
541 LOG_FERR2(LS_WARNING, ParseEvent, ret,
542 current_packet->payload_length);
543 PacketBuffer::DeleteAllPackets(&packet_list);
544 return kDtmfParsingError;
545 }
546 if (dtmf_buffer_->InsertEvent(event) != DtmfBuffer::kOK) {
547 LOG_FERR0(LS_WARNING, InsertEvent);
548 PacketBuffer::DeleteAllPackets(&packet_list);
549 return kDtmfInsertError;
550 }
551 // TODO(hlundin): Let the destructor of Packet handle the payload.
552 delete [] current_packet->payload;
553 delete current_packet;
554 it = packet_list.erase(it);
555 } else {
556 ++it;
557 }
558 }
559
560 // Check for FEC in packets, and separate payloads into several packets.
561 int ret = payload_splitter_->SplitFec(&packet_list, decoder_database_.get());
562 if (ret != PayloadSplitter::kOK) {
563 LOG_FERR1(LS_WARNING, SplitFec, packet_list.size());
564 PacketBuffer::DeleteAllPackets(&packet_list);
565 switch (ret) {
566 case PayloadSplitter::kUnknownPayloadType:
567 return kUnknownRtpPayloadType;
568 default:
569 return kOtherError;
570 }
571 }
572
573 // Split payloads into smaller chunks. This also verifies that all payloads
574 // are of a known payload type. SplitAudio() method is protected against
575 // sync-packets.
576 ret = payload_splitter_->SplitAudio(&packet_list, *decoder_database_);
577 if (ret != PayloadSplitter::kOK) {
578 LOG_FERR1(LS_WARNING, SplitAudio, packet_list.size());
579 PacketBuffer::DeleteAllPackets(&packet_list);
580 switch (ret) {
581 case PayloadSplitter::kUnknownPayloadType:
582 return kUnknownRtpPayloadType;
583 case PayloadSplitter::kFrameSplitError:
584 return kFrameSplitError;
585 default:
586 return kOtherError;
587 }
588 }
589
590 // Update bandwidth estimate, if the packet is not sync-packet.
591 if (!packet_list.empty() && !packet_list.front()->sync_packet) {
592 // The list can be empty here if we got nothing but DTMF payloads.
593 AudioDecoder* decoder =
594 decoder_database_->GetDecoder(main_header.payloadType);
595 assert(decoder); // Should always get a valid object, since we have
596 // already checked that the payload types are known.
597 decoder->IncomingPacket(packet_list.front()->payload,
598 packet_list.front()->payload_length,
599 packet_list.front()->header.sequenceNumber,
600 packet_list.front()->header.timestamp,
601 receive_timestamp);
602 }
603
604 // Insert packets in buffer.
605 int temp_bufsize = packet_buffer_->NumPacketsInBuffer();
606 ret = packet_buffer_->InsertPacketList(
607 &packet_list,
608 *decoder_database_,
609 ¤t_rtp_payload_type_,
610 ¤t_cng_rtp_payload_type_);
611 if (ret == PacketBuffer::kFlushed) {
612 // Reset DSP timestamp etc. if packet buffer flushed.
613 new_codec_ = true;
614 update_sample_rate_and_channels = true;
615 LOG_F(LS_WARNING) << "Packet buffer flushed";
616 } else if (ret != PacketBuffer::kOK) {
617 LOG_FERR1(LS_WARNING, InsertPacketList, packet_list.size());
618 PacketBuffer::DeleteAllPackets(&packet_list);
619 return kOtherError;
620 }
621 if (current_rtp_payload_type_ != 0xFF) {
622 const DecoderDatabase::DecoderInfo* dec_info =
623 decoder_database_->GetDecoderInfo(current_rtp_payload_type_);
624 if (!dec_info) {
625 assert(false); // Already checked that the payload type is known.
626 }
627 }
628
629 if (update_sample_rate_and_channels && !packet_buffer_->Empty()) {
630 // We do not use |current_rtp_payload_type_| to |set payload_type|, but
631 // get the next RTP header from |packet_buffer_| to obtain the payload type.
632 // The reason for it is the following corner case. If NetEq receives a
633 // CNG packet with a sample rate different than the current CNG then it
634 // flushes its buffer, assuming send codec must have been changed. However,
635 // payload type of the hypothetically new send codec is not known.
636 const RTPHeader* rtp_header = packet_buffer_->NextRtpHeader();
637 assert(rtp_header);
638 int payload_type = rtp_header->payloadType;
639 AudioDecoder* decoder = decoder_database_->GetDecoder(payload_type);
640 assert(decoder); // Payloads are already checked to be valid.
641 const DecoderDatabase::DecoderInfo* decoder_info =
642 decoder_database_->GetDecoderInfo(payload_type);
643 assert(decoder_info);
644 if (decoder_info->fs_hz != fs_hz_ ||
645 decoder->channels() != algorithm_buffer_->Channels())
646 SetSampleRateAndChannels(decoder_info->fs_hz, decoder->channels());
647 }
648
649 // TODO(hlundin): Move this code to DelayManager class.
650 const DecoderDatabase::DecoderInfo* dec_info =
651 decoder_database_->GetDecoderInfo(main_header.payloadType);
652 assert(dec_info); // Already checked that the payload type is known.
653 delay_manager_->LastDecoderType(dec_info->codec_type);
654 if (delay_manager_->last_pack_cng_or_dtmf() == 0) {
655 // Calculate the total speech length carried in each packet.
656 temp_bufsize = packet_buffer_->NumPacketsInBuffer() - temp_bufsize;
657 temp_bufsize *= decoder_frame_length_;
658
659 if ((temp_bufsize > 0) &&
660 (temp_bufsize != decision_logic_->packet_length_samples())) {
661 decision_logic_->set_packet_length_samples(temp_bufsize);
662 delay_manager_->SetPacketAudioLength((1000 * temp_bufsize) / fs_hz_);
663 }
664
665 // Update statistics.
666 if ((int32_t) (main_header.timestamp - timestamp_) >= 0 &&
667 !new_codec_) {
668 // Only update statistics if incoming packet is not older than last played
669 // out packet, and if new codec flag is not set.
670 delay_manager_->Update(main_header.sequenceNumber, main_header.timestamp,
671 fs_hz_);
672 }
673 } else if (delay_manager_->last_pack_cng_or_dtmf() == -1) {
674 // This is first "normal" packet after CNG or DTMF.
675 // Reset packet time counter and measure time until next packet,
676 // but don't update statistics.
677 delay_manager_->set_last_pack_cng_or_dtmf(0);
678 delay_manager_->ResetPacketIatCount();
679 }
680 return 0;
681 }
682
GetAudioInternal(size_t max_length,int16_t * output,int * samples_per_channel,int * num_channels)683 int NetEqImpl::GetAudioInternal(size_t max_length, int16_t* output,
684 int* samples_per_channel, int* num_channels) {
685 PacketList packet_list;
686 DtmfEvent dtmf_event;
687 Operations operation;
688 bool play_dtmf;
689 int return_value = GetDecision(&operation, &packet_list, &dtmf_event,
690 &play_dtmf);
691 if (return_value != 0) {
692 LOG_FERR1(LS_WARNING, GetDecision, return_value);
693 assert(false);
694 last_mode_ = kModeError;
695 return return_value;
696 }
697 LOG(LS_VERBOSE) << "GetDecision returned operation=" << operation <<
698 " and " << packet_list.size() << " packet(s)";
699
700 AudioDecoder::SpeechType speech_type;
701 int length = 0;
702 int decode_return_value = Decode(&packet_list, &operation,
703 &length, &speech_type);
704
705 assert(vad_.get());
706 bool sid_frame_available =
707 (operation == kRfc3389Cng && !packet_list.empty());
708 vad_->Update(decoded_buffer_.get(), length, speech_type,
709 sid_frame_available, fs_hz_);
710
711 algorithm_buffer_->Clear();
712 switch (operation) {
713 case kNormal: {
714 DoNormal(decoded_buffer_.get(), length, speech_type, play_dtmf);
715 break;
716 }
717 case kMerge: {
718 DoMerge(decoded_buffer_.get(), length, speech_type, play_dtmf);
719 break;
720 }
721 case kExpand: {
722 return_value = DoExpand(play_dtmf);
723 break;
724 }
725 case kAccelerate: {
726 return_value = DoAccelerate(decoded_buffer_.get(), length, speech_type,
727 play_dtmf);
728 break;
729 }
730 case kPreemptiveExpand: {
731 return_value = DoPreemptiveExpand(decoded_buffer_.get(), length,
732 speech_type, play_dtmf);
733 break;
734 }
735 case kRfc3389Cng:
736 case kRfc3389CngNoPacket: {
737 return_value = DoRfc3389Cng(&packet_list, play_dtmf);
738 break;
739 }
740 case kCodecInternalCng: {
741 // This handles the case when there is no transmission and the decoder
742 // should produce internal comfort noise.
