If you’re trying to sync audio with video, for example because you recorded them on different devices, you acquired them via different sources (that don’t share a timebase), or you’re playing them back through devices with different latencies, you traditionally need some kind of interactive successive approximation where you twist some kind of knob or something until you get it right.
A difficulty with this is that only occasionally are there events in the video that you can associate with high precision with the audio. So you have to wait for one of them to happen once for each approximation.
The ideal solution is probably to find a short segment of video with such an event and repeat it at as high a frequency as possible as you adjust the audio–video lag. Around 20 Hz, rhythms — pulse trains — fade into continuous tones. So you need something a little slower than that, which possibly limits you to about 50 ms feedback latency on the knob. Also, you probably need several frames of video to successfully interpolate movement, and video is commonly interpolated at rates as low as 24 Hz, so you might not be able to do better than, say, 6 Hz or 4 Hz for the repetition.
However, at these rates, the impulsive event you’re looking to synchronize the sound with — ideally a clapboard, but often in practice a spoken plosive or something — may be overwhelmed by the much larger impulsive event of jumping back in time by 167 ms or 250 ms. A possible solution is to play both the video and the audio segment boustrophedonically: first forward, then backward, then forward again, and so on. This eliminates the first-order discontinuities, though the remaining second-order discontinuities — objects instantly reversing their direction of motion — may be disturbing.
The knob should probably directly control the video displacement, not the audio displacement, to avoid screwing with the pitch.