2 * Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
11 #ifndef WEBRTC_MODULES_AUDIO_CODING_NETEQ_NETEQ_IMPL_H_
12 #define WEBRTC_MODULES_AUDIO_CODING_NETEQ_NETEQ_IMPL_H_
16 #include "webrtc/base/constructormagic.h"
17 #include "webrtc/modules/audio_coding/neteq/audio_multi_vector.h"
18 #include "webrtc/modules/audio_coding/neteq/defines.h"
19 #include "webrtc/modules/audio_coding/neteq/interface/neteq.h"
20 #include "webrtc/modules/audio_coding/neteq/packet.h" // Declare PacketList.
21 #include "webrtc/modules/audio_coding/neteq/random_vector.h"
22 #include "webrtc/modules/audio_coding/neteq/rtcp.h"
23 #include "webrtc/modules/audio_coding/neteq/statistics_calculator.h"
24 #include "webrtc/system_wrappers/interface/scoped_ptr.h"
25 #include "webrtc/system_wrappers/interface/thread_annotations.h"
26 #include "webrtc/typedefs.h"
30 // Forward declarations.
32 class BackgroundNoise;
33 class BufferLevelFilter;
35 class CriticalSectionWrapper;
37 class DecoderDatabase;
39 class DelayPeakDetector;
41 class DtmfToneGenerator;
46 class PayloadSplitter;
48 class PreemptiveExpand;
51 class TimestampScaler;
52 struct AccelerateFactory;
55 struct PreemptiveExpandFactory;
57 class NetEqImpl : public webrtc::NetEq {
59 // Creates a new NetEqImpl object. The object will assume ownership of all
60 // injected dependencies, and will delete them when done.
61 NetEqImpl(const NetEq::Config& config,
62 BufferLevelFilter* buffer_level_filter,
63 DecoderDatabase* decoder_database,
64 DelayManager* delay_manager,
65 DelayPeakDetector* delay_peak_detector,
66 DtmfBuffer* dtmf_buffer,
67 DtmfToneGenerator* dtmf_tone_generator,
68 PacketBuffer* packet_buffer,
69 PayloadSplitter* payload_splitter,
70 TimestampScaler* timestamp_scaler,
71 AccelerateFactory* accelerate_factory,
72 ExpandFactory* expand_factory,
73 PreemptiveExpandFactory* preemptive_expand_factory,
74 bool create_components = true);
78 // Inserts a new packet into NetEq. The |receive_timestamp| is an indication
79 // of the time when the packet was received, and should be measured with
80 // the same tick rate as the RTP timestamp of the current payload.
81 // Returns 0 on success, -1 on failure.
82 virtual int InsertPacket(const WebRtcRTPHeader& rtp_header,
83 const uint8_t* payload,
85 uint32_t receive_timestamp);
87 // Inserts a sync-packet into packet queue. Sync-packets are decoded to
88 // silence and are intended to keep AV-sync intact in an event of long packet
89 // losses when Video NACK is enabled but Audio NACK is not. Clients of NetEq
90 // might insert sync-packet when they observe that buffer level of NetEq is
91 // decreasing below a certain threshold, defined by the application.
92 // Sync-packets should have the same payload type as the last audio payload
93 // type, i.e. they cannot have DTMF or CNG payload type, nor a codec change
94 // can be implied by inserting a sync-packet.
95 // Returns kOk on success, kFail on failure.
96 virtual int InsertSyncPacket(const WebRtcRTPHeader& rtp_header,
97 uint32_t receive_timestamp);
99 // Instructs NetEq to deliver 10 ms of audio data. The data is written to
100 // |output_audio|, which can hold (at least) |max_length| elements.
101 // The number of channels that were written to the output is provided in
102 // the output variable |num_channels|, and each channel contains
103 // |samples_per_channel| elements. If more than one channel is written,
104 // the samples are interleaved.
105 // The speech type is written to |type|, if |type| is not NULL.
106 // Returns kOK on success, or kFail in case of an error.
107 virtual int GetAudio(size_t max_length, int16_t* output_audio,
108 int* samples_per_channel, int* num_channels,
109 NetEqOutputType* type);
111 // Associates |rtp_payload_type| with |codec| and stores the information in
112 // the codec database. Returns kOK on success, kFail on failure.
