- fix DTX decoding

src/engine/media/MT_AudioReceiver.cpp

@@ -28,8 +28,7 @@ using namespace MT;
 // ----------------- RtpBuffer::Packet --------------
 RtpBuffer::Packet::Packet(const std::shared_ptr<RTPPacket>& packet, std::chrono::milliseconds timelength, int samplerate)
     :mRtp(packet), mTimelength(timelength), mSamplerate(samplerate)
-{
+{}
-}
 
 std::shared_ptr<RTPPacket> RtpBuffer::Packet::rtp() const
 {
@@ -66,6 +65,7 @@ RtpBuffer::RtpBuffer(Statistics& stat)
 
 RtpBuffer::~RtpBuffer()
 {
+    if (mAddCounter)
         ICELogDebug(<< "Number of add packets: " << mAddCounter << ", number of retrieved packets " << mReturnedCounter);
 }
 
@@ -129,7 +129,7 @@ std::shared_ptr<RtpBuffer::Packet> RtpBuffer::add(const std::shared_ptr<jrtplib:
     mStat.mSsrc = static_cast<uint16_t>(packet->GetSSRC());
 
     // Update jitter
-    ICELogMedia(<< "Adding new packet into jitter buffer");
+    ICELogMedia(<< "Adding new packet seqno " << packet->GetSequenceNumber() << " into jitter buffer");
     mAddCounter++;
 
     // Look for maximum&minimal sequence number; check for dublicates
@@ -138,7 +138,7 @@ std::shared_ptr<RtpBuffer::Packet> RtpBuffer::add(const std::shared_ptr<jrtplib:
     // New sequence number
     unsigned newSeqno = packet->GetExtendedSequenceNumber();
 
-    for (std::shared_ptr<Packet>& p: mPacketList)
+    for (auto& p: mPacketList)
     {
         unsigned seqno = p->rtp()->GetExtendedSequenceNumber();
 
@@ -171,7 +171,7 @@ std::shared_ptr<RtpBuffer::Packet> RtpBuffer::add(const std::shared_ptr<jrtplib:
         available = findTimelength();
 
         if (available > mHigh)
-            ICELogMedia(<< "Available " << available << "ms with limit " << mHigh << "ms");
+            ICELogMedia(<< "Available " << available << " with limit " << mHigh);
 
         return p;
     }
@@ -346,16 +346,14 @@ int RtpBuffer::getNumberOfAddPackets() const
 //-------------- Receiver ---------------
 Receiver::Receiver(Statistics& stat)
     :mStat(stat)
-{
+{}
-}
 
 Receiver::~Receiver()
-{
+{}
-}
 
 //-------------- AudioReceiver ----------------
 AudioReceiver::AudioReceiver(const CodecList::Settings& settings, MT::Statistics &stat)
-    :Receiver(stat), mBuffer(stat), mDtmfBuffer(stat), mCodecSettings(settings), mCodecList(settings), mDtmfReceiver(stat)
+    :Receiver(stat), mRtpBuffer(stat), mDtmfBuffer(stat), mCodecSettings(settings), mCodecList(settings), mDtmfReceiver(stat)
 {
     // Init resamplers
     mResampler8.start(AUDIO_CHANNELS, 8000, AUDIO_SAMPLERATE);
@@ -367,12 +365,17 @@ AudioReceiver::AudioReceiver(const CodecList::Settings& settings, MT::Statistics
     mCodecList.setSettings(settings);
     mCodecList.fillCodecMap(mCodecMap);
 
-    mAvailable.setCapacity(AUDIO_SAMPLERATE * sizeof(short));
+    // 10 seconds is the maximum length of decoded audio in single step
+    // It is important - DTX may produce silence up to few seconds easily
+    mAvailable.setCapacity(AUDIO_SAMPLERATE * 10 * sizeof(short));
 
     mDtmfBuffer.setPrebuffer(0ms);
     mDtmfBuffer.setLow(0ms);
     mDtmfBuffer.setHigh(1ms);
 
+    // Avoid collecting too much data
+    mRtpBuffer.setHigh(240ms);
+
 #if defined(DUMP_DECODED)
     mDecodedDump = std::make_shared<Audio::WavFileWriter>();
     mDecodedDump->open("decoded.wav", 8000 /*G711*/, AUDIO_CHANNELS);
@@ -386,6 +389,11 @@ AudioReceiver::~AudioReceiver()
     mResampler32.stop();
     mResampler48.stop();
     mDecodedDump.reset();
+
+    if (mRequestedAudio != 0ms)
+        ICELogDebug(<< "Requested " << mRequestedAudio << ", produced " << mProducedAudio);
+    if (mDecodeCount)
+        ICELogDebug(<< "Average interval between packet decoding " << mIntervalBetweenDecode / mDecodeCount);
 }
 