743 // TODO(hlundin): Write test for codec-internal CNG.
744 DoCodecInternalCng();
745 break;
746 }
747 case kDtmf: {
748 // TODO(hlundin): Write test for this.
749 return_value = DoDtmf(dtmf_event, &play_dtmf);
750 break;
751 }
752 case kAlternativePlc: {
753 // TODO(hlundin): Write test for this.
754 DoAlternativePlc(false);
755 break;
756 }
757 case kAlternativePlcIncreaseTimestamp: {
758 // TODO(hlundin): Write test for this.
759 DoAlternativePlc(true);
760 break;
761 }
762 case kAudioRepetitionIncreaseTimestamp: {
763 // TODO(hlundin): Write test for this.
764 sync_buffer_->IncreaseEndTimestamp(output_size_samples_);
765 // Skipping break on purpose. Execution should move on into the
766 // next case.
767 }
768 case kAudioRepetition: {
769 // TODO(hlundin): Write test for this.
770 // Copy last |output_size_samples_| from |sync_buffer_| to
771 // |algorithm_buffer|.
772 algorithm_buffer_->PushBackFromIndex(
773 *sync_buffer_, sync_buffer_->Size() - output_size_samples_);
774 expand_->Reset();
775 break;
776 }
777 case kUndefined: {
778 LOG_F(LS_ERROR) << "Invalid operation kUndefined.";
779 assert(false); // This should not happen.
780 last_mode_ = kModeError;
781 return kInvalidOperation;
782 }
783 } // End of switch.
784 if (return_value < 0) {
785 return return_value;
786 }
787
788 if (last_mode_ != kModeRfc3389Cng) {
789 comfort_noise_->Reset();
790 }
791
792 // Copy from |algorithm_buffer| to |sync_buffer_|.
793 sync_buffer_->PushBack(*algorithm_buffer_);
794
795 // Extract data from |sync_buffer_| to |output|.
796 size_t num_output_samples_per_channel = output_size_samples_;
797 size_t num_output_samples = output_size_samples_ * sync_buffer_->Channels();
798 if (num_output_samples > max_length) {
799 LOG(LS_WARNING) << "Output array is too short. " << max_length << " < " <<
800 output_size_samples_ << " * " << sync_buffer_->Channels();
801 num_output_samples = max_length;
802 num_output_samples_per_channel = static_cast<int>(
803 max_length / sync_buffer_->Channels());
804 }
805 int samples_from_sync = static_cast<int>(
806 sync_buffer_->GetNextAudioInterleaved(num_output_samples_per_channel,
807 output));
808 *num_channels = static_cast<int>(sync_buffer_->Channels());
809 LOG(LS_VERBOSE) << "Sync buffer (" << *num_channels << " channel(s)):" <<
810 " insert " << algorithm_buffer_->Size() << " samples, extract " <<
811 samples_from_sync << " samples";
812 if (samples_from_sync != output_size_samples_) {
813 LOG_F(LS_ERROR) << "samples_from_sync != output_size_samples_";
814 // TODO(minyue): treatment of under-run, filling zeros
815 memset(output, 0, num_output_samples * sizeof(int16_t));
816 *samples_per_channel = output_size_samples_;
817 return kSampleUnderrun;
818 }
819 *samples_per_channel = output_size_samples_;
820
821 // Should always have overlap samples left in the |sync_buffer_|.
822 assert(sync_buffer_->FutureLength() >= expand_->overlap_length());
823
824 if (play_dtmf) {
825 return_value = DtmfOverdub(dtmf_event, sync_buffer_->Channels(), output);
826 }
827
828 // Update the background noise parameters if last operation wrote data
829 // straight from the decoder to the |sync_buffer_|. That is, none of the
830 // operations that modify the signal can be followed by a parameter update.
831 if ((last_mode_ == kModeNormal) ||
832 (last_mode_ == kModeAccelerateFail) ||
833 (last_mode_ == kModePreemptiveExpandFail) ||
834 (last_mode_ == kModeRfc3389Cng) ||
835 (last_mode_ == kModeCodecInternalCng)) {
836 background_noise_->Update(*sync_buffer_, *vad_.get());
837 }
838
839 if (operation == kDtmf) {
840 // DTMF data was written the end of |sync_buffer_|.
841 // Update index to end of DTMF data in |sync_buffer_|.
842 sync_buffer_->set_dtmf_index(sync_buffer_->Size());
843 }
844
845 if (last_mode_ != kModeExpand) {
846 // If last operation was not expand, calculate the |playout_timestamp_| from
847 // the |sync_buffer_|. However, do not update the |playout_timestamp_| if it
848 // would be moved "backwards".
849 uint32_t temp_timestamp = sync_buffer_->end_timestamp() -
850 static_cast<uint32_t>(sync_buffer_->FutureLength());
851 if (static_cast<int32_t>(temp_timestamp - playout_timestamp_) > 0) {
852 playout_timestamp_ = temp_timestamp;
853 }
854 } else {
855 // Use dead reckoning to estimate the |playout_timestamp_|.
856 playout_timestamp_ += output_size_samples_;
857 }
858
859 if (decode_return_value) return decode_return_value;
860 return return_value;
861 }
862
GetDecision(Operations * operation,PacketList * packet_list,DtmfEvent * dtmf_event,bool * play_dtmf)863 int NetEqImpl::GetDecision(Operations* operation,
864 PacketList* packet_list,
865 DtmfEvent* dtmf_event,
866 bool* play_dtmf) {
867 // Initialize output variables.
868 *play_dtmf = false;
869 *operation = kUndefined;
870
871 // Increment time counters.
872 packet_buffer_->IncrementWaitingTimes();
873 stats_.IncreaseCounter(output_size_samples_, fs_hz_);
874
875 assert(sync_buffer_.get());
876 uint32_t end_timestamp = sync_buffer_->end_timestamp();
877 if (!new_codec_) {
878 packet_buffer_->DiscardOldPackets(end_timestamp);
879 }
880 const RTPHeader* header = packet_buffer_->NextRtpHeader();
881
882 if (decision_logic_->CngRfc3389On() || last_mode_ == kModeRfc3389Cng) {
883 // Because of timestamp peculiarities, we have to "manually" disallow using
884 // a CNG packet with the same timestamp as the one that was last played.
885 // This can happen when using redundancy and will cause the timing to shift.
886 while (header && decoder_database_->IsComfortNoise(header->payloadType) &&
887 (end_timestamp >= header->timestamp ||
888 end_timestamp + decision_logic_->generated_noise_samples() >
889 header->timestamp)) {
890 // Don't use this packet, discard it.
891 if (packet_buffer_->DiscardNextPacket() != PacketBuffer::kOK) {
892 assert(false); // Must be ok by design.
893 }
894 // Check buffer again.
895 if (!new_codec_) {
896 packet_buffer_->DiscardOldPackets(end_timestamp);
897 }
898 header = packet_buffer_->NextRtpHeader();
899 }
900 }
901
902 assert(expand_.get());
903 const int samples_left = static_cast<int>(sync_buffer_->FutureLength() -
904 expand_->overlap_length());
905 if (last_mode_ == kModeAccelerateSuccess ||
906 last_mode_ == kModeAccelerateLowEnergy ||
907 last_mode_ == kModePreemptiveExpandSuccess ||
908 last_mode_ == kModePreemptiveExpandLowEnergy) {
909 // Subtract (samples_left + output_size_samples_) from sampleMemory.
