[time-nuts] an interesting timing problem

[Joseph Gwinn]

On Wed, 06 May 2020 12:00:02 -0400, time-nuts-request at lists.febo.com 
wrote:
Re: time-nuts Digest, Vol 190, Issue 10
> 
> ------------------------------
> 
> Message: 6
> Date: Wed, 6 May 2020 07:00:51 -0700
> From: jimlux <jimlux at earthlink.net>
> To: Discussion of precise time and frequency measurement
> 	<time-nuts at lists.febo.com>
> Subject: [time-nuts] an interesting timing problem
> Message-ID: <cfdf4164-1a45-a6e2-9538-4eb8b9f746b8 at earthlink.net>
> Content-Type: text/plain; charset=utf-8; format=flowed
> 
> Given that there's a lot more people spending time zooming, webexing, 
> teaming, skype, facetime, etc. these days, I'm curious if anyone has 
> figured out to *quantify* the issues of lag, desynchronization, etc.
> 
> How would one go about instrumenting it (without access to the source 
> code or servers involved)?
> 
> There's two areas of some interest to me:
> 1) there's several studies that say that when voice and image aren't 
> perfectly synchronized, particularly if it's not a consistent delay, or 
> if there are gaps and jumps, that it is more stressful and creates a 
> cognitive workload that does not exist with actual in-person meetings 
> (the "why am I more tired after a day of telework than the real thing")
> 
> 2) If you wanted to do group music playing or singing, relative timing 
> among the streams is critical.  Is there a threshold where it all breaks 
> down?  For instance, in an orchestra or choir, one has visual cues from 
> the conductor, but most people do not sing or play using the conductor 
> as a metronome triggering the next measure's notes. They also listen to 
> the players around them (or perhaps on the other side of the stage, some 
> 30-40 milliseconds late)
> 
> 
> I can think of ways to "test" a given teleconferencing system (blinking 
> LEDs in a pattern, tone bursts on audio), but I think there's some 
> challenges in things like compression algorithms (do they have constant 
> latency?) and highly structured test signals might not measure the same 
> as actual video and audio.
> 
> I will note that there are subjective difference among the various 
> tools, and there's differing effects from compression artifacts and 
> bandwidth/packet transport.

I recall digging into this some time ago for audio, in particular 
music, and found the answer in the Computer Music (meaning synthetic 
music) literature.  There was a founding book, but I don't recall the 
details, and things have probably evolved since then.

As I recall, if the time offset between instruments exceeded something 
like 20 milliseconds, the error became audible.

Joe Gwinn