 // Update codec settings
@@ -450,7 +458,7 @@ Codec* AudioReceiver::add(const std::shared_ptr<jrtplib::RTPPacket>& p)
         payloadLength = p->GetPayloadLength(),
         ptype = p->GetPayloadType();
 
-    ICELogMedia(<< "Adding packet No " << p->GetSequenceNumber());
+    // ICELogMedia(<< "Adding packet No " << p->GetSequenceNumber());
 
     // Increase codec counter
     mStat.mCodecCount[ptype]++;
@@ -508,12 +516,12 @@ Codec* AudioReceiver::add(const std::shared_ptr<jrtplib::RTPPacket>& p)
         {
             // It will cause statistics to report about bad RTP packet
             // I have to replay last packet payload here to avoid report about lost packet
-            mBuffer.add(p, std::chrono::milliseconds(time_length), samplerate);
+            mRtpBuffer.add(p, std::chrono::milliseconds(time_length), samplerate);
             return nullptr;
         }
 
         // Queue packet to buffer
-        mBuffer.add(p, std::chrono::milliseconds(time_length), samplerate).get();
+        mRtpBuffer.add(p, std::chrono::milliseconds(time_length), samplerate).get();
     }
     return codec;
 }
@@ -533,8 +541,12 @@ void AudioReceiver::processDecoded(Audio::DataWindow& output, DecodeOptions opti
 
 void AudioReceiver::produceSilence(std::chrono::milliseconds length, Audio::DataWindow& output, DecodeOptions options)
 {
+    if (!mCodec)
+        return;
+
     // Fill mDecodeBuffer as much as needed and call processDecoded()
     // Depending on used codec mono or stereo silence should be produced
+
     size_t chunks = length.count() / 10;
     size_t tail = length.count() % 10;
     size_t chunk_size = 10 * sizeof(int16_t) * mCodec->samplerate() / 1000 * mCodec->channels();
@@ -635,7 +647,8 @@ AudioReceiver::DecodeResult AudioReceiver::decodePacketTo(Audio::DataWindow& out
     auto& rtp = *packet->rtp(); // Syntax sugar
 
     mFailedCount = 0;
-    // Check if we need to emit silence or CNG - previously CNG packet was detected. Emit CNG audio here if needed.
+
+    // Check if we need to emit silence - it may happen in the case if next packet has RTP timestamp much beyond the previous one; maybe DTX was active.
     if (mLastPacketTimestamp && mLastPacketTimeLength && mCodec)
     {
          int units = rtp.GetTimestamp() - *mLastPacketTimestamp;
@@ -643,7 +656,8 @@ AudioReceiver::DecodeResult AudioReceiver::decodePacketTo(Audio::DataWindow& out
          if (milliseconds > mLastPacketTimeLength)
          {
              auto silenceLength = std::chrono::milliseconds(milliseconds - mLastPacketTimeLength);
-
+             ICELogDebug(<< "Emit " << silenceLength << " silence while requested " << options.mElapsed);
+             silenceLength = std::min(silenceLength, options.mElapsed);
              if (mCngPacket && options.mFillGapByCNG)
                  produceCNG(silenceLength, output, options);
              else
@@ -677,6 +691,7 @@ AudioReceiver::DecodeResult AudioReceiver::decodePacketTo(Audio::DataWindow& out
                 mDecodedLength = 0;
             else
             {
+                ICELogDebug(<< "Decoding CNG");
                 mCngPacket = packet;
                 mCngDecoder.decode3389(rtp.GetPayloadData(), rtp.GetPayloadLength());
 
@@ -775,7 +790,7 @@ AudioReceiver::DecodeResult AudioReceiver::decodeEmptyTo(Audio::DataWindow& outp
         else
         {
             // Emit silence if codec information is available - it is to properly handle the gaps
-            auto avail = output.getTimeLength(fmt.rate(), fmt.channels());
+            auto avail = output.getTimeLength(fmt);
             if (options.mElapsed > avail)
                 output.addZero(fmt.sizeFromTime(options.mElapsed - avail));
         }
@@ -785,86 +800,18 @@ AudioReceiver::DecodeResult AudioReceiver::decodeEmptyTo(Audio::DataWindow& outp
     return {.mStatus = DecodeResult::Status::Skip};
 }
 