910 decision_logic_->AddSampleMemory(-(samples_left + output_size_samples_));
911 }
912
913 // Check if it is time to play a DTMF event.
914 if (dtmf_buffer_->GetEvent(end_timestamp +
915 decision_logic_->generated_noise_samples(),
916 dtmf_event)) {
917 *play_dtmf = true;
918 }
919
920 // Get instruction.
921 assert(sync_buffer_.get());
922 assert(expand_.get());
923 *operation = decision_logic_->GetDecision(*sync_buffer_,
924 *expand_,
925 decoder_frame_length_,
926 header,
927 last_mode_,
928 *play_dtmf,
929 &reset_decoder_);
930
931 // Check if we already have enough samples in the |sync_buffer_|. If so,
932 // change decision to normal, unless the decision was merge, accelerate, or
933 // preemptive expand.
934 if (samples_left >= output_size_samples_ &&
935 *operation != kMerge &&
936 *operation != kAccelerate &&
937 *operation != kPreemptiveExpand) {
938 *operation = kNormal;
939 return 0;
940 }
941
942 decision_logic_->ExpandDecision(*operation);
943
944 // Check conditions for reset.
945 if (new_codec_ || *operation == kUndefined) {
946 // The only valid reason to get kUndefined is that new_codec_ is set.
947 assert(new_codec_);
948 if (*play_dtmf && !header) {
949 timestamp_ = dtmf_event->timestamp;
950 } else {
951 assert(header);
952 if (!header) {
953 LOG_F(LS_ERROR) << "Packet missing where it shouldn't.";
954 return -1;
955 }
956 timestamp_ = header->timestamp;
957 if (*operation == kRfc3389CngNoPacket
958 #ifndef LEGACY_BITEXACT
959 // Without this check, it can happen that a non-CNG packet is sent to
960 // the CNG decoder as if it was a SID frame. This is clearly a bug,
961 // but is kept for now to maintain bit-exactness with the test
962 // vectors.
963 && decoder_database_->IsComfortNoise(header->payloadType)
964 #endif
965 ) {
966 // Change decision to CNG packet, since we do have a CNG packet, but it
967 // was considered too early to use. Now, use it anyway.
968 *operation = kRfc3389Cng;
969 } else if (*operation != kRfc3389Cng) {
970 *operation = kNormal;
971 }
972 }
973 // Adjust |sync_buffer_| timestamp before setting |end_timestamp| to the
974 // new value.
975 sync_buffer_->IncreaseEndTimestamp(timestamp_ - end_timestamp);
976 end_timestamp = timestamp_;
977 new_codec_ = false;
978 decision_logic_->SoftReset();
979 buffer_level_filter_->Reset();
980 delay_manager_->Reset();
981 stats_.ResetMcu();
982 }
983
984 int required_samples = output_size_samples_;
985 const int samples_10_ms = 80 * fs_mult_;
986 const int samples_20_ms = 2 * samples_10_ms;
987 const int samples_30_ms = 3 * samples_10_ms;
988
989 switch (*operation) {
990 case kExpand: {
991 timestamp_ = end_timestamp;
992 return 0;
993 }
994 case kRfc3389CngNoPacket:
995 case kCodecInternalCng: {
996 return 0;
997 }
998 case kDtmf: {
999 // TODO(hlundin): Write test for this.
1000 // Update timestamp.
1001 timestamp_ = end_timestamp;
1002 if (decision_logic_->generated_noise_samples() > 0 &&
1003 last_mode_ != kModeDtmf) {
1004 // Make a jump in timestamp due to the recently played comfort noise.
1005 uint32_t timestamp_jump = decision_logic_->generated_noise_samples();
1006 sync_buffer_->IncreaseEndTimestamp(timestamp_jump);
1007 timestamp_ += timestamp_jump;
1008 }
1009 decision_logic_->set_generated_noise_samples(0);
1010 return 0;
1011 }
1012 case kAccelerate: {
1013 // In order to do a accelerate we need at least 30 ms of audio data.
1014 if (samples_left >= samples_30_ms) {
1015 // Already have enough data, so we do not need to extract any more.
1016 decision_logic_->set_sample_memory(samples_left);
1017 decision_logic_->set_prev_time_scale(true);
1018 return 0;
1019 } else if (samples_left >= samples_10_ms &&
1020 decoder_frame_length_ >= samples_30_ms) {
1021 // Avoid decoding more data as it might overflow the playout buffer.
1022 *operation = kNormal;
1023 return 0;
1024 } else if (samples_left < samples_20_ms &&
1025 decoder_frame_length_ < samples_30_ms) {
1026 // Build up decoded data by decoding at least 20 ms of audio data. Do
1027 // not perform accelerate yet, but wait until we only need to do one
1028 // decoding.
1029 required_samples = 2 * output_size_samples_;
1030 *operation = kNormal;
1031 }
1032 // If none of the above is true, we have one of two possible situations:
1033 // (1) 20 ms <= samples_left < 30 ms and decoder_frame_length_ < 30 ms; or
1034 // (2) samples_left < 10 ms and decoder_frame_length_ >= 30 ms.
1035 // In either case, we move on with the accelerate decision, and decode one
1036 // frame now.
1037 break;
1038 }
1039 case kPreemptiveExpand: {
1040 // In order to do a preemptive expand we need at least 30 ms of decoded
1041 // audio data.
1042 if ((samples_left >= samples_30_ms) ||
1043 (samples_left >= samples_10_ms &&
1044 decoder_frame_length_ >= samples_30_ms)) {
1045 // Already have enough data, so we do not need to extract any more.
1046 // Or, avoid decoding more data as it might overflow the playout buffer.
1047 // Still try preemptive expand, though.
1048 decision_logic_->set_sample_memory(samples_left);
1049 decision_logic_->set_prev_time_scale(true);
1050 return 0;
1051 }
1052 if (samples_left < samples_20_ms &&
1053 decoder_frame_length_ < samples_30_ms) {
1054 // Build up decoded data by decoding at least 20 ms of audio data.
1055 // Still try to perform preemptive expand.
1056 required_samples = 2 * output_size_samples_;
1057 }
1058 // Move on with the preemptive expand decision.
1059 break;
1060 }
1061 case kMerge: {
1062 required_samples =
1063 std::max(merge_->RequiredFutureSamples(), required_samples);
1064 break;
1065 }
1066 default: {
1067 // Do nothing.
1068 }
1069 }
1070
1071 // Get packets from buffer.
1072 int extracted_samples = 0;
1073 if (header &&
1074 *operation != kAlternativePlc &&
1075 *operation != kAlternativePlcIncreaseTimestamp &&
1076 *operation != kAudioRepetition &&
1077 *operation != kAudioRepetitionIncreaseTimestamp) {
1078 sync_buffer_->IncreaseEndTimestamp(header->timestamp - end_timestamp);
1079 if (decision_logic_->CngOff()) {
1080 // Adjustment of timestamp only corresponds to an actual packet loss
1081 // if comfort noise is not played. If comfort noise was just played,
1082 // this adjustment of timestamp is only done to get back in sync with the
1083 // stream timestamp; no loss to report.
1084 stats_.LostSamples(header->timestamp - end_timestamp);
1085 }
1086
1087 if (*operation != kRfc3389Cng) {
1088 // We are about to decode and use a non-CNG packet.
1089 decision_logic_->SetCngOff();
1090 }
1091 // Reset CNG timestamp as a new packet will be delivered.
1092 // (Also if this is a CNG packet, since playedOutTS is updated.)
1093 decision_logic_->set_generated_noise_samples(0);
1094
1095 extracted_samples = ExtractPackets(required_samples, packet_list);
1096 if (extracted_samples < 0) {
1097 LOG_F(LS_WARNING) << "Failed to extract packets from buffer.";
1098 return kPacketBufferCorruption;
1099 }
1100 }
1101
1102 if (*operation == kAccelerate ||
1103 *operation == kPreemptiveExpand) {
1104 decision_logic_->set_sample_memory(samples_left + extracted_samples);
1105 decision_logic_->set_prev_time_scale(true);
1106 }
1107
1108 if (*operation == kAccelerate) {
1109 // Check that we have enough data (30ms) to do accelerate.
