Design a Karaoke Streaming App

Problem statement

Design a karaoke streaming platform that delivers instrumental audio tracks, time-synchronized lyrics with word-level highlighting, optional real-time pitch detection and scoring, recording and playback of user performances, and social sharing—all while respecting music licensing and DRM constraints.

This is not a simple music streaming problem. A karaoke session requires three tightly coupled real-time streams: (1) the instrumental audio at broadcast quality, (2) lyric text that highlights in precise synchronization with the audio timeline, and (3) optionally, a microphone input pipeline that performs pitch detection against a reference melody to produce a live score. The synchronization tolerance between audio playback and lyric highlight is under 50 milliseconds—any perceptible drift destroys the user experience.

Why the problem is distinctive

A music streaming backend can buffer and retry. A karaoke session cannot tolerate drift between audio and lyrics because the user is actively singing along. The design therefore separates media delivery from synchronization control. Media delivery is an eventually-consistent CDN problem: fetch audio chunks, buffer, adapt bitrate. Synchronization control is a real-time local problem: the client maintains a single authoritative clock, aligns lyric events to audio sample positions, and compensates for network jitter without cloud round-trips.

The scoring subsystem adds a third constraint: microphone audio must be captured, analyzed for fundamental frequency, compared against a reference pitch track, and scored—all within 100-200ms to feel responsive. This processing must happen on-device because round-trip latency to a server makes real-time feedback impossible.

Public operating baseline versus design assumptions

Public evidence establishes that the category is operationally real. Smule reports over 100 million downloads and approximately 70 million monthly active users at peak, with users creating over 5 billion performances. StarMaker reports 100 million downloads across 200 countries. KaraFun maintains a catalog of over 50,000 licensed karaoke tracks. Apple Music Sing launched in December 2022 as a built-in karaoke feature across Apple devices.

For capacity planning, this answer explicitly assumes a mature platform with 50 million registered users, 8 million monthly active users, 500,000 concurrent karaoke sessions at peak, and 2 million song plays per day. Unless a number is tied to a citation, it is a stated design assumption, target, budget, or illustrative threshold.

The four architectural planes

1. Media delivery plane: CDN distribution, adaptive bitrate streaming, audio encoding, DRM enforcement, and offline caching.
2. Synchronization plane: lyric timeline management, word-level highlighting, client-side clock discipline, and drift compensation.
3. Interaction plane: microphone capture, pitch detection, scoring algorithm, recording, and performance playback.
4. Platform plane: catalog management, licensing, user profiles, social features, recommendations, and analytics.

A strong interview answer keeps these planes separate. It allows the platform plane to degrade without affecting active sessions, and it permits the scoring plane to be optional without breaking the core sing-along experience.