The proposed method allows to quantify the audio-to-video alignement manually by giving the user the ability to vizualize the decoded frames and audio waveform side by side and place time makers on it. The delay is computed in both network domain (packet capture timestamps) and media domain (RTP timestamps).
- Setup a test pattern generator in physical space or in the network. The media should consist of identifiable video events synced with audio events like a clap or a dedicated pattern.
- Capture both audio and video simultaneously and analyze in LIST
- In
Compare streamspage:- select the capture file
- select either video or audio a
reference - select the other one as
main - name the analysis, exple
AV something - press
Compare
- Then interpret the results
| Results | Description |
|---|---|
| AV Delay (capture time) | This is diff between audio time marker and video time marker. Precision of +/-0.5 field/frame is due the unability to accuratly date natural a event captured by a camera sensor. |
| AV Delay (RTP time) | Same but calculated from RTP time markers |
- On frame selection, the video capture time marker corresponds to the capture timestamp of 1st the packet of the frame/field. This works perfectly for a pattern generator, but not 100% accurate for natural feed.
- On audio cursor moved, the audio capture time marker is computed from the the new cursor position on the waveform (relative time) offset by the capture timestamps of the 1st packet of audio stream.
- For both audio and video, RTP time marker = capture time marker - pkt_time_vs_rtp_time (from influxDB).