1110 if (extracted_samples + samples_left < samples_30_ms) {
1111 // TODO(hlundin): Write test for this.
1112 // Not enough, do normal operation instead.
1113 *operation = kNormal;
1114 }
1115 }
1116
1117 timestamp_ = end_timestamp;
1118 return 0;
1119 }
1120
Decode(PacketList * packet_list,Operations * operation,int * decoded_length,AudioDecoder::SpeechType * speech_type)1121 int NetEqImpl::Decode(PacketList* packet_list, Operations* operation,
1122 int* decoded_length,
1123 AudioDecoder::SpeechType* speech_type) {
1124 *speech_type = AudioDecoder::kSpeech;
1125 AudioDecoder* decoder = NULL;
1126 if (!packet_list->empty()) {
1127 const Packet* packet = packet_list->front();
1128 int payload_type = packet->header.payloadType;
1129 if (!decoder_database_->IsComfortNoise(payload_type)) {
1130 decoder = decoder_database_->GetDecoder(payload_type);
1131 assert(decoder);
1132 if (!decoder) {
1133 LOG_FERR1(LS_WARNING, GetDecoder, payload_type);
1134 PacketBuffer::DeleteAllPackets(packet_list);
1135 return kDecoderNotFound;
1136 }
1137 bool decoder_changed;
1138 decoder_database_->SetActiveDecoder(payload_type, &decoder_changed);
1139 if (decoder_changed) {
1140 // We have a new decoder. Re-init some values.
1141 const DecoderDatabase::DecoderInfo* decoder_info = decoder_database_
1142 ->GetDecoderInfo(payload_type);
1143 assert(decoder_info);
1144 if (!decoder_info) {
1145 LOG_FERR1(LS_WARNING, GetDecoderInfo, payload_type);
1146 PacketBuffer::DeleteAllPackets(packet_list);
1147 return kDecoderNotFound;
1148 }
1149 // If sampling rate or number of channels has changed, we need to make
1150 // a reset.
1151 if (decoder_info->fs_hz != fs_hz_ ||
1152 decoder->channels() != algorithm_buffer_->Channels()) {
1153 // TODO(tlegrand): Add unittest to cover this event.
1154 SetSampleRateAndChannels(decoder_info->fs_hz, decoder->channels());
1155 }
1156 sync_buffer_->set_end_timestamp(timestamp_);
1157 playout_timestamp_ = timestamp_;
1158 }
1159 }
1160 }
1161
1162 if (reset_decoder_) {
1163 // TODO(hlundin): Write test for this.
1164 // Reset decoder.
1165 if (decoder) {
1166 decoder->Init();
1167 }
1168 // Reset comfort noise decoder.
1169 AudioDecoder* cng_decoder = decoder_database_->GetActiveCngDecoder();
1170 if (cng_decoder) {
1171 cng_decoder->Init();
1172 }
1173 reset_decoder_ = false;
1174 }
1175
1176 #ifdef LEGACY_BITEXACT
1177 // Due to a bug in old SignalMCU, it could happen that CNG operation was
1178 // decided, but a speech packet was provided. The speech packet will be used
1179 // to update the comfort noise decoder, as if it was a SID frame, which is
1180 // clearly wrong.
1181 if (*operation == kRfc3389Cng) {
1182 return 0;
1183 }
1184 #endif
1185
1186 *decoded_length = 0;
1187 // Update codec-internal PLC state.
1188 if ((*operation == kMerge) && decoder && decoder->HasDecodePlc()) {
1189 decoder->DecodePlc(1, &decoded_buffer_[*decoded_length]);
1190 }
1191
1192 int return_value = DecodeLoop(packet_list, operation, decoder,
1193 decoded_length, speech_type);
1194
1195 if (*decoded_length < 0) {
1196 // Error returned from the decoder.
1197 *decoded_length = 0;
1198 sync_buffer_->IncreaseEndTimestamp(decoder_frame_length_);
1199 int error_code = 0;
1200 if (decoder)
1201 error_code = decoder->ErrorCode();
1202 if (error_code != 0) {
1203 // Got some error code from the decoder.
1204 decoder_error_code_ = error_code;
1205 return_value = kDecoderErrorCode;
1206 } else {
1207 // Decoder does not implement error codes. Return generic error.
1208 return_value = kOtherDecoderError;
1209 }
1210 LOG_FERR2(LS_WARNING, DecodeLoop, error_code, packet_list->size());
1211 *operation = kExpand; // Do expansion to get data instead.
1212 }
1213 if (*speech_type != AudioDecoder::kComfortNoise) {
1214 // Don't increment timestamp if codec returned CNG speech type
1215 // since in this case, the we will increment the CNGplayedTS counter.
1216 // Increase with number of samples per channel.
1217 assert(*decoded_length == 0 ||
1218 (decoder && decoder->channels() == sync_buffer_->Channels()));
1219 sync_buffer_->IncreaseEndTimestamp(
1220 *decoded_length / static_cast<int>(sync_buffer_->Channels()));
1221 }
1222 return return_value;
1223 }
1224
DecodeLoop(PacketList * packet_list,Operations * operation,AudioDecoder * decoder,int * decoded_length,AudioDecoder::SpeechType * speech_type)1225 int NetEqImpl::DecodeLoop(PacketList* packet_list, Operations* operation,
1226 AudioDecoder* decoder, int* decoded_length,
1227 AudioDecoder::SpeechType* speech_type) {
1228 Packet* packet = NULL;
1229 if (!packet_list->empty()) {
1230 packet = packet_list->front();
1231 }
1232 // Do decoding.
1233 while (packet &&
1234 !decoder_database_->IsComfortNoise(packet->header.payloadType)) {
1235 assert(decoder); // At this point, we must have a decoder object.
1236 // The number of channels in the |sync_buffer_| should be the same as the
1237 // number decoder channels.
1238 assert(sync_buffer_->Channels() == decoder->channels());
1239 assert(decoded_buffer_length_ >= kMaxFrameSize * decoder->channels());
1240 assert(*operation == kNormal || *operation == kAccelerate ||
1241 *operation == kMerge || *operation == kPreemptiveExpand);
1242 packet_list->pop_front();
1243 int payload_length = packet->payload_length;
1244 int16_t decode_length;
1245 if (packet->sync_packet) {
1246 // Decode to silence with the same frame size as the last decode.
1247 LOG(LS_VERBOSE) << "Decoding sync-packet: " <<
1248 " ts=" << packet->header.timestamp <<
1249 ", sn=" << packet->header.sequenceNumber <<
1250 ", pt=" << static_cast<int>(packet->header.payloadType) <<
1251 ", ssrc=" << packet->header.ssrc <<
1252 ", len=" << packet->payload_length;
1253 memset(&decoded_buffer_[*decoded_length], 0, decoder_frame_length_ *
1254 decoder->channels() * sizeof(decoded_buffer_[0]));
1255 decode_length = decoder_frame_length_;
1256 } else if (!packet->primary) {
1257 // This is a redundant payload; call the special decoder method.
1258 LOG(LS_VERBOSE) << "Decoding packet (redundant):" <<
1259 " ts=" << packet->header.timestamp <<
1260 ", sn=" << packet->header.sequenceNumber <<
1261 ", pt=" << static_cast<int>(packet->header.payloadType) <<
1262 ", ssrc=" << packet->header.ssrc <<
1263 ", len=" << packet->payload_length;
1264 decode_length = decoder->DecodeRedundant(
1265 packet->payload, packet->payload_length,
1266 &decoded_buffer_[*decoded_length], speech_type);
1267 } else {
1268 LOG(LS_VERBOSE) << "Decoding packet: ts=" << packet->header.timestamp <<
1269 ", sn=" << packet->header.sequenceNumber <<
1270 ", pt=" << static_cast<int>(packet->header.payloadType) <<
1271 ", ssrc=" << packet->header.ssrc <<
1272 ", len=" << packet->payload_length;
1273 decode_length = decoder->Decode(packet->payload,
1274 packet->payload_length,
1275 &decoded_buffer_[*decoded_length],
1276 speech_type);
1277 }
1278
1279 delete[] packet->payload;
1280 delete packet;
1281 packet = NULL;
1282 if (decode_length > 0) {
1283 *decoded_length += decode_length;
1284 // Update |decoder_frame_length_| with number of samples per channel.