113 virtual int RegisterPayloadType(enum NetEqDecoder codec,
114 uint8_t rtp_payload_type);
116 // Provides an externally created decoder object |decoder| to insert in the
117 // decoder database. The decoder implements a decoder of type |codec| and
118 // associates it with |rtp_payload_type|. Returns kOK on success, kFail on
120 virtual int RegisterExternalDecoder(AudioDecoder* decoder,
121 enum NetEqDecoder codec,
122 uint8_t rtp_payload_type);
124 // Removes |rtp_payload_type| from the codec database. Returns 0 on success,
126 virtual int RemovePayloadType(uint8_t rtp_payload_type);
128 virtual bool SetMinimumDelay(int delay_ms);
130 virtual bool SetMaximumDelay(int delay_ms);
132 virtual int LeastRequiredDelayMs() const;
134 virtual int SetTargetDelay() { return kNotImplemented; }
136 virtual int TargetDelay() { return kNotImplemented; }
138 virtual int CurrentDelay() { return kNotImplemented; }
140 // Sets the playout mode to |mode|.
141 virtual void SetPlayoutMode(NetEqPlayoutMode mode);
143 // Returns the current playout mode.
144 virtual NetEqPlayoutMode PlayoutMode() const;
146 // Writes the current network statistics to |stats|. The statistics are reset
148 virtual int NetworkStatistics(NetEqNetworkStatistics* stats);
150 // Writes the last packet waiting times (in ms) to |waiting_times|. The number
151 // of values written is no more than 100, but may be smaller if the interface
152 // is polled again before 100 packets has arrived.
153 virtual void WaitingTimes(std::vector<int>* waiting_times);
155 // Writes the current RTCP statistics to |stats|. The statistics are reset
156 // and a new report period is started with the call.
157 virtual void GetRtcpStatistics(RtcpStatistics* stats);
159 // Same as RtcpStatistics(), but does not reset anything.
160 virtual void GetRtcpStatisticsNoReset(RtcpStatistics* stats);
162 // Enables post-decode VAD. When enabled, GetAudio() will return
163 // kOutputVADPassive when the signal contains no speech.
164 virtual void EnableVad();
166 // Disables post-decode VAD.
167 virtual void DisableVad();
169 virtual bool GetPlayoutTimestamp(uint32_t* timestamp);
171 virtual int SetTargetNumberOfChannels() { return kNotImplemented; }
173 virtual int SetTargetSampleRate() { return kNotImplemented; }
175 // Returns the error code for the last occurred error. If no error has
176 // occurred, 0 is returned.
177 virtual int LastError();
179 // Returns the error code last returned by a decoder (audio or comfort noise).
180 // When LastError() returns kDecoderErrorCode or kComfortNoiseErrorCode, check
181 // this method to get the decoder's error code.
182 virtual int LastDecoderError();
184 // Flushes both the packet buffer and the sync buffer.
185 virtual void FlushBuffers();
187 virtual void PacketBufferStatistics(int* current_num_packets,
188 int* max_num_packets) const;
190 // Get sequence number and timestamp of the latest RTP.
191 // This method is to facilitate NACK.
192 virtual int DecodedRtpInfo(int* sequence_number, uint32_t* timestamp) const;
194 // This accessor method is only intended for testing purposes.
195 virtual const SyncBuffer* sync_buffer_for_test() const;
198 static const int kOutputSizeMs = 10;
199 static const int kMaxFrameSize = 2880; // 60 ms @ 48 kHz.
200 // TODO(hlundin): Provide a better value for kSyncBufferSize.
201 static const int kSyncBufferSize = 2 * kMaxFrameSize;
203 // Inserts a new packet into NetEq. This is used by the InsertPacket method
204 // above. Returns 0 on success, otherwise an error code.
205 // TODO(hlundin): Merge this with InsertPacket above?
206 int InsertPacketInternal(const WebRtcRTPHeader& rtp_header,
207 const uint8_t* payload,
209 uint32_t receive_timestamp,
211 EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
213 // Delivers 10 ms of audio data. The data is written to |output|, which can
214 // hold (at least) |max_length| elements. The number of channels that were
215 // written to the output is provided in the output variable |num_channels|,
216 // and each channel contains |samples_per_channel| elements. If more than one
217 // channel is written, the samples are interleaved.
218 // Returns 0 on success, otherwise an error code.
219 int GetAudioInternal(size_t max_length,
221 int* samples_per_channel,
222 int* num_channels) EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
224 // Provides a decision to the GetAudioInternal method. The decision what to
225 // do is written to |operation|. Packets to decode are written to
226 // |packet_list|, and a DTMF event to play is written to |dtmf_event|. When
227 // DTMF should be played, |play_dtmf| is set to true by the method.
228 // Returns 0 on success, otherwise an error code.
