Audio Video Synchronization in Content Stitching

Videos streaming over the internet often have independent secondary contents, such as advertisements, stitched into main content. Artifacts free playback requires normalizing audio and video (A/V) durations at stitching locations and overall A/V duration in main and secondary contents. In this paper, we analyze why A/V encoding practice often yields unaligned streams and how content stitching can cause A/V drifting over time. We present algorithm "Forever" which generates stitching friendly content that minimizes A/V misalignment. The algorithm, which proves that pattern template manifest in DASH standard is a degenerated case, has been successfully deployed on thousands of devices and media platforms in Prime Video and AWS media services, achieving artifacts free playback everywhere, on low-, medium- and high-end devices at scale.