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 #ifndef MODULES_AUDIO_CODING_NETEQ_NETEQ_IMPL_H_
12 #define MODULES_AUDIO_CODING_NETEQ_NETEQ_IMPL_H_
13 
14 #include <memory>
15 #include <string>
16 
17 #include "api/optional.h"
18 #include "modules/audio_coding/neteq/audio_multi_vector.h"
19 #include "modules/audio_coding/neteq/defines.h"
20 #include "modules/audio_coding/neteq/include/neteq.h"
21 #include "modules/audio_coding/neteq/packet.h"  // Declare PacketList.
22 #include "modules/audio_coding/neteq/random_vector.h"
23 #include "modules/audio_coding/neteq/rtcp.h"
24 #include "modules/audio_coding/neteq/statistics_calculator.h"
25 #include "modules/audio_coding/neteq/tick_timer.h"
26 #include "modules/include/module_common_types.h"
27 #include "rtc_base/constructormagic.h"
28 #include "rtc_base/criticalsection.h"
29 #include "rtc_base/thread_annotations.h"
30 #include "typedefs.h"  // NOLINT(build/include)
31 
32 namespace webrtc {
33 
34 // Forward declarations.
35 class Accelerate;
36 class BackgroundNoise;
37 class BufferLevelFilter;
38 class ComfortNoise;
39 class DecisionLogic;
40 class DecoderDatabase;
41 class DelayManager;
42 class DelayPeakDetector;
43 class DtmfBuffer;
44 class DtmfToneGenerator;
45 class Expand;
46 class Merge;
47 class NackTracker;
48 class Normal;
49 class PacketBuffer;
50 class RedPayloadSplitter;
51 class PostDecodeVad;
52 class PreemptiveExpand;
53 class RandomVector;
54 class SyncBuffer;
55 class TimestampScaler;
56 struct AccelerateFactory;
57 struct DtmfEvent;
58 struct ExpandFactory;
59 struct PreemptiveExpandFactory;
60 
61 class NetEqImpl : public webrtc::NetEq {
62  public:
63   enum class OutputType {
64     kNormalSpeech,
65     kPLC,
66     kCNG,
67     kPLCCNG,
68     kVadPassive
69   };
70 
71   enum ErrorCodes {
72     kNoError = 0,
73     kOtherError,
74     kUnknownRtpPayloadType,
75     kDecoderNotFound,
76     kInvalidPointer,
77     kAccelerateError,
78     kPreemptiveExpandError,
79     kComfortNoiseErrorCode,
80     kDecoderErrorCode,
81     kOtherDecoderError,
82     kInvalidOperation,
83     kDtmfParsingError,
84     kDtmfInsertError,
85     kSampleUnderrun,
86     kDecodedTooMuch,
87     kRedundancySplitError,
88     kPacketBufferCorruption
89   };
90 
91   struct Dependencies {
92     // The constructor populates the Dependencies struct with the default
93     // implementations of the objects. They can all be replaced by the user
94     // before sending the struct to the NetEqImpl constructor. However, there
95     // are dependencies between some of the classes inside the struct, so
96     // swapping out one may make it necessary to re-create another one.
97     explicit Dependencies(
98         const NetEq::Config& config,
99         const rtc::scoped_refptr<AudioDecoderFactory>& decoder_factory);
100     ~Dependencies();
101 
102     std::unique_ptr<TickTimer> tick_timer;
103     std::unique_ptr<BufferLevelFilter> buffer_level_filter;
104     std::unique_ptr<DecoderDatabase> decoder_database;
105     std::unique_ptr<DelayPeakDetector> delay_peak_detector;
106     std::unique_ptr<DelayManager> delay_manager;
107     std::unique_ptr<DtmfBuffer> dtmf_buffer;
108     std::unique_ptr<DtmfToneGenerator> dtmf_tone_generator;
109     std::unique_ptr<PacketBuffer> packet_buffer;
110     std::unique_ptr<RedPayloadSplitter> red_payload_splitter;
111     std::unique_ptr<TimestampScaler> timestamp_scaler;
112     std::unique_ptr<AccelerateFactory> accelerate_factory;
113     std::unique_ptr<ExpandFactory> expand_factory;
114     std::unique_ptr<PreemptiveExpandFactory> preemptive_expand_factory;
115   };
116 
117   // Creates a new NetEqImpl object.
