A clean studio master proves very little. These examples are picked for the opposite reason — surface noise, crowd noise, phone captures, mixes that bury the vocal. Every one shows the real output, timed to the letter.
Download the karaoke video, the audio and the raw timings for any of them.
We are finishing the reference set below. Each slot needs its audio, its rendered karaoke video and its raw timings before it goes live — we would rather show you nothing than a cherry-picked best case.
Try it on your own audioThe baseline. Sparse arrangement, vocal well forward — if alignment fails here, nothing else matters.
Josh Woodward · source · CC BY 4.0 Trimmed to a short excerpt for this page.
A fuller band pressing on the vocal — the case where generic speech models start to drift.
Josh Woodward · source · CC BY 4.0 Trimmed to a short excerpt for this page.
Surface noise, a century of wear and an acoustic recording horn. The hardest audio on this page.
Billy Murray · source · Public domain Trimmed to a short excerpt. Lyrics from the 1917 sheet music (also public domain).
The clean take above, put in a room: reverb tail and off-mic drift.
Josh Woodward · source · CC BY 4.0 Reverb and off-mic drift added by Tonael to simulate a live capture.
Same take again, downsampled and clipped. The worst signal we accept — and the point of the comparison.
Josh Woodward · source · CC BY 4.0 Band-limited and downsampled to 8 kHz by Tonael to simulate a phone recording.
Fast phrasing at 170 bpm with little melodic anchor — where per-letter timing has the least room for error.
Josh Woodward · source · CC BY 4.0 Trimmed to a short excerpt for this page.
Ours is picked to be hard, but yours is the one you care about. New accounts get 10 free minutes — no card required.
Start free