[Kdenlive-devel] Feature request: Sync External Audio Automatically

[Brian Matherly]

>>  The ITU did a study and found that the threshold of detectability of lip 

> sync errors is about +45 ms to –125 ms (audio early to audio late) and that the 
> threshold of acceptability is about +90 ms to –185 ms. Apparently it is 
> generally more tolerable for the audio to be slightly delayed than for it to be 
> slightly early.
>> 
>> 
>>  If you only synchronize on video frame boundaries, then the worst case 
> scenario would be +20ms or -20ms for 25fps video (that is +/- 1/2 video frame). 
> The ITU research supports Dan's comment that most people can't even 
> detect that much error. That may be acceptable if your project will be consumed 
> directly. But if the output of your project is destined for further processing 
> (like being transcoded by another system, or being sent through a broadcast 
> chain), the down stream systems may add to the AV sync error. If the error 
> stacks up, it could exceed the thresholds of detectability or possibly even the 
> threshold of acceptability.
>> 
>>  So the amount of A/V error that is appropriate for your project depends on 
> what you plan on doing with it.
>> 
>>  ~BM
> 
> Thanks a lot for this information!
> Did they also test detectability of sync errors of hard sound effects? 
> Could it be lower there?

My understanding of the study is that they took a series of samples across many different types of clips (talking heads, hard sound effects, etc) and across many different people and they applied some statistical averages to the results. So it is possible that some types of sounds are more detectable than others - just as some people are more sensitive to errors than others.

> Maybe I'm a little too sensitive here, I hate watching videos in German 
> since they usually are badly synched (obviously ... different language) 
> and I often have to force myself to look somewhere else than to the 
> actors' faces in the first minutes.
> So we could start with frame accuracy first.

It stands to reason that you could start with frame accuracy, and then see if there is any demand for sub-frame accuracy.

~Brian