118   NetEqImpl(const NetEq::Config& config,
119             Dependencies&& deps,
120             bool create_components = true);
121 
122   ~NetEqImpl() override;
123 
124   // Inserts a new packet into NetEq. The |receive_timestamp| is an indication
125   // of the time when the packet was received, and should be measured with
126   // the same tick rate as the RTP timestamp of the current payload.
127   // Returns 0 on success, -1 on failure.
128   int InsertPacket(const RTPHeader& rtp_header,
129                    rtc::ArrayView<const uint8_t> payload,
130                    uint32_t receive_timestamp) override;
131 
132   void InsertEmptyPacket(const RTPHeader& rtp_header) override;
133 
134   int GetAudio(AudioFrame* audio_frame, bool* muted) override;
135 
136   void SetCodecs(const std::map<int, SdpAudioFormat>& codecs) override;
137 
138   int RegisterPayloadType(NetEqDecoder codec,
139                           const std::string& codec_name,
140                           uint8_t rtp_payload_type) override;
141 
142   int RegisterExternalDecoder(AudioDecoder* decoder,
143                               NetEqDecoder codec,
144                               const std::string& codec_name,
145                               uint8_t rtp_payload_type) override;
146 
147   bool RegisterPayloadType(int rtp_payload_type,
148                            const SdpAudioFormat& audio_format) override;
149 
150   // Removes |rtp_payload_type| from the codec database. Returns 0 on success,
151   // -1 on failure.
152   int RemovePayloadType(uint8_t rtp_payload_type) override;
153 
154   void RemoveAllPayloadTypes() override;
155 
156   bool SetMinimumDelay(int delay_ms) override;
157 
158   bool SetMaximumDelay(int delay_ms) override;
159 
160   int LeastRequiredDelayMs() const override;
161 
162   int SetTargetDelay() override;
163 
164   int TargetDelayMs() const override;
165 
166   int CurrentDelayMs() const override;
167 
168   int FilteredCurrentDelayMs() const override;
169 
170   // Sets the playout mode to |mode|.
171   // Deprecated.
172   // TODO(henrik.lundin) Delete.
173   void SetPlayoutMode(NetEqPlayoutMode mode) override;
174 
175   // Returns the current playout mode.
176   // Deprecated.
177   // TODO(henrik.lundin) Delete.
178   NetEqPlayoutMode PlayoutMode() const override;
179 
180   // Writes the current network statistics to |stats|. The statistics are reset
181   // after the call.
182   int NetworkStatistics(NetEqNetworkStatistics* stats) override;
183 
184   // Writes the current RTCP statistics to |stats|. The statistics are reset
185   // and a new report period is started with the call.
186   void GetRtcpStatistics(RtcpStatistics* stats) override;
187 
188   NetEqLifetimeStatistics GetLifetimeStatistics() const override;
189 
190   // Same as RtcpStatistics(), but does not reset anything.
191   void GetRtcpStatisticsNoReset(RtcpStatistics* stats) override;
192 
193   // Enables post-decode VAD. When enabled, GetAudio() will return
194   // kOutputVADPassive when the signal contains no speech.
195   void EnableVad() override;
196 
197   // Disables post-decode VAD.
198   void DisableVad() override;
199 
200   rtc::Optional<uint32_t> GetPlayoutTimestamp() const override;
201 
202   int last_output_sample_rate_hz() const override;
203 
204   rtc::Optional<CodecInst> GetDecoder(int payload_type) const override;
205 
206   rtc::Optional<SdpAudioFormat> GetDecoderFormat(
207       int payload_type) const override;
208 
209   int SetTargetNumberOfChannels() override;
210 
211   int SetTargetSampleRate() override;
212 
213   // Flushes both the packet buffer and the sync buffer.
214   void FlushBuffers() override;
215 
216   void PacketBufferStatistics(int* current_num_packets,
217                               int* max_num_packets) const override;
218 
219   void EnableNack(size_t max_nack_list_size) override;
220 
221   void DisableNack() override;
222 
223   std::vector<uint16_t> GetNackList(int64_t round_trip_time_ms) const override;
224 
225   std::vector<uint32_t> LastDecodedTimestamps() const override;
226 
227   int SyncBufferSizeMs() const override;
228 
229   // This accessor method is only intended for testing purposes.