1285 decoder_frame_length_ = decode_length /
1286 static_cast<int>(decoder->channels());
1287 LOG(LS_VERBOSE) << "Decoded " << decode_length << " samples (" <<
1288 decoder->channels() << " channel(s) -> " << decoder_frame_length_ <<
1289 " samples per channel)";
1290 } else if (decode_length < 0) {
1291 // Error.
1292 LOG_FERR2(LS_WARNING, Decode, decode_length, payload_length);
1293 *decoded_length = -1;
1294 PacketBuffer::DeleteAllPackets(packet_list);
1295 break;
1296 }
1297 if (*decoded_length > static_cast<int>(decoded_buffer_length_)) {
1298 // Guard against overflow.
1299 LOG_F(LS_WARNING) << "Decoded too much.";
1300 PacketBuffer::DeleteAllPackets(packet_list);
1301 return kDecodedTooMuch;
1302 }
1303 if (!packet_list->empty()) {
1304 packet = packet_list->front();
1305 } else {
1306 packet = NULL;
1307 }
1308 } // End of decode loop.
1309
1310 // If the list is not empty at this point, either a decoding error terminated
1311 // the while-loop, or list must hold exactly one CNG packet.
1312 assert(packet_list->empty() || *decoded_length < 0 ||
1313 (packet_list->size() == 1 && packet &&
1314 decoder_database_->IsComfortNoise(packet->header.payloadType)));
1315 return 0;
1316 }
1317
DoNormal(const int16_t * decoded_buffer,size_t decoded_length,AudioDecoder::SpeechType speech_type,bool play_dtmf)1318 void NetEqImpl::DoNormal(const int16_t* decoded_buffer, size_t decoded_length,
1319 AudioDecoder::SpeechType speech_type, bool play_dtmf) {
1320 assert(normal_.get());
1321 assert(mute_factor_array_.get());
1322 normal_->Process(decoded_buffer, decoded_length, last_mode_,
1323 mute_factor_array_.get(), algorithm_buffer_.get());
1324 if (decoded_length != 0) {
1325 last_mode_ = kModeNormal;
1326 }
1327
1328 // If last packet was decoded as an inband CNG, set mode to CNG instead.
1329 if ((speech_type == AudioDecoder::kComfortNoise)
1330 || ((last_mode_ == kModeCodecInternalCng)
1331 && (decoded_length == 0))) {
1332 // TODO(hlundin): Remove second part of || statement above.
1333 last_mode_ = kModeCodecInternalCng;
1334 }
1335
1336 if (!play_dtmf) {
1337 dtmf_tone_generator_->Reset();
1338 }
1339 }
1340
DoMerge(int16_t * decoded_buffer,size_t decoded_length,AudioDecoder::SpeechType speech_type,bool play_dtmf)1341 void NetEqImpl::DoMerge(int16_t* decoded_buffer, size_t decoded_length,
1342 AudioDecoder::SpeechType speech_type, bool play_dtmf) {
1343 assert(mute_factor_array_.get());
1344 assert(merge_.get());
1345 int new_length = merge_->Process(decoded_buffer, decoded_length,
1346 mute_factor_array_.get(),
1347 algorithm_buffer_.get());
1348
1349 // Update in-call and post-call statistics.
1350 if (expand_->MuteFactor(0) == 0) {
1351 // Expand generates only noise.
1352 stats_.ExpandedNoiseSamples(new_length - static_cast<int>(decoded_length));
1353 } else {
1354 // Expansion generates more than only noise.
1355 stats_.ExpandedVoiceSamples(new_length - static_cast<int>(decoded_length));
1356 }
1357
1358 last_mode_ = kModeMerge;
1359 // If last packet was decoded as an inband CNG, set mode to CNG instead.
1360 if (speech_type == AudioDecoder::kComfortNoise) {
1361 last_mode_ = kModeCodecInternalCng;
1362 }
1363 expand_->Reset();
1364 if (!play_dtmf) {
1365 dtmf_tone_generator_->Reset();
1366 }
1367 }
1368
DoExpand(bool play_dtmf)1369 int NetEqImpl::DoExpand(bool play_dtmf) {
1370 while ((sync_buffer_->FutureLength() - expand_->overlap_length()) <
1371 static_cast<size_t>(output_size_samples_)) {
1372 algorithm_buffer_->Clear();
1373 int return_value = expand_->Process(algorithm_buffer_.get());
1374 int length = static_cast<int>(algorithm_buffer_->Size());
1375
1376 // Update in-call and post-call statistics.
1377 if (expand_->MuteFactor(0) == 0) {
1378 // Expand operation generates only noise.
1379 stats_.ExpandedNoiseSamples(length);
1380 } else {
1381 // Expand operation generates more than only noise.
1382 stats_.ExpandedVoiceSamples(length);
1383 }
1384
1385 last_mode_ = kModeExpand;
1386
1387 if (return_value < 0) {
1388 return return_value;
1389 }
1390
1391 sync_buffer_->PushBack(*algorithm_buffer_);
1392 algorithm_buffer_->Clear();
1393 }
1394 if (!play_dtmf) {
1395 dtmf_tone_generator_->Reset();
1396 }
1397 return 0;
1398 }
1399
DoAccelerate(int16_t * decoded_buffer,size_t decoded_length,AudioDecoder::SpeechType speech_type,bool play_dtmf)1400 int NetEqImpl::DoAccelerate(int16_t* decoded_buffer, size_t decoded_length,
1401 AudioDecoder::SpeechType speech_type,
1402 bool play_dtmf) {
1403 const size_t required_samples = 240 * fs_mult_; // Must have 30 ms.
1404 size_t borrowed_samples_per_channel = 0;
1405 size_t num_channels = algorithm_buffer_->Channels();
1406 size_t decoded_length_per_channel = decoded_length / num_channels;
1407 if (decoded_length_per_channel < required_samples) {
1408 // Must move data from the |sync_buffer_| in order to get 30 ms.
1409 borrowed_samples_per_channel = static_cast<int>(required_samples -
1410 decoded_length_per_channel);
1411 memmove(&decoded_buffer[borrowed_samples_per_channel * num_channels],
1412 decoded_buffer,
1413 sizeof(int16_t) * decoded_length);
1414 sync_buffer_->ReadInterleavedFromEnd(borrowed_samples_per_channel,
1415 decoded_buffer);
1416 decoded_length = required_samples * num_channels;
1417 }
1418
1419 int16_t samples_removed;
1420 Accelerate::ReturnCodes return_code = accelerate_->Process(
1421 decoded_buffer, decoded_length, algorithm_buffer_.get(),
1422 &samples_removed);
1423 stats_.AcceleratedSamples(samples_removed);
1424 switch (return_code) {
1425 case Accelerate::kSuccess:
1426 last_mode_ = kModeAccelerateSuccess;
1427 break;
1428 case Accelerate::kSuccessLowEnergy:
1429 last_mode_ = kModeAccelerateLowEnergy;
1430 break;
1431 case Accelerate::kNoStretch:
1432 last_mode_ = kModeAccelerateFail;
1433 break;
1434 case Accelerate::kError:
1435 // TODO(hlundin): Map to kModeError instead?
1436 last_mode_ = kModeAccelerateFail;
1437 return kAccelerateError;
1438 }
1439
1440 if (borrowed_samples_per_channel > 0) {
1441 // Copy borrowed samples back to the |sync_buffer_|.
1442 size_t length = algorithm_buffer_->Size();
1443 if (length < borrowed_samples_per_channel) {
1444 // This destroys the beginning of the buffer, but will not cause any
1445 // problems.
