The voice-match score
A party score for rhythm and timing — never for how your voice sounds. A ten-year-old dubbing Arnold is meant to be able to score 100.
Every dub gets a number out of 100, and every line inside it gets its own. It's worth knowing what that number is actually measuring, because it is narrower than it looks — and honest about being narrow.
What it compares
For each line, the scorer takes your recording and the original actor's delivery of that same line and compares how their loudness moves over time. Not pitch, not tone, not the words — just the rise and fall of energy, in slices of forty milliseconds. Each side is normalised to itself first, so a whisper and a shout are judged on shape, not volume. If your take is essentially silent, it scores zero.
The four things it looks at
Shape. How closely the contour of your loudness follows theirs. Before measuring, the scorer slides your take back and forth (up to about a quarter of a second either way) and tries a few tempos (from 20% slower to 25% faster) to find the best fit — so being a little late, or a little quick, doesn't cost you the whole line. This is the single biggest component.
Timing. Of the moments where the original actor is speaking, how often are you speaking too? And of the moments where they're quiet, how often are you quiet? Both matter, and they're combined so that neither can carry the other: fill every pause with sound and you lose half of it. Timing doesn't add to your score so much as gate it — a badly timed take is capped low however good its shape.
Fill. How much you talk compared to how much they did. Talking twice as much costs the same as talking half as much.
Coverage. Whether you got through the line. A take that stops a third of the way in is scored against the whole line, not the third you did — so a beautifully read first ten seconds of a minute-long speech doesn't score as a finished performance.
How they combine
Shape and fill are blended into a performance figure (shape counts for most of it). That's multiplied by timing — with a floor, so an honest take that's merely mistimed doesn't collapse to zero — and by coverage. The result is then curved upward a little, which lifts the middle of the scale so that ordinary, decent dubs read as a comfortable pass rather than a fail.
The grades
- 90 and up — Uncanny double.
- 80 — Certified voice twin.
- 65 — Nailed the rhythm.
- 50 — Solid dub.
- 35 — Creative liberties.
- Below that — Off-script legend.
Why everyone gets about 60
Here's the honest part. Across every dub ever posted to the gallery, the average score is about 60, and nine in ten land between 45 and 74. Nothing has ever scored 80. The two top grades are, in practice, out of reach — they exist as a ceiling, not a target.
That's because of what the score measures. It rewards speech-shaped sound arriving at about the right times, and it's good at that: the original audio scores exactly 100 against itself, and white noise scores in the twenties. But it barely tells a right performance from a wrong one — reading the correct line and reading a different line of the same length score within a few points of each other. Short lines score higher than long ones almost regardless of who's reading them. Which line you performed moves the number by about six points; how long the line was moves it by about fifteen.
So treat it the way it's meant: a rhythm game's score, fun to chase on retakes of the same scene, and not a verdict on your acting. The gallery agrees — how a dub is voted has essentially no relationship to its score.
What it deliberately ignores
The score never listens to pitch or timbre. That's a choice: a deep voice should be able to match a child's line, and an impression should be judged by the people watching it, not by a formula. The one thing that would let the score tell a real performance from a well-timed mumble is the melody of the line — the way it rises and falls in pitch — and that's the direction it may grow. For now, the number is about time, and it says so.
Line by line
Tap the score on any premiere to open every line's own number, and the match: the two loudness curves the scorer compared, drawn on top of each other with your take already slid and stretched to where the alignment put it. Where they agree the line glows white. It's the scorer's own view of your take — if something looks wrong there, it's a scoring question, and you can see exactly which line it is.
In a Roulette or Private Room dub, each player's total is the average over the lines they recorded, and the dub's own badge is the average across every recorded line, weighted so a player who took fifteen lines counts more than one who took one.