-AudioReceiver::DecodeResult AudioReceiver::getAudioTo(Audio::DataWindow& output, DecodeOptions options)
+void MT::AudioReceiver::processDtmf()
+{
+    if (mDtmfBuffer.getCount())
     {
-    DecodeResult result = {.mStatus = DecodeResult::Status::Skip};
-
-    // Process RFC2833 here; it doesn't result in any audio - only callbacks and statistics
         auto fr = mDtmfBuffer.fetch();
         if (fr.mPacket && fr.mStatus == RtpBuffer::FetchResult::Status::RegularPacket)
             mDtmfReceiver.add(fr.mPacket->rtp());
-
-
-    auto produced = 0ms;
-    if (mAvailable.filled() && mCodec && options.mElapsed != 0ms)
-    {
-        Audio::Format fmt = options.mResampleToMainRate ? Audio::Format(AUDIO_SAMPLERATE, 1) : mCodec->getAudioFormat();
-        auto initiallyAvailable = mCodec ? mAvailable.getTimeLength(fmt.rate(), fmt.channels()) : 0ms;
-        if (initiallyAvailable != 0ms)
-        {
-            std::chrono::milliseconds resultTime = std::min(initiallyAvailable, options.mElapsed);
-            auto resultLen = fmt.sizeFromTime(resultTime);
-            mAvailable.moveTo(output, resultLen);
-            produced += resultTime;
-
-            // Maybe request is satisfied ?
-            if (produced >= options.mElapsed)
-                return {.mStatus = DecodeResult::Status::Ok, .mSamplerate = fmt.rate(), .mChannels = fmt.channels()};
     }
 }
 
-    std::chrono::milliseconds decoded = 0ms;
+void MT::AudioReceiver::updateDecodingTimeStatistics()
-    do
 {
-        // Get next packet from buffer
-        RtpBuffer::ResultList rl;
-        RtpBuffer::FetchResult fr = mBuffer.fetch();
-        // ICELogDebug(<< fr.toString() << " " << mBuffer.findTimelength());
-
-        switch (fr.mStatus)
-        {
-        case RtpBuffer::FetchResult::Status::Gap:           result = decodeGapTo(mAvailable, options);                  break;
-        case RtpBuffer::FetchResult::Status::NoPacket:      result = decodeEmptyTo(mAvailable, options);                break;
-        case RtpBuffer::FetchResult::Status::RegularPacket: result = decodePacketTo(mAvailable, options, fr.mPacket);   break;
-        default:
-            assert(0);
-        }
-
-        // Was there decoding at all ?
-        if (!mCodec)
-            break; // No sense to continue - we have no information at all
-
-        Audio::Format fmt = options.mResampleToMainRate ? Audio::Format(AUDIO_SAMPLERATE, 1) : mCodec->getAudioFormat();
-        result.mSamplerate = fmt.rate();
-        result.mChannels = fmt.channels();
-
-        // Have we anything interesting in the buffer ?
-        auto bufferAvailable = mAvailable.getTimeLength(fmt.rate(), fmt.channels());
-        if (bufferAvailable == 0ms)
-            break; // No sense to continue - decoding / CNG / PLC stopped totally
-
-        // How much data should be moved to result buffer ?
-        if (options.mElapsed != 0ms)
-        {
-            std::chrono::milliseconds resultTime = std::min(bufferAvailable, options.mElapsed - produced);
-            auto resultLen = fmt.sizeFromTime(resultTime);
-            mAvailable.moveTo(output, resultLen);
-            produced += resultTime;
-        }
-        else
-            mAvailable.moveTo(output, mAvailable.filled());
-
-        decoded += bufferAvailable;
-    }
-    while (produced < options.mElapsed);
-
-    if (produced != 0ms)
-        result.mStatus = DecodeResult::Status::Ok;
-
-    // Time statistics
-    if (result.mStatus == DecodeResult::Status::Ok)
-    {
-        // Decode statistics
     if (!mDecodeTimestamp)
         mDecodeTimestamp = std::chrono::steady_clock::now();
     else
@@ -874,6 +821,87 @@ AudioReceiver::DecodeResult AudioReceiver::getAudioTo(Audio::DataWindow& output,
         mDecodeTimestamp = t;
     }
 }
+
+AudioReceiver::DecodeResult AudioReceiver::getAudioTo(Audio::DataWindow& output, DecodeOptions options)
+{
+    // ICELogDebug(<< "getAudioTo() for " << options.mElapsed);
+    assert (options.mElapsed != 0ms);
+
+    // Increase counter of requested audio
+    mRequestedAudio += options.mElapsed;
+
+    DecodeResult result = {.mStatus = DecodeResult::Status::Skip};
+
+    // Process RFC2833 here; it doesn't result in any audio - only callbacks and statistics
+    processDtmf();
+
+    // How much time length audio we produced here
+    auto produced = 0ms;
+    Audio::Format fmt;
+
+    // Have we anything from the previous decode attempts ?
+    if (mAvailable.filled())
+    {
+        // Find what audio format is used in mAvailable data
+        fmt = options.mResampleToMainRate ? Audio::Format(AUDIO_SAMPLERATE, 1) : mCodec->getAudioFormat();
+
+        // How much milliseconds are available ?
+        auto availTime = mAvailable.getTimeLength(fmt);
+        if (availTime != 0ms)
+        {
+            // How much we can consume from the mAvailable buffer ?
+            std::chrono::milliseconds resultTime = std::min(availTime, options.mElapsed);
+
+            // Number of bytes
+            mAvailable.moveTo(output, fmt.sizeFromTime(resultTime));
+
+            // Increase the counter of produced milliseconds
+            produced += resultTime;
+        }
+    }
+
+    while (produced < options.mElapsed)
+    {
+        // Get next packet from buffer
+        RtpBuffer::FetchResult fr = mRtpBuffer.fetch();
+
+        // Decode to mAvailable buffer
+        switch (fr.mStatus)
+        {
+        case RtpBuffer::FetchResult::Status::Gap:           result = decodeGapTo(mAvailable, options.decreaseElapsedBy(produced));                                                      break;
+        case RtpBuffer::FetchResult::Status::NoPacket:      result = decodeEmptyTo(mAvailable, options.decreaseElapsedBy(produced));                                                    break;
+        case RtpBuffer::FetchResult::Status::RegularPacket: result = decodePacketTo(mAvailable, options.decreaseElapsedBy(produced), fr.mPacket);   updateDecodeIntervalStatistics();   break;
+        default:
+            assert(0);
+        }
+
+        // Was there decoding at all ?
+        if (!mCodec)
+            break; // No sense to continue - we have no information at all
+
+        fmt = options.mResampleToMainRate ? Audio::Format(AUDIO_SAMPLERATE, 1) : mCodec->getAudioFormat();
+        result.mSamplerate  = fmt.rate();
+        result.mChannels    = fmt.channels();
+
+        // How much milliseconds we have in audio buffer ?
+        auto bufferAvailable = mAvailable.getTimeLength(fmt);
+        if (bufferAvailable == 0ms)
+            break; // No sense to continue - decoding / CNG / PLC stopped totally
+
+        // How much data should be moved to result buffer ?
+        std::chrono::milliseconds resultTime = std::min(bufferAvailable, options.mElapsed - produced);
+        mAvailable.moveTo(output, fmt.sizeFromTime(resultTime));
+        produced += resultTime;
+    }
+
+    if (produced != 0ms)
+    {
+        result.mStatus = DecodeResult::Status::Ok;
+        updateDecodingTimeStatistics();
+    }
+
+    mProducedAudio += produced;
+    // ICELogDebug(<< "Requested " << options.mElapsed << ", produced " << produced << ", remains " << mAvailable.getTimeLength(fmt) << ", packets " << getRtpBuffer().getCount());
     return result;
 }
 
