Review numbers, dates and units in AI narration

Separate the value from its spoken form

Consider a product explanation containing 1.5 litres, 3/4 and a specific calendar date. Have the content owner approve the value and meaning before deciding its spoken form. A fraction and a date can share a written pattern while conveying different information, so clarify context. Natural delivery does not repair an incorrect source value. Conversely, a correct value still needs correction if listeners reasonably hear another quantity. This review has a different purpose from maintaining a proper-name pronunciation dictionary: it protects the meaning of numbers, dates and units throughout the release workflow.

Resolve ambiguity in the speech script

ElevenLabs documents spelling ambiguous numbers, dates and symbols out in the intended words. Supply context identifying the calendar, currency or type of quantity instead of relying on an isolated numeral. Decide how regional dates, separators, decimal marks, model identifiers and serial numbers should be read. Reconcile the expanded wording with the approved value so normalization does not introduce a semantic change. Even when using automatic processing, do not assume the same symbol has identical meaning in every language. Review the actual voice and passage rather than treating a normalization feature as proof of a correct numerical instruction.

Verify the actual normalization scope

ElevenLabs describes normalization separately for website Text to Speech, Studio and API generation. Do not present an API parameter as an assumed website control. Identify the actual production path and current setting, then rerun important quantities after a voice or model change. With Murf or WellSaid, another service’s notation should not be assumed portable. Preserve approved spelling alongside any speech-specific expansion and the method used. Feature labels are less important than an output that communicates the intended value. This record lets a later operator repeat the correction without guessing how the original narration transformed its numerical source.

Reconcile captions with the spoken value

A speech script may spell a date out while captions retain a concise official form. Reconcile both with the source value. A unit separated into the next caption or a screen changing too soon can obscure what the quantity describes even when the audio is correct. Perform a listening pass against the approved values and a separate muted visual review. For multiple languages, give each reviewer the intended value and context rather than copying another language’s written pattern. This checks numerical equivalence across audio and presentation without forcing all outputs to use an identical surface format.

Record the cost of the correction

Simulate a quantity changing from 1.5 to 1.8. Check generated characters and modifiers for ElevenCreative, affected-sub-block VGT for Murf, and downloaded or combined audio treatment for WellSaid. Do not assume every service meters only the final replacement sentence. Review the patch beside its neighbors and update captions, visuals and destinations. Earlier factual approval can reduce avoidable regeneration, but confirm the actual account rules before estimating the correction. Leave an unknown extra charge unresolved rather than treating it as zero. The useful work log connects the changed value with the exact production operations needed to release it.

Preserve a human-readable value record

Retain source value, intended meaning, spoken form, language, review date and released revision together. Keep a human-readable record even if a dictionary or settings export exists. Retrieve required audio and captions before changing access, then trial a passage containing important quantities with any replacement voice. Do not assume the previous reading is reproduced automatically. Ask another operator to verify the original value and make a correction using the retained package. This prevents the rendered video from becoming the only source when nobody can identify which quantity was approved or why its spoken form was chosen.

• Japanese and English narration: scripts, speech, captions and updates
• Tools for managing a narration pronunciation dictionary
• Automatic caption quality checks before delivery