Design Audio Streaming & Playback

The prompt looks like the Netflix or YouTube streaming question, and that resemblance is the trap. Reach for per-title transcode ladders and 4K adaptive bitrate and you have built the wrong machine. A minute of Spotify audio at its top tier is roughly two and a half megabytes. A minute of 4K video can run hundreds of megabits per second, so the per-second byte cost differs by about a hundred to one. When each item is that small, encoding stops being the expensive part. The expensive parts are breadth and latency. The catalog runs to a hundred million tracks plus several million podcasts. Any of them might be the next thing a listener taps, and it has to start the instant they do.

So the first sentence out of your mouth should reset the framing. This is a delivery problem about catalog breadth and switch latency, plus offline sync and a play-logging pipeline, not a transcode problem about resolution ladders. Spotify settled this years ago. It encodes each track a handful of times into Ogg Vorbis at 96, 160, and 320 kilobits per second, then stores those files once. The real engineering goes into getting the right segment to the listener's ear with no audible gap. Everything downstream follows from naming that early.

The Spotify angle

The first sentence out of your mouth should reset the interviewer's frame from video to audio. Spotify interviewers build this system for a living, so a candidate who opens with transcode ladders and 4K renditions is reaching for a video diagram that does not fit. The grounding line: audio is roughly a hundredth of the bytes per second of video, so storage and encoding are cheap. The hard part is making any track in a hundred-million-track catalog start instantly and keep playing through network dips and track changes. Say that, then name Ogg Vorbis at 96, 160, and 320 kbps as the rendition set, and the answer shows you know the actual system rather than a generic CDN template.