The Opus RFC seems to recommend a frame size of 20ms for most applications. For wideband speech, the sweet spot range is recommended to be 16-20kbps. 20ms frames => 50 frames per second. For a VoIP application, the header overhead per frame (IP+UDP+RTP+SRTP) is 44bytes => 17.6kbps at 50 frames per second. So a 20ms frame size seems to cause a roughly 100% overhead of header data. Therefore, higher frame sizes seem to make more sense to reduce this overhead. Is there any significant downside in using a higher frame size for VoIP applications? Secondly, the sweet spot is mentioned as 16-20kbps for wideband speech "for a 20ms frame". Assuming this is just the codec bitrate (without any headers), does the frame size really influence the Opus bitrate? Thanks, Manpreet. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.xiph.org/pipermail/opus/attachments/20141104/286c0cc7/attachment.htm
Manpreet Singh wrote:> this overhead. Is there any significant downside in using a higher frame > size for VoIP applications?The only real downside is latency. If you're targeting a latency of 150 ms and your network stack, jitter buffer, etc., add 2-3 packets worth of latency, then 60 ms frames put you in danger of blowing your latency budget even if your network is infinitely fast.> Secondly, the sweet spot is mentioned as 16-20kbps for wideband speech > "for a 20ms frame". Assuming this is just the codec bitrate (without any > headers), does the frame size really influence the Opus bitrate?Frame sizes larger than 20 ms will have similar quality at similar bitrates. If you want smaller frames, you will have to increase the bitrate to maintain the same quality.