1446 sync_buffer_->ReplaceAtIndex(*algorithm_buffer_,
1447 sync_buffer_->Size() -
1448 borrowed_samples_per_channel);
1449 sync_buffer_->PushFrontZeros(borrowed_samples_per_channel - length);
1450 algorithm_buffer_->PopFront(length);
1451 assert(algorithm_buffer_->Empty());
1452 } else {
1453 sync_buffer_->ReplaceAtIndex(*algorithm_buffer_,
1454 borrowed_samples_per_channel,
1455 sync_buffer_->Size() -
1456 borrowed_samples_per_channel);
1457 algorithm_buffer_->PopFront(borrowed_samples_per_channel);
1458 }
1459 }
1460
1461 // If last packet was decoded as an inband CNG, set mode to CNG instead.
1462 if (speech_type == AudioDecoder::kComfortNoise) {
1463 last_mode_ = kModeCodecInternalCng;
1464 }
1465 if (!play_dtmf) {
1466 dtmf_tone_generator_->Reset();
1467 }
1468 expand_->Reset();
1469 return 0;
1470 }
1471
DoPreemptiveExpand(int16_t * decoded_buffer,size_t decoded_length,AudioDecoder::SpeechType speech_type,bool play_dtmf)1472 int NetEqImpl::DoPreemptiveExpand(int16_t* decoded_buffer,
1473 size_t decoded_length,
1474 AudioDecoder::SpeechType speech_type,
1475 bool play_dtmf) {
1476 const size_t required_samples = 240 * fs_mult_; // Must have 30 ms.
1477 size_t num_channels = algorithm_buffer_->Channels();
1478 int borrowed_samples_per_channel = 0;
1479 int old_borrowed_samples_per_channel = 0;
1480 size_t decoded_length_per_channel = decoded_length / num_channels;
1481 if (decoded_length_per_channel < required_samples) {
1482 // Must move data from the |sync_buffer_| in order to get 30 ms.
1483 borrowed_samples_per_channel = static_cast<int>(required_samples -
1484 decoded_length_per_channel);
1485 // Calculate how many of these were already played out.
1486 old_borrowed_samples_per_channel = static_cast<int>(
1487 borrowed_samples_per_channel - sync_buffer_->FutureLength());
1488 old_borrowed_samples_per_channel = std::max(
1489 0, old_borrowed_samples_per_channel);
1490 memmove(&decoded_buffer[borrowed_samples_per_channel * num_channels],
1491 decoded_buffer,
1492 sizeof(int16_t) * decoded_length);
1493 sync_buffer_->ReadInterleavedFromEnd(borrowed_samples_per_channel,
1494 decoded_buffer);
1495 decoded_length = required_samples * num_channels;
1496 }
1497
1498 int16_t samples_added;
1499 PreemptiveExpand::ReturnCodes return_code = preemptive_expand_->Process(
1500 decoded_buffer, static_cast<int>(decoded_length),
1501 old_borrowed_samples_per_channel,
1502 algorithm_buffer_.get(), &samples_added);
1503 stats_.PreemptiveExpandedSamples(samples_added);
1504 switch (return_code) {
1505 case PreemptiveExpand::kSuccess:
1506 last_mode_ = kModePreemptiveExpandSuccess;
1507 break;
1508 case PreemptiveExpand::kSuccessLowEnergy:
1509 last_mode_ = kModePreemptiveExpandLowEnergy;
1510 break;
1511 case PreemptiveExpand::kNoStretch:
1512 last_mode_ = kModePreemptiveExpandFail;
1513 break;
1514 case PreemptiveExpand::kError:
1515 // TODO(hlundin): Map to kModeError instead?
1516 last_mode_ = kModePreemptiveExpandFail;
1517 return kPreemptiveExpandError;
1518 }
1519
1520 if (borrowed_samples_per_channel > 0) {
1521 // Copy borrowed samples back to the |sync_buffer_|.
1522 sync_buffer_->ReplaceAtIndex(
1523 *algorithm_buffer_, borrowed_samples_per_channel,
1524 sync_buffer_->Size() - borrowed_samples_per_channel);
1525 algorithm_buffer_->PopFront(borrowed_samples_per_channel);
1526 }
1527
1528 // If last packet was decoded as an inband CNG, set mode to CNG instead.
1529 if (speech_type == AudioDecoder::kComfortNoise) {
1530 last_mode_ = kModeCodecInternalCng;
1531 }
1532 if (!play_dtmf) {
1533 dtmf_tone_generator_->Reset();
1534 }
1535 expand_->Reset();
1536 return 0;
1537 }
1538
DoRfc3389Cng(PacketList * packet_list,bool play_dtmf)1539 int NetEqImpl::DoRfc3389Cng(PacketList* packet_list, bool play_dtmf) {
1540 if (!packet_list->empty()) {
1541 // Must have exactly one SID frame at this point.
1542 assert(packet_list->size() == 1);
1543 Packet* packet = packet_list->front();
1544 packet_list->pop_front();
1545 if (!decoder_database_->IsComfortNoise(packet->header.payloadType)) {
1546 #ifdef LEGACY_BITEXACT
1547 // This can happen due to a bug in GetDecision. Change the payload type
1548 // to a CNG type, and move on. Note that this means that we are in fact
1549 // sending a non-CNG payload to the comfort noise decoder for decoding.
1550 // Clearly wrong, but will maintain bit-exactness with legacy.
1551 if (fs_hz_ == 8000) {
1552 packet->header.payloadType =
1553 decoder_database_->GetRtpPayloadType(kDecoderCNGnb);
1554 } else if (fs_hz_ == 16000) {
1555 packet->header.payloadType =
1556 decoder_database_->GetRtpPayloadType(kDecoderCNGwb);
1557 } else if (fs_hz_ == 32000) {
1558 packet->header.payloadType =
1559 decoder_database_->GetRtpPayloadType(kDecoderCNGswb32kHz);
1560 } else if (fs_hz_ == 48000) {
1561 packet->header.payloadType =
1562 decoder_database_->GetRtpPayloadType(kDecoderCNGswb48kHz);
1563 }
1564 assert(decoder_database_->IsComfortNoise(packet->header.payloadType));
1565 #else
1566 LOG(LS_ERROR) << "Trying to decode non-CNG payload as CNG.";
1567 return kOtherError;
1568 #endif
1569 }
1570 // UpdateParameters() deletes |packet|.
1571 if (comfort_noise_->UpdateParameters(packet) ==
1572 ComfortNoise::kInternalError) {
1573 LOG_FERR0(LS_WARNING, UpdateParameters);
1574 algorithm_buffer_->Zeros(output_size_samples_);
1575 return -comfort_noise_->internal_error_code();
1576 }
1577 }
1578 int cn_return = comfort_noise_->Generate(output_size_samples_,
1579 algorithm_buffer_.get());
1580 expand_->Reset();
1581 last_mode_ = kModeRfc3389Cng;
1582 if (!play_dtmf) {
1583 dtmf_tone_generator_->Reset();
1584 }
1585 if (cn_return == ComfortNoise::kInternalError) {
1586 LOG_FERR1(LS_WARNING, comfort_noise_->Generate, cn_return);
1587 decoder_error_code_ = comfort_noise_->internal_error_code();
1588 return kComfortNoiseErrorCode;
1589 } else if (cn_return == ComfortNoise::kUnknownPayloadType) {
1590 LOG_FERR1(LS_WARNING, comfort_noise_->Generate, cn_return);
1591 return kUnknownRtpPayloadType;
1592 }
1593 return 0;
1594 }
1595
DoCodecInternalCng()1596 void NetEqImpl::DoCodecInternalCng() {
1597 int length = 0;
1598 // TODO(hlundin): Will probably need a longer buffer for multi-channel.