229 int GetDecision(Operations* operation,
230 PacketList* packet_list,
231 DtmfEvent* dtmf_event,
232 bool* play_dtmf) EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
234 // Decodes the speech packets in |packet_list|, and writes the results to
235 // |decoded_buffer|, which is allocated to hold |decoded_buffer_length|
236 // elements. The length of the decoded data is written to |decoded_length|.
237 // The speech type -- speech or (codec-internal) comfort noise -- is written
238 // to |speech_type|. If |packet_list| contains any SID frames for RFC 3389
239 // comfort noise, those are not decoded.
240 int Decode(PacketList* packet_list,
241 Operations* operation,
243 AudioDecoder::SpeechType* speech_type)
244 EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
246 // Sub-method to Decode(). Performs the actual decoding.
247 int DecodeLoop(PacketList* packet_list,
248 Operations* operation,
249 AudioDecoder* decoder,
251 AudioDecoder::SpeechType* speech_type)
252 EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
254 // Sub-method which calls the Normal class to perform the normal operation.
255 void DoNormal(const int16_t* decoded_buffer,
256 size_t decoded_length,
257 AudioDecoder::SpeechType speech_type,
258 bool play_dtmf) EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
260 // Sub-method which calls the Merge class to perform the merge operation.
261 void DoMerge(int16_t* decoded_buffer,
262 size_t decoded_length,
263 AudioDecoder::SpeechType speech_type,
264 bool play_dtmf) EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
266 // Sub-method which calls the Expand class to perform the expand operation.
267 int DoExpand(bool play_dtmf) EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
269 // Sub-method which calls the Accelerate class to perform the accelerate
271 int DoAccelerate(int16_t* decoded_buffer,
272 size_t decoded_length,
273 AudioDecoder::SpeechType speech_type,
274 bool play_dtmf) EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
276 // Sub-method which calls the PreemptiveExpand class to perform the
277 // preemtive expand operation.
278 int DoPreemptiveExpand(int16_t* decoded_buffer,
279 size_t decoded_length,
280 AudioDecoder::SpeechType speech_type,
281 bool play_dtmf) EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
283 // Sub-method which calls the ComfortNoise class to generate RFC 3389 comfort
284 // noise. |packet_list| can either contain one SID frame to update the
285 // noise parameters, or no payload at all, in which case the previously
286 // received parameters are used.
287 int DoRfc3389Cng(PacketList* packet_list, bool play_dtmf)
288 EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
290 // Calls the audio decoder to generate codec-internal comfort noise when
291 // no packet was received.
292 void DoCodecInternalCng() EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
294 // Calls the DtmfToneGenerator class to generate DTMF tones.
295 int DoDtmf(const DtmfEvent& dtmf_event, bool* play_dtmf)
296 EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
298 // Produces packet-loss concealment using alternative methods. If the codec
299 // has an internal PLC, it is called to generate samples. Otherwise, the
300 // method performs zero-stuffing.
301 void DoAlternativePlc(bool increase_timestamp)
302 EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
304 // Overdub DTMF on top of |output|.
305 int DtmfOverdub(const DtmfEvent& dtmf_event,
307 int16_t* output) const EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
309 // Extracts packets from |packet_buffer_| to produce at least
310 // |required_samples| samples. The packets are inserted into |packet_list|.
311 // Returns the number of samples that the packets in the list will produce, or
312 // -1 in case of an error.
313 int ExtractPackets(int required_samples, PacketList* packet_list)
314 EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
316 // Resets various variables and objects to new values based on the sample rate
317 // |fs_hz| and |channels| number audio channels.
318 void SetSampleRateAndChannels(int fs_hz, size_t channels)
319 EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
321 // Returns the output type for the audio produced by the latest call to
323 NetEqOutputType LastOutputType() EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
325 // Updates Expand and Merge.
326 virtual void UpdatePlcComponents(int fs_hz, size_t channels)
327 EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
329 // Creates DecisionLogic object for the given mode.
