Numeric dubbing errors: worked examples
A flag is a place to listen, not a verdict.
The examples below are synthetic demonstrations of the current rules, not customer incidents or an accuracy benchmark. Each pair assumes nearby timestamps and the stated number formats. You can paste the text into the timestamped JSON example in the import guide.
A price gains a zero
Original: 29 dollars
Dub: 290 dollars
One review item. Confirm the price in both recordings before correcting the dub.
A small count changes
Original: Repeat 2 times.
Dub: Repeat 5 times.
One review item. Small counts and step numbers are included; intentional renumbering needs a human decision.
An extra quantity appears
Original: Add 120 grams.
Dub: Add 120 grams and 50 grams.
The 120-gram pair matches. The additional 50 grams creates a dubbed-side review item.
A valid conversion
Original: 150 pounds
Dub: 68 kilograms
No review item: these supported mass units agree within the rounding tolerance.
A unit changes the amount
Original: 120 grams
Dub: 120 kilograms
One review item. Equal written digits do not make these quantities equivalent.
A decimal convention changes
Original: 1.5 grams (English)
Dub: 1,5 gramos (Spanish)
No review item when the corresponding numeric formats are selected.
Cases a clean report can miss
Number words: “one hundred grams” and “two hundred grams” contain no written digits. Neither amount is inspected. Check the occurrence counts before interpreting an empty report.
Swapped roles: “120 grams in the morning, 240 at night” can match “240 grams in the morning, 120 at night” if both occurrences carry equivalent units and fall in the same timing window. The checker does not establish which quantity belongs to which action. Listen for meaning and context.
Unsupported units: matching digits beside units outside the supported list do not validate those units. Currency exchange is also not calculated: $120 and €120 are not treated as an equivalent conversion.
Why one mismatch may produce one card
A changed amount leaves an unmatched occurrence in each transcript. The tool groups nearby unmatched values into a single review item to avoid two cards for the same possible substitution. That grouping uses timing, not semantic understanding; inspect the displayed text and both track timestamps. A remaining dubbed quantity gets its own card.
Decide what actually needs changing
Replay both recordings. If the original and dubbed speech agree but one transcript is wrong, correct the transcript and rerun. If the speech differs, ask a reviewer who understands both languages whether it is a valid localization or a real error. Keep uncertain cases open. The two deliberate mistakes in the downloadable practice pair help you learn the interface before reviewing real work.