230   const SyncBuffer* sync_buffer_for_test() const;
231   Operations last_operation_for_test() const;
232 
233  protected:
234   static const int kOutputSizeMs = 10;
235   static const size_t kMaxFrameSize = 5760;  // 120 ms @ 48 kHz.
236   // TODO(hlundin): Provide a better value for kSyncBufferSize.
237   // Current value is kMaxFrameSize + 60 ms * 48 kHz, which is enough for
238   // calculating correlations of current frame against history.
239   static const size_t kSyncBufferSize = kMaxFrameSize + 60 * 48;
240 
241   // Inserts a new packet into NetEq. This is used by the InsertPacket method
242   // above. Returns 0 on success, otherwise an error code.
243   // TODO(hlundin): Merge this with InsertPacket above?
244   int InsertPacketInternal(const RTPHeader& rtp_header,
245                            rtc::ArrayView<const uint8_t> payload,
246                            uint32_t receive_timestamp)
247       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
248 
249   // Delivers 10 ms of audio data. The data is written to |audio_frame|.
250   // Returns 0 on success, otherwise an error code.
251   int GetAudioInternal(AudioFrame* audio_frame, bool* muted)
252       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
253 
254   // Provides a decision to the GetAudioInternal method. The decision what to
255   // do is written to |operation|. Packets to decode are written to
256   // |packet_list|, and a DTMF event to play is written to |dtmf_event|. When
257   // DTMF should be played, |play_dtmf| is set to true by the method.
258   // Returns 0 on success, otherwise an error code.
259   int GetDecision(Operations* operation,
260                   PacketList* packet_list,
261                   DtmfEvent* dtmf_event,
262                   bool* play_dtmf) RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
263 
264   // Decodes the speech packets in |packet_list|, and writes the results to
265   // |decoded_buffer|, which is allocated to hold |decoded_buffer_length|
266   // elements. The length of the decoded data is written to |decoded_length|.
267   // The speech type -- speech or (codec-internal) comfort noise -- is written
268   // to |speech_type|. If |packet_list| contains any SID frames for RFC 3389
269   // comfort noise, those are not decoded.
270   int Decode(PacketList* packet_list,
271              Operations* operation,
272              int* decoded_length,
273              AudioDecoder::SpeechType* speech_type)
274       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
275 
276   // Sub-method to Decode(). Performs codec internal CNG.
277   int DecodeCng(AudioDecoder* decoder,
278                 int* decoded_length,
279                 AudioDecoder::SpeechType* speech_type)
280       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
281 
282   // Sub-method to Decode(). Performs the actual decoding.
283   int DecodeLoop(PacketList* packet_list,
284                  const Operations& operation,
285                  AudioDecoder* decoder,
286                  int* decoded_length,
287                  AudioDecoder::SpeechType* speech_type)
288       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
289 
290   // Sub-method which calls the Normal class to perform the normal operation.
291   void DoNormal(const int16_t* decoded_buffer,
292                 size_t decoded_length,
293                 AudioDecoder::SpeechType speech_type,
294                 bool play_dtmf) RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
295 
296   // Sub-method which calls the Merge class to perform the merge operation.
297   void DoMerge(int16_t* decoded_buffer,
298                size_t decoded_length,
299                AudioDecoder::SpeechType speech_type,
300                bool play_dtmf) RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
301 
302   // Sub-method which calls the Expand class to perform the expand operation.
303   int DoExpand(bool play_dtmf) RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
304 
305   // Sub-method which calls the Accelerate class to perform the accelerate
306   // operation.
307   int DoAccelerate(int16_t* decoded_buffer,
308                    size_t decoded_length,
309                    AudioDecoder::SpeechType speech_type,
310                    bool play_dtmf,
311                    bool fast_accelerate)
312       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
313 
314   // Sub-method which calls the PreemptiveExpand class to perform the
315   // preemtive expand operation.
316   int DoPreemptiveExpand(int16_t* decoded_buffer,
317                          size_t decoded_length,
318                          AudioDecoder::SpeechType speech_type,
319                          bool play_dtmf)
320       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
321 
322   // Sub-method which calls the ComfortNoise class to generate RFC 3389 comfort
323   // noise. |packet_list| can either contain one SID frame to update the
324   // noise parameters, or no payload at all, in which case the previously
325   // received parameters are used.