1599 int16_t decoded_buffer[kMaxFrameSize];
1600 AudioDecoder* decoder = decoder_database_->GetActiveDecoder();
1601 if (decoder) {
1602 const uint8_t* dummy_payload = NULL;
1603 AudioDecoder::SpeechType speech_type;
1604 length = decoder->Decode(dummy_payload, 0, decoded_buffer, &speech_type);
1605 }
1606 assert(mute_factor_array_.get());
1607 normal_->Process(decoded_buffer, length, last_mode_, mute_factor_array_.get(),
1608 algorithm_buffer_.get());
1609 last_mode_ = kModeCodecInternalCng;
1610 expand_->Reset();
1611 }
1612
DoDtmf(const DtmfEvent & dtmf_event,bool * play_dtmf)1613 int NetEqImpl::DoDtmf(const DtmfEvent& dtmf_event, bool* play_dtmf) {
1614 // This block of the code and the block further down, handling |dtmf_switch|
1615 // are commented out. Otherwise playing out-of-band DTMF would fail in VoE
1616 // test, DtmfTest.ManualSuccessfullySendsOutOfBandTelephoneEvents. This is
1617 // equivalent to |dtmf_switch| always be false.
1618 //
1619 // See http://webrtc-codereview.appspot.com/1195004/ for discussion
1620 // On this issue. This change might cause some glitches at the point of
1621 // switch from audio to DTMF. Issue 1545 is filed to track this.
1622 //
1623 // bool dtmf_switch = false;
1624 // if ((last_mode_ != kModeDtmf) && dtmf_tone_generator_->initialized()) {
1625 // // Special case; see below.
1626 // // We must catch this before calling Generate, since |initialized| is
1627 // // modified in that call.
1628 // dtmf_switch = true;
1629 // }
1630
1631 int dtmf_return_value = 0;
1632 if (!dtmf_tone_generator_->initialized()) {
1633 // Initialize if not already done.
1634 dtmf_return_value = dtmf_tone_generator_->Init(fs_hz_, dtmf_event.event_no,
1635 dtmf_event.volume);
1636 }
1637
1638 if (dtmf_return_value == 0) {
1639 // Generate DTMF signal.
1640 dtmf_return_value = dtmf_tone_generator_->Generate(output_size_samples_,
1641 algorithm_buffer_.get());
1642 }
1643
1644 if (dtmf_return_value < 0) {
1645 algorithm_buffer_->Zeros(output_size_samples_);
1646 return dtmf_return_value;
1647 }
1648
1649 // if (dtmf_switch) {
1650 // // This is the special case where the previous operation was DTMF
1651 // // overdub, but the current instruction is "regular" DTMF. We must make
1652 // // sure that the DTMF does not have any discontinuities. The first DTMF
1653 // // sample that we generate now must be played out immediately, therefore
1654 // // it must be copied to the speech buffer.
1655 // // TODO(hlundin): This code seems incorrect. (Legacy.) Write test and
1656 // // verify correct operation.
1657 // assert(false);
1658 // // Must generate enough data to replace all of the |sync_buffer_|
1659 // // "future".
1660 // int required_length = sync_buffer_->FutureLength();
1661 // assert(dtmf_tone_generator_->initialized());
1662 // dtmf_return_value = dtmf_tone_generator_->Generate(required_length,
1663 // algorithm_buffer_);
1664 // assert((size_t) required_length == algorithm_buffer_->Size());
1665 // if (dtmf_return_value < 0) {
1666 // algorithm_buffer_->Zeros(output_size_samples_);
1667 // return dtmf_return_value;
1668 // }
1669 //
1670 // // Overwrite the "future" part of the speech buffer with the new DTMF
1671 // // data.
1672 // // TODO(hlundin): It seems that this overwriting has gone lost.
1673 // // Not adapted for multi-channel yet.
1674 // assert(algorithm_buffer_->Channels() == 1);
1675 // if (algorithm_buffer_->Channels() != 1) {
1676 // LOG(LS_WARNING) << "DTMF not supported for more than one channel";
1677 // return kStereoNotSupported;
1678 // }
1679 // // Shuffle the remaining data to the beginning of algorithm buffer.
1680 // algorithm_buffer_->PopFront(sync_buffer_->FutureLength());
1681 // }
1682
1683 sync_buffer_->IncreaseEndTimestamp(output_size_samples_);
1684 expand_->Reset();
1685 last_mode_ = kModeDtmf;
1686
1687 // Set to false because the DTMF is already in the algorithm buffer.
1688 *play_dtmf = false;
1689 return 0;
1690 }
1691
DoAlternativePlc(bool increase_timestamp)1692 void NetEqImpl::DoAlternativePlc(bool increase_timestamp) {
1693 AudioDecoder* decoder = decoder_database_->GetActiveDecoder();
1694 int length;
1695 if (decoder && decoder->HasDecodePlc()) {
1696 // Use the decoder's packet-loss concealment.
1697 // TODO(hlundin): Will probably need a longer buffer for multi-channel.
1698 int16_t decoded_buffer[kMaxFrameSize];
1699 length = decoder->DecodePlc(1, decoded_buffer);
1700 if (length > 0) {
1701 algorithm_buffer_->PushBackInterleaved(decoded_buffer, length);
1702 } else {
1703 length = 0;
1704 }
1705 } else {
1706 // Do simple zero-stuffing.
1707 length = output_size_samples_;
1708 algorithm_buffer_->Zeros(length);
1709 // By not advancing the timestamp, NetEq inserts samples.
1710 stats_.AddZeros(length);
1711 }
1712 if (increase_timestamp) {
1713 sync_buffer_->IncreaseEndTimestamp(length);
1714 }
1715 expand_->Reset();
1716 }
1717
DtmfOverdub(const DtmfEvent & dtmf_event,size_t num_channels,int16_t * output) const1718 int NetEqImpl::DtmfOverdub(const DtmfEvent& dtmf_event, size_t num_channels,
1719 int16_t* output) const {
1720 size_t out_index = 0;
1721 int overdub_length = output_size_samples_; // Default value.
1722
1723 if (sync_buffer_->dtmf_index() > sync_buffer_->next_index()) {
1724 // Special operation for transition from "DTMF only" to "DTMF overdub".
1725 out_index = std::min(
1726 sync_buffer_->dtmf_index() - sync_buffer_->next_index(),
1727 static_cast<size_t>(output_size_samples_));
1728 overdub_length = output_size_samples_ - static_cast<int>(out_index);
1729 }
1730
1731 AudioMultiVector dtmf_output(num_channels);
1732 int dtmf_return_value = 0;
1733 if (!dtmf_tone_generator_->initialized()) {
1734 dtmf_return_value = dtmf_tone_generator_->Init(fs_hz_, dtmf_event.event_no,
1735 dtmf_event.volume);
1736 }
1737 if (dtmf_return_value == 0) {
1738 dtmf_return_value = dtmf_tone_generator_->Generate(overdub_length,
1739 &dtmf_output);
1740 assert((size_t) overdub_length == dtmf_output.Size());
1741 }
1742 dtmf_output.ReadInterleaved(overdub_length, &output[out_index]);
1743 return dtmf_return_value < 0 ? dtmf_return_value : 0;
1744 }
1745
ExtractPackets(int required_samples,PacketList * packet_list)1746 int NetEqImpl::ExtractPackets(int required_samples, PacketList* packet_list) {
1747 bool first_packet = true;
1748 uint8_t prev_payload_type = 0;
1749 uint32_t prev_timestamp = 0;
1750 uint16_t prev_sequence_number = 0;
1751 bool next_packet_available = false;
1752
1753 const RTPHeader* header = packet_buffer_->NextRtpHeader();
1754 assert(header);
1755 if (!header) {
1756 return -1;
1757 }
1758 uint32_t first_timestamp = header->timestamp;
1759 int extracted_samples = 0;
1760
1761 // Packet extraction loop.
1762 do {
1763 timestamp_ = header->timestamp;
1764 int discard_count = 0;
1765 Packet* packet = packet_buffer_->GetNextPacket(&discard_count);
1766 // |header| may be invalid after the |packet_buffer_| operation.
1767 header = NULL;
1768 if (!packet) {
1769 LOG_FERR1(LS_ERROR, GetNextPacket, discard_count) <<
1770 "Should always be able to extract a packet here";
1771 assert(false); // Should always be able to extract a packet here.