@@ -987,43 +1015,16 @@ AudioReceiver::MediaInfo AudioReceiver::infoFor(jrtplib::RTPPacket& p)
     return {packetTime, codec->samplerate()};
 }
 
-// int AudioReceiver::timelengthFor(jrtplib::RTPPacket& p)
+void AudioReceiver::updateDecodeIntervalStatistics()
-// {
+{
-//     CodecMap::iterator codecIter = mCodecMap.find(p.GetPayloadType());
+    auto now = std::chrono::steady_clock::now();
-//     if (codecIter == mCodecMap.end())
+    if (mLastDecodeTimestamp)
-//         return 0;
+    {
-
+        mIntervalBetweenDecode += std::chrono::duration_cast<std::chrono::microseconds>(now - *mLastDecodeTimestamp);
-//     PCodec codec = codecIter->second;
+        mDecodeCount ++;
-//     if (codec)
+    }
-//     {
+    mLastDecodeTimestamp = now;
-//         int frame_count = 0;
+}
-//         if (codec->rtpLength() != 0)
-//         {
-//             frame_count = static_cast<int>(p.GetPayloadLength() / codec->rtpLength());
-//             if (p.GetPayloadType() == 9/*G729A silence*/ && p.GetPayloadLength() % codec->rtpLength())
-//                 frame_count++;
-//         }
-//         else
-//             frame_count = 1;
-
-//         return frame_count * codec->frameTime();
-//     }
-//     else
-//         return 0;
-// }
-
-// int AudioReceiver::samplerateFor(jrtplib::RTPPacket& p)
-// {
-//     CodecMap::iterator codecIter = mCodecMap.find(p.GetPayloadType());
-//     if (codecIter != mCodecMap.end())
-//     {
-//         PCodec codec = codecIter->second;
-//         if (codec)
-//             return codec->samplerate();
-//     }
-
-//     return 8000;
-// }
 