326   int DoRfc3389Cng(PacketList* packet_list, bool play_dtmf)
327       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
328 
329   // Calls the audio decoder to generate codec-internal comfort noise when
330   // no packet was received.
331   void DoCodecInternalCng(const int16_t* decoded_buffer, size_t decoded_length)
332       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
333 
334   // Calls the DtmfToneGenerator class to generate DTMF tones.
335   int DoDtmf(const DtmfEvent& dtmf_event, bool* play_dtmf)
336       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
337 
338   // Produces packet-loss concealment using alternative methods. If the codec
339   // has an internal PLC, it is called to generate samples. Otherwise, the
340   // method performs zero-stuffing.
341   void DoAlternativePlc(bool increase_timestamp)
342       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
343 
344   // Overdub DTMF on top of |output|.
345   int DtmfOverdub(const DtmfEvent& dtmf_event,
346                   size_t num_channels,
347                   int16_t* output) const
348       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
349 
350   // Extracts packets from |packet_buffer_| to produce at least
351   // |required_samples| samples. The packets are inserted into |packet_list|.
352   // Returns the number of samples that the packets in the list will produce, or
353   // -1 in case of an error.
354   int ExtractPackets(size_t required_samples, PacketList* packet_list)
355       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
356 
357   // Resets various variables and objects to new values based on the sample rate
358   // |fs_hz| and |channels| number audio channels.
359   void SetSampleRateAndChannels(int fs_hz, size_t channels)
360       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
361 
362   // Returns the output type for the audio produced by the latest call to
363   // GetAudio().
364   OutputType LastOutputType() RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
365 
366   // Updates Expand and Merge.
367   virtual void UpdatePlcComponents(int fs_hz, size_t channels)
368       RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
369 
370   // Creates DecisionLogic object with the mode given by |playout_mode_|.
371   virtual void CreateDecisionLogic() RTC_EXCLUSIVE_LOCKS_REQUIRED(crit_sect_);
372 
373   rtc::CriticalSection crit_sect_;
374   const std::unique_ptr<TickTimer> tick_timer_ RTC_GUARDED_BY(crit_sect_);
375   const std::unique_ptr<BufferLevelFilter> buffer_level_filter_
376       RTC_GUARDED_BY(crit_sect_);
377   const std::unique_ptr<DecoderDatabase> decoder_database_
378       RTC_GUARDED_BY(crit_sect_);
379   const std::unique_ptr<DelayManager> delay_manager_ RTC_GUARDED_BY(crit_sect_);
380   const std::unique_ptr<DelayPeakDetector> delay_peak_detector_
381       RTC_GUARDED_BY(crit_sect_);
382   const std::unique_ptr<DtmfBuffer> dtmf_buffer_ RTC_GUARDED_BY(crit_sect_);
383   const std::unique_ptr<DtmfToneGenerator> dtmf_tone_generator_
384       RTC_GUARDED_BY(crit_sect_);
385   const std::unique_ptr<PacketBuffer> packet_buffer_ RTC_GUARDED_BY(crit_sect_);
386   const std::unique_ptr<RedPayloadSplitter> red_payload_splitter_
387       RTC_GUARDED_BY(crit_sect_);
388   const std::unique_ptr<TimestampScaler> timestamp_scaler_
389       RTC_GUARDED_BY(crit_sect_);
