/** @file * * Wireshark - Network traffic analyzer * By Gerald Combs * Copyright 1998 Gerald Combs * * SPDX-License-Identifier: GPL-2.0-or-later */ #ifndef RTPAUDIOSTREAM_H #define RTPAUDIOSTREAM_H #include "config.h" #ifdef QT_MULTIMEDIA_LIB #include #include #include #include #include #include #include #include #include #include #include #include #include class QAudioFormat; #if (QT_VERSION >= QT_VERSION_CHECK(6, 0, 0)) class QAudioSink; #else class QAudioOutput; #endif class QIODevice; class RtpAudioStream : public QObject { Q_OBJECT public: enum TimingMode { JitterBuffer, RtpTimestamp, Uninterrupted }; explicit RtpAudioStream(QObject *parent, rtpstream_id_t *id, bool stereo_required); ~RtpAudioStream(); bool isMatch(const rtpstream_id_t *id) const; bool isMatch(const struct _packet_info *pinfo, const struct _rtp_info *rtp_info) const; void addRtpPacket(const struct _packet_info *pinfo, const struct _rtp_info *rtp_info); void clearPackets(); void reset(double global_start_time); AudioRouting getAudioRouting(); void setAudioRouting(AudioRouting audio_routing); #if (QT_VERSION >= QT_VERSION_CHECK(6, 0, 0)) void decode(QAudioDevice out_device); #else void decode(QAudioDeviceInfo out_device); #endif double startRelTime() const { return start_rel_time_; } double stopRelTime() const { return stop_rel_time_; } unsigned sampleRate() const { return first_sample_rate_; } unsigned playRate() const { return audio_out_rate_; } void setRequestedPlayRate(unsigned new_rate) { audio_requested_out_rate_ = new_rate; } const QStringList payloadNames() const; /** * @brief Return a list of visual timestamps. * @return A set of timestamps suitable for passing to QCPGraph::setData. */ const QVector visualTimestamps(bool relative = true); /** * @brief Return a list of visual samples. There will be fewer visual samples * per second (1000) than the actual audio. * @param y_offset Y axis offset to be used for stacking graphs. * @return A set of values suitable for passing to QCPGraph::setData. */ const QVector visualSamples(int y_offset = 0); /** * @brief Return a list of out-of-sequence timestamps. * @return A set of timestamps suitable for passing to QCPGraph::setData. */ const QVector outOfSequenceTimestamps(bool relative = true); int outOfSequence() { return static_cast(out_of_seq_timestamps_.size()); } /** * @brief Return a list of out-of-sequence samples. Y value is constant. * @param y_offset Y axis offset to be used for stacking graphs. * @return A set of values suitable for passing to QCPGraph::setData. */ const QVector outOfSequenceSamples(int y_offset = 0); /** * @brief Return a list of jitter dropped timestamps. * @return A set of timestamps suitable for passing to QCPGraph::setData. */ const QVector jitterDroppedTimestamps(bool relative = true); int jitterDropped() { return static_cast(jitter_drop_timestamps_.size()); } /** * @brief Return a list of jitter dropped samples. Y value is constant. * @param y_offset Y axis offset to be used for stacking graphs. * @return A set of values suitable for passing to QCPGraph::setData. */ const QVector jitterDroppedSamples(int y_offset = 0); /** * @brief Return a list of wrong timestamps. * @return A set of timestamps suitable for passing to QCPGraph::setData. */ const QVector wrongTimestampTimestamps(bool relative = true); int wrongTimestamps() { return static_cast(wrong_timestamp_timestamps_.size()); } /** * @brief Return a list of wrong timestamp samples. Y value is constant. * @param y_offset Y axis offset to be used for stacking graphs. * @return A set of values suitable for passing to QCPGraph::setData. */ const QVector wrongTimestampSamples(int y_offset = 0); /** * @brief Return a list of inserted silence timestamps. * @return A set of timestamps suitable for passing to QCPGraph::setData. */ const QVector insertedSilenceTimestamps(bool relative = true); int insertedSilences() { return