1 views collected around this technical thread.
The Tubi Android team replaced a dual‑player architecture with a single Media3/ExoPlayer instance, achieving smoother video‑ad transitions, reduced buffering, lower error rates, and improved business metrics, while also addressing decoder‑initialization failures, non‑key‑frame ad buffering, and subtitle timing issues.
ExoPlayer enables both MergingMediaSource‑based multi‑track merging and adaptive streaming (DASH, HLS, smoothing‑stream), letting developers choose between low‑cost CDN merging or bandwidth‑driven track switching, with detailed components, manifest parsing, bandwidth‑meter customization, and decoder reuse to ensure smooth bitrate changes.
The article explains how ExoPlayer can achieve low‑cost multi‑stream switching by merging multiple media sources and dynamically selecting tracks with DefaultTrackSelector, using decoder restarts to switch video and audio qualities, and compares this approach to adaptive DASH/HLS methods for practical Android development.
The article explains audio‑video synchronization fundamentals, user tolerance limits, four sync strategies, and then details ExoPlayer’s audio‑master architecture—two clocks, media‑clock selection, frame‑dropping logic, and how to inject custom clocks—while also addressing Android AudioTrack jitter and work‑arounds.
The NetEase News team improved video playback by analyzing the production and playback pipelines, prioritizing reading and buffering, and applying client‑side techniques such as preloading, fine‑grained caching and adaptive buffer control, selecting ExoPlayer for Android, optimizing file layout and CDN strategies, and using data‑driven measurements, ultimately reaching a 94% instant‑play rate with minimal stutter and failure.
The Android app’s HLS video produced an electric‑like noise because FFmpeg’s av_find_best_stream mistakenly selected the second audio track in a dual‑stream file, a track lacking 1‑8 kHz frequencies, whereas iOS and PC used the correct first track; fixing requires publishing videos with a single audio track and adding detection tools.