390   const std::unique_ptr<PostDecodeVad> vad_ RTC_GUARDED_BY(crit_sect_);
391   const std::unique_ptr<ExpandFactory> expand_factory_
392       RTC_GUARDED_BY(crit_sect_);
393   const std::unique_ptr<AccelerateFactory> accelerate_factory_
394       RTC_GUARDED_BY(crit_sect_);
395   const std::unique_ptr<PreemptiveExpandFactory> preemptive_expand_factory_
396       RTC_GUARDED_BY(crit_sect_);
397 
398   std::unique_ptr<BackgroundNoise> background_noise_ RTC_GUARDED_BY(crit_sect_);
399   std::unique_ptr<DecisionLogic> decision_logic_ RTC_GUARDED_BY(crit_sect_);
400   std::unique_ptr<AudioMultiVector> algorithm_buffer_
401       RTC_GUARDED_BY(crit_sect_);
402   std::unique_ptr<SyncBuffer> sync_buffer_ RTC_GUARDED_BY(crit_sect_);
403   std::unique_ptr<Expand> expand_ RTC_GUARDED_BY(crit_sect_);
404   std::unique_ptr<Normal> normal_ RTC_GUARDED_BY(crit_sect_);
405   std::unique_ptr<Merge> merge_ RTC_GUARDED_BY(crit_sect_);
406   std::unique_ptr<Accelerate> accelerate_ RTC_GUARDED_BY(crit_sect_);
407   std::unique_ptr<PreemptiveExpand> preemptive_expand_
408       RTC_GUARDED_BY(crit_sect_);
409   RandomVector random_vector_ RTC_GUARDED_BY(crit_sect_);
410   std::unique_ptr<ComfortNoise> comfort_noise_ RTC_GUARDED_BY(crit_sect_);
411   Rtcp rtcp_ RTC_GUARDED_BY(crit_sect_);
412   StatisticsCalculator stats_ RTC_GUARDED_BY(crit_sect_);
413   int fs_hz_ RTC_GUARDED_BY(crit_sect_);
414   int fs_mult_ RTC_GUARDED_BY(crit_sect_);
415   std::atomic<int> last_output_sample_rate_hz_;
416   size_t output_size_samples_ RTC_GUARDED_BY(crit_sect_);
417   size_t decoder_frame_length_ RTC_GUARDED_BY(crit_sect_);
418   Modes last_mode_ RTC_GUARDED_BY(crit_sect_);
419   Operations last_operation_ RTC_GUARDED_BY(crit_sect_);
420   std::unique_ptr<int16_t[]> mute_factor_array_ RTC_GUARDED_BY(crit_sect_);
421   size_t decoded_buffer_length_ RTC_GUARDED_BY(crit_sect_);
422   std::unique_ptr<int16_t[]> decoded_buffer_ RTC_GUARDED_BY(crit_sect_);
423   uint32_t playout_timestamp_ RTC_GUARDED_BY(crit_sect_);
424   bool new_codec_ RTC_GUARDED_BY(crit_sect_);
425   uint32_t timestamp_ RTC_GUARDED_BY(crit_sect_);
426   bool reset_decoder_ RTC_GUARDED_BY(crit_sect_);
427   rtc::Optional<uint8_t> current_rtp_payload_type_ RTC_GUARDED_BY(crit_sect_);
428   rtc::Optional<uint8_t> current_cng_rtp_payload_type_
429       RTC_GUARDED_BY(crit_sect_);
430   uint32_t ssrc_ RTC_GUARDED_BY(crit_sect_);
431   bool first_packet_ RTC_GUARDED_BY(crit_sect_);
432   const BackgroundNoiseMode background_noise_mode_ RTC_GUARDED_BY(crit_sect_);
433   NetEqPlayoutMode playout_mode_ RTC_GUARDED_BY(crit_sect_);
434   bool enable_fast_accelerate_ RTC_GUARDED_BY(crit_sect_);
435   std::unique_ptr<NackTracker> nack_ RTC_GUARDED_BY(crit_sect_);
436   bool nack_enabled_ RTC_GUARDED_BY(crit_sect_);
437   const bool enable_muted_state_ RTC_GUARDED_BY(crit_sect_);
438   AudioFrame::VADActivity last_vad_activity_ RTC_GUARDED_BY(crit_sect_) =
439       AudioFrame::kVadPassive;
440   std::unique_ptr<TickTimer::Stopwatch> generated_noise_stopwatch_
441       RTC_GUARDED_BY(crit_sect_);
442   std::vector<uint32_t> last_decoded_timestamps_ RTC_GUARDED_BY(crit_sect_);
443   const bool use_dtx_delay_fix_ RTC_GUARDED_BY(crit_sect_);
444 
445  private:
446   RTC_DISALLOW_COPY_AND_ASSIGN(NetEqImpl);
447 };
448 
449 }  // namespace webrtc
450 #endif  // MODULES_AUDIO_CODING_NETEQ_NETEQ_IMPL_H_
451