330 virtual void CreateDecisionLogic(NetEqPlayoutMode mode)
331 EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
333 const scoped_ptr<CriticalSectionWrapper> crit_sect_;
334 const scoped_ptr<BufferLevelFilter> buffer_level_filter_
335 GUARDED_BY(crit_sect_);
336 const scoped_ptr<DecoderDatabase> decoder_database_ GUARDED_BY(crit_sect_);
337 const scoped_ptr<DelayManager> delay_manager_ GUARDED_BY(crit_sect_);
338 const scoped_ptr<DelayPeakDetector> delay_peak_detector_
339 GUARDED_BY(crit_sect_);
340 const scoped_ptr<DtmfBuffer> dtmf_buffer_ GUARDED_BY(crit_sect_);
341 const scoped_ptr<DtmfToneGenerator> dtmf_tone_generator_
342 GUARDED_BY(crit_sect_);
343 const scoped_ptr<PacketBuffer> packet_buffer_ GUARDED_BY(crit_sect_);
344 const scoped_ptr<PayloadSplitter> payload_splitter_ GUARDED_BY(crit_sect_);
345 const scoped_ptr<TimestampScaler> timestamp_scaler_ GUARDED_BY(crit_sect_);
346 const scoped_ptr<PostDecodeVad> vad_ GUARDED_BY(crit_sect_);
347 const scoped_ptr<ExpandFactory> expand_factory_ GUARDED_BY(crit_sect_);
348 const scoped_ptr<AccelerateFactory> accelerate_factory_
349 GUARDED_BY(crit_sect_);
350 const scoped_ptr<PreemptiveExpandFactory> preemptive_expand_factory_
351 GUARDED_BY(crit_sect_);
353 scoped_ptr<BackgroundNoise> background_noise_ GUARDED_BY(crit_sect_);
354 scoped_ptr<DecisionLogic> decision_logic_ GUARDED_BY(crit_sect_);
355 scoped_ptr<AudioMultiVector> algorithm_buffer_ GUARDED_BY(crit_sect_);
356 scoped_ptr<SyncBuffer> sync_buffer_ GUARDED_BY(crit_sect_);
357 scoped_ptr<Expand> expand_ GUARDED_BY(crit_sect_);
358 scoped_ptr<Normal> normal_ GUARDED_BY(crit_sect_);
359 scoped_ptr<Merge> merge_ GUARDED_BY(crit_sect_);
360 scoped_ptr<Accelerate> accelerate_ GUARDED_BY(crit_sect_);
361 scoped_ptr<PreemptiveExpand> preemptive_expand_ GUARDED_BY(crit_sect_);
362 RandomVector random_vector_ GUARDED_BY(crit_sect_);
363 scoped_ptr<ComfortNoise> comfort_noise_ GUARDED_BY(crit_sect_);
364 Rtcp rtcp_ GUARDED_BY(crit_sect_);
365 StatisticsCalculator stats_ GUARDED_BY(crit_sect_);
366 int fs_hz_ GUARDED_BY(crit_sect_);
367 int fs_mult_ GUARDED_BY(crit_sect_);
368 int output_size_samples_ GUARDED_BY(crit_sect_);
369 int decoder_frame_length_ GUARDED_BY(crit_sect_);
370 Modes last_mode_ GUARDED_BY(crit_sect_);
371 scoped_ptr<int16_t[]> mute_factor_array_ GUARDED_BY(crit_sect_);
372 size_t decoded_buffer_length_ GUARDED_BY(crit_sect_);
373 scoped_ptr<int16_t[]> decoded_buffer_ GUARDED_BY(crit_sect_);
374 uint32_t playout_timestamp_ GUARDED_BY(crit_sect_);
375 bool new_codec_ GUARDED_BY(crit_sect_);
376 uint32_t timestamp_ GUARDED_BY(crit_sect_);
377 bool reset_decoder_ GUARDED_BY(crit_sect_);
378 uint8_t current_rtp_payload_type_ GUARDED_BY(crit_sect_);
379 uint8_t current_cng_rtp_payload_type_ GUARDED_BY(crit_sect_);
380 uint32_t ssrc_ GUARDED_BY(crit_sect_);
381 bool first_packet_ GUARDED_BY(crit_sect_);
382 int error_code_ GUARDED_BY(crit_sect_); // Store last error code.
383 int decoder_error_code_ GUARDED_BY(crit_sect_);
384 const BackgroundNoiseMode background_noise_mode_ GUARDED_BY(crit_sect_);
386 // These values are used by NACK module to estimate time-to-play of
387 // a missing packet. Occasionally, NetEq might decide to decode more
388 // than one packet. Therefore, these values store sequence number and
389 // timestamp of the first packet pulled from the packet buffer. In
390 // such cases, these values do not exactly represent the sequence number
391 // or timestamp associated with a 10ms audio pulled from NetEq. NACK
392 // module is designed to compensate for this.
393 int decoded_packet_sequence_number_ GUARDED_BY(crit_sect_);
394 uint32_t decoded_packet_timestamp_ GUARDED_BY(crit_sect_);
397 DISALLOW_COPY_AND_ASSIGN(NetEqImpl);
400 } // namespace webrtc
401 #endif // WEBRTC_MODULES_AUDIO_CODING_NETEQ_NETEQ_IMPL_H_