1772 return -1;
1773 }
1774 stats_.PacketsDiscarded(discard_count);
1775 // Store waiting time in ms; packets->waiting_time is in "output blocks".
1776 stats_.StoreWaitingTime(packet->waiting_time * kOutputSizeMs);
1777 assert(packet->payload_length > 0);
1778 packet_list->push_back(packet); // Store packet in list.
1779
1780 if (first_packet) {
1781 first_packet = false;
1782 decoded_packet_sequence_number_ = prev_sequence_number =
1783 packet->header.sequenceNumber;
1784 decoded_packet_timestamp_ = prev_timestamp = packet->header.timestamp;
1785 prev_payload_type = packet->header.payloadType;
1786 }
1787
1788 // Store number of extracted samples.
1789 int packet_duration = 0;
1790 AudioDecoder* decoder = decoder_database_->GetDecoder(
1791 packet->header.payloadType);
1792 if (decoder) {
1793 if (packet->sync_packet) {
1794 packet_duration = decoder_frame_length_;
1795 } else {
1796 packet_duration = packet->primary ?
1797 decoder->PacketDuration(packet->payload, packet->payload_length) :
1798 decoder->PacketDurationRedundant(packet->payload,
1799 packet->payload_length);
1800 }
1801 } else {
1802 LOG_FERR1(LS_WARNING, GetDecoder, packet->header.payloadType) <<
1803 "Could not find a decoder for a packet about to be extracted.";
1804 assert(false);
1805 }
1806 if (packet_duration <= 0) {
1807 // Decoder did not return a packet duration. Assume that the packet
1808 // contains the same number of samples as the previous one.
1809 packet_duration = decoder_frame_length_;
1810 }
1811 extracted_samples = packet->header.timestamp - first_timestamp +
1812 packet_duration;
1813
1814 // Check what packet is available next.
1815 header = packet_buffer_->NextRtpHeader();
1816 next_packet_available = false;
1817 if (header && prev_payload_type == header->payloadType) {
1818 int16_t seq_no_diff = header->sequenceNumber - prev_sequence_number;
1819 int32_t ts_diff = header->timestamp - prev_timestamp;
1820 if (seq_no_diff == 1 ||
1821 (seq_no_diff == 0 && ts_diff == decoder_frame_length_)) {
1822 // The next sequence number is available, or the next part of a packet
1823 // that was split into pieces upon insertion.
1824 next_packet_available = true;
1825 }
1826 prev_sequence_number = header->sequenceNumber;
1827 }
1828 } while (extracted_samples < required_samples && next_packet_available);
1829
1830 return extracted_samples;
1831 }
1832
UpdatePlcComponents(int fs_hz,size_t channels)1833 void NetEqImpl::UpdatePlcComponents(int fs_hz, size_t channels) {
1834 // Delete objects and create new ones.
1835 expand_.reset(expand_factory_->Create(background_noise_.get(),
1836 sync_buffer_.get(), &random_vector_,
1837 fs_hz, channels));
1838 merge_.reset(new Merge(fs_hz, channels, expand_.get(), sync_buffer_.get()));
1839 }
1840
SetSampleRateAndChannels(int fs_hz,size_t channels)1841 void NetEqImpl::SetSampleRateAndChannels(int fs_hz, size_t channels) {
1842 LOG_API2(fs_hz, channels);
1843 // TODO(hlundin): Change to an enumerator and skip assert.
1844 assert(fs_hz == 8000 || fs_hz == 16000 || fs_hz == 32000 || fs_hz == 48000);
1845 assert(channels > 0);
1846
1847 fs_hz_ = fs_hz;
1848 fs_mult_ = fs_hz / 8000;
1849 output_size_samples_ = kOutputSizeMs * 8 * fs_mult_;
1850 decoder_frame_length_ = 3 * output_size_samples_; // Initialize to 30ms.
1851
1852 last_mode_ = kModeNormal;
1853
1854 // Create a new array of mute factors and set all to 1.
1855 mute_factor_array_.reset(new int16_t[channels]);
1856 for (size_t i = 0; i < channels; ++i) {
1857 mute_factor_array_[i] = 16384; // 1.0 in Q14.
1858 }
1859
1860 // Reset comfort noise decoder, if there is one active.
1861 AudioDecoder* cng_decoder = decoder_database_->GetActiveCngDecoder();
1862 if (cng_decoder) {
1863 cng_decoder->Init();
1864 }
1865
1866 // Reinit post-decode VAD with new sample rate.
1867 assert(vad_.get()); // Cannot be NULL here.
1868 vad_->Init();
1869
1870 // Delete algorithm buffer and create a new one.
1871 algorithm_buffer_.reset(new AudioMultiVector(channels));
1872
1873 // Delete sync buffer and create a new one.
1874 sync_buffer_.reset(new SyncBuffer(channels, kSyncBufferSize * fs_mult_));
1875
1876
1877 // Delete BackgroundNoise object and create a new one, while preserving its
1878 // mode.
1879 NetEqBackgroundNoiseMode current_mode = kBgnOn;
1880 if (background_noise_.get())
1881 current_mode = background_noise_->mode();
1882 background_noise_.reset(new BackgroundNoise(channels));
1883 background_noise_->set_mode(current_mode);
1884
1885 // Reset random vector.
1886 random_vector_.Reset();
1887
1888 UpdatePlcComponents(fs_hz, channels);
1889
1890 // Move index so that we create a small set of future samples (all 0).
1891 sync_buffer_->set_next_index(sync_buffer_->next_index() -
1892 expand_->overlap_length());
1893
1894 normal_.reset(new Normal(fs_hz, decoder_database_.get(), *background_noise_,
1895 expand_.get()));
1896 accelerate_.reset(
1897 accelerate_factory_->Create(fs_hz, channels, *background_noise_));
1898 preemptive_expand_.reset(preemptive_expand_factory_->Create(
1899 fs_hz, channels,
1900 *background_noise_,
1901 static_cast<int>(expand_->overlap_length())));
1902
1903 // Delete ComfortNoise object and create a new one.
1904 comfort_noise_.reset(new ComfortNoise(fs_hz, decoder_database_.get(),
1905 sync_buffer_.get()));
1906
1907 // Verify that |decoded_buffer_| is long enough.
1908 if (decoded_buffer_length_ < kMaxFrameSize * channels) {
1909 // Reallocate to larger size.
1910 decoded_buffer_length_ = kMaxFrameSize * channels;
1911 decoded_buffer_.reset(new int16_t[decoded_buffer_length_]);
1912 }
1913
1914 // Create DecisionLogic if it is not created yet, then communicate new sample
1915 // rate and output size to DecisionLogic object.
1916 if (!decision_logic_.get()) {
1917 CreateDecisionLogic(kPlayoutOn);
1918 }
1919 decision_logic_->SetSampleRate(fs_hz_, output_size_samples_);
1920 }
1921
LastOutputType()1922 NetEqOutputType NetEqImpl::LastOutputType() {
1923 assert(vad_.get());
1924 assert(expand_.get());
1925 if (last_mode_ == kModeCodecInternalCng || last_mode_ == kModeRfc3389Cng) {
1926 return kOutputCNG;
1927 } else if (last_mode_ == kModeExpand && expand_->MuteFactor(0) == 0) {
1928 // Expand mode has faded down to background noise only (very long expand).
1929 return kOutputPLCtoCNG;
1930 } else if (last_mode_ == kModeExpand) {
1931 return kOutputPLC;
1932 } else if (vad_->running() && !vad_->active_speech()) {
1933 return kOutputVADPassive;
1934 } else {
1935 return kOutputNormal;
1936 }
1937 }
1938
CreateDecisionLogic(NetEqPlayoutMode mode)1939 void NetEqImpl::CreateDecisionLogic(NetEqPlayoutMode mode) {
1940 decision_logic_.reset(DecisionLogic::Create(fs_hz_, output_size_samples_,
1941 mode,
1942 decoder_database_.get(),
1943 *packet_buffer_.get(),
1944 delay_manager_.get(),
1945 buffer_level_filter_.get()));
1946 }
1947 } // namespace webrtc
1948