 // ----------------------- DtmfReceiver -------------------
 DtmfReceiver::DtmfReceiver(Statistics& stat)

src/engine/media/MT_AudioReceiver.h

@@ -122,6 +122,7 @@ protected:
     std::optional<uint32_t> mLastSeqno;
     std::optional<jrtplib::RTPTime> mLastReceiveTime;
 
+
     // To calculate average interval between packet add. It is close to jitter but more useful in debugging.
     float mLastAddTime = 0.0f;
 };
@@ -169,10 +170,22 @@ public:
 
     struct DecodeOptions
     {
+        bool mRealtimeProcessing = false;               // Target PCAP parsing by default
         bool mResampleToMainRate = true;                // Resample all decoded audio to AUDIO_SAMPLERATE
         bool mFillGapByCNG = false;                     // Use CNG information if available
         bool mSkipDecode = false;                       // Don't do decode, just dry run - fetch packets, remove them from the jitter buffer
         std::chrono::milliseconds mElapsed = 0ms;       // How much milliseconds should be decoded; zero value means "decode just next packet from the buffer"
+        DecodeOptions decreaseElapsedBy(std::chrono::milliseconds delta)
+        {
+            return
+            {
+                .mRealtimeProcessing = mRealtimeProcessing,
+                .mResampleToMainRate = mResampleToMainRate,
+                .mFillGapByCNG = mFillGapByCNG,
+                .mSkipDecode = mSkipDecode,
+                .mElapsed = std::max(mElapsed - delta, 0ms)
+            };
+        }
     };
 
     struct DecodeResult
@@ -193,7 +206,7 @@ public:
 
     // Looks for codec by payload type
     Codec*      findCodec(int payloadType);
-    RtpBuffer&  getRtpBuffer() { return mBuffer; }
+    RtpBuffer&  getRtpBuffer() { return mRtpBuffer; }
 
     // Returns size of AudioReceiver's instance in bytes (including size of all data + codecs + etc.)
     int getSize() const;
@@ -205,14 +218,12 @@ public:
     };
     MediaInfo infoFor(jrtplib::RTPPacket& p);
 
-    // // Returns timelength for given packet
+    void processDtmf();
-    // int timelengthFor(jrtplib::RTPPacket& p);
 
-    // // Return samplerate for given packet
+    void updateDecodingTimeStatistics();
-    // int samplerateFor(jrtplib::RTPPacket& p);
 
 protected:
-    RtpBuffer                           mBuffer;                // Jitter buffer itself
+    RtpBuffer                           mRtpBuffer;             // RTP jitter buffer itself; here are audio packets
     RtpBuffer                           mDtmfBuffer;            // These two (mDtmfBuffer / mDtmfReceiver) are for our analyzer stack only; in normal softphone logic DTMF packets goes via SingleAudioStream::mDtmfReceiver
     DtmfReceiver                        mDtmfReceiver;
 
@@ -258,6 +269,9 @@ protected:
     float mIntervalSum = 0.0f;
     int mIntervalCount = 0;
 
+    std::chrono::milliseconds mRequestedAudio = 0ms;
+    std::chrono::milliseconds mProducedAudio = 0ms;
+
     // Zero rate will make audio mono but resampling will be skipped
     void makeMonoAndResample(int rate, int channels);
 
@@ -272,6 +286,12 @@ protected:
     DecodeResult decodeGapTo(Audio::DataWindow& output, DecodeOptions options);
     DecodeResult decodePacketTo(Audio::DataWindow& output, DecodeOptions options, const std::shared_ptr<RtpBuffer::Packet>& p);
     DecodeResult decodeEmptyTo(Audio::DataWindow& output, DecodeOptions options);
+
+    std::optional<std::chrono::steady_clock::time_point> mLastDecodeTimestamp;
+    std::chrono::microseconds mIntervalBetweenDecode = 0us;
+    size_t mDecodeCount = 0;
+    void updateDecodeIntervalStatistics();
+
 };
 
 }

src/engine/media/MT_CodecList.cpp

@@ -137,7 +137,6 @@ std::string CodecList::Settings::toString() const
         oss << "OPUS ptype: " << spec.mPayloadType << ", rate: " << spec.mRate << ", channels: " << spec.mChannels << std::endl;
     }
 
-
     return oss.str();
 }