static_cast(silence_timestamps_.size()); } /** * @brief Return a list of wrong timestamp samples. Y value is constant. * @param y_offset Y axis offset to be used for stacking graphs. * @return A set of values suitable for passing to QCPGraph::setData. */ const QVector insertedSilenceSamples(int y_offset = 0); quint32 nearestPacket(double timestamp, bool is_relative = true); QRgb color() { return color_; } void setColor(QRgb color) { color_ = color; } QAudio::State outputState() const; void setJitterBufferSize(int jitter_buffer_size) { jitter_buffer_size_ = jitter_buffer_size; } void setTimingMode(TimingMode timing_mode) { timing_mode_ = timing_mode; } void setStartPlayTime(double start_play_time) { start_play_time_ = start_play_time; } #if (QT_VERSION >= QT_VERSION_CHECK(6, 0, 0)) bool prepareForPlay(QAudioDevice out_device); #else bool prepareForPlay(QAudioDeviceInfo out_device); #endif void startPlaying(); void pausePlaying(); void stopPlaying(); void seekPlaying(qint64 samples); void setStereoRequired(bool stereo_required) { stereo_required_ = stereo_required; } qint16 getMaxSampleValue() { return max_sample_val_; } void setMaxSampleValue(int16_t max_sample_val) { max_sample_val_used_ = max_sample_val; } void seekSample(qint64 samples); qint64 readSample(SAMPLE *sample); qint64 getLeadSilenceSamples() { return prepend_samples_; } qint64 getTotalSamples() { return (audio_file_->getTotalSamples()); } qint64 getEndOfSilenceSample() { return (audio_file_->getEndOfSilenceSample()); } double getEndOfSilenceTime() { return (double)getEndOfSilenceSample() / (double)playRate(); } qint64 convertTimeToSamples(double time) { return (qint64)(time * playRate()); } bool savePayload(QIODevice *file); unsigned getHash() { return rtpstream_id_to_hash(&(id_)); } rtpstream_id_t *getID() { return &(id_); } QString getIDAsQString(); rtpstream_info_t *getStreamInfo() { return &rtpstream_; } signals: void processedSecs(double secs); void playbackError(const QString error_msg); void finishedPlaying(RtpAudioStream *stream, QAudio::Error error); private: // Used to identify unique streams. // The GTK+ UI also uses the call number + current channel. rtpstream_id_t id_; rtpstream_info_t rtpstream_; bool first_packet_; QVectorrtp_packets_; RtpAudioFile *audio_file_; // Stores waveform samples in sparse file QIODevice *temp_file_; struct _GHashTable *decoders_hash_; double global_start_rel_time_; double start_abs_offset_; double start_rel_time_; double stop_rel_time_; qint64 prepend_samples_; // Count of silence samples at begin of the stream to align with other streams AudioRouting audio_routing_; bool stereo_required_; quint32 first_sample_rate_; quint32 audio_out_rate_; quint32 audio_requested_out_rate_; QSet payload_names_; struct SpeexResamplerState_ *visual_resampler_; QMap packet_timestamps_; QVector visual_samples_; QVector out_of_seq_timestamps_; QVector jitter_drop_timestamps_; QVector wrong_timestamp_timestamps_; QVector silence_timestamps_; qint16 max_sample_val_; qint16 max_sample_val_used_; QRgb color_; int jitter_buffer_size_; TimingMode timing_mode_; double start_play_time_; const QString formatDescription(const QAudioFormat & format); QString currentOutputDevice(); #if (QT_VERSION >= QT_VERSION_CHECK(6, 0, 0)) QAudioSink *audio_output_; void decodeAudio(QAudioDevice out_device); quint32 calculateAudioOutRate(QAudioDevice out_device, unsigned int sample_rate, unsigned int requested_out_rate); #else QAudioOutput *audio_output_; void decodeAudio(QAudioDeviceInfo out_device); quint32 calculateAudioOutRate(QAudioDeviceInfo out_device, unsigned int sample_rate, unsigned int requested_out_rate); #endif void decodeVisual(); SAMPLE *resizeBufferIfNeeded(SAMPLE *buff, int32_t *buff_bytes, qint64 requested_size); private slots: void outputStateChanged(QAudio::State new_state); void delayedStopStream(); }; #endif // QT_MULTIMEDIA_LIB #endif // RTPAUDIOSTREAM_H