How to translate an audio guide

Lock the source before anything is translated

This is the whole game. Every change made to the source script after translation has started is paid for again in every language, and it is paid for twice if that language has already been voiced. A script that is still being tweaked by a curator while translation is in progress is the single most reliable way to overrun an audio guide budget.

Locking it means a named person signs it off and further changes go into a list for the next revision rather than into the file. That role is uncomfortable to assign and it saves more money than any other decision in the project.

Translate for the ear, not the page

Tell the translator the text will be spoken aloud, because it changes what they produce. Written translations favour longer sentences and subordinate clauses that are hard to say and harder to follow when you cannot re-read them. Spoken translation wants shorter units, plainer connectives, and a rhythm someone can perform.

Send them the audio of the source language if you have it. Hearing where the original pauses, emphasises, and slows down tells a translator things the transcript does not, and it costs nothing to include.

Length is the practical consequence. Some languages expand against English by ten to twenty percent, which turns a ninety second stop into nearly two minutes. If the pacing of a stop matters, say so, and let the translator condense rather than discovering it in the recording session.

• Tell the translator it will be spoken, not read

• Send the source audio alongside the script

• Supply a list of names with the pronunciation you want

• Flag any stop where the length is fixed

• Say which words should not be translated at all

Names are where guides get judged

Place names, personal names, Indigenous names, and object names are the words a visitor who speaks that language will notice first, and they are the words machine translation and synthetic narration both get wrong most often. A name that should have been left alone gets translated into a phrase; a name that should have been pronounced one way gets pronounced as if it were English.

Prepare a name list once, before any translation or recording, giving the spelling, whether it is translated, and how it is said. It takes an afternoon, it applies to every language, and it is the difference between a guide that sounds careful and one that sounds automated.

Reviewing a language nobody on staff reads

You need a second person for every language: one to translate and one to check. Without that, you are publishing text you cannot evaluate, and the first person to evaluate it will be a visitor.

Reviewers do not have to be professional linguists. A community organisation, a local language school, a tour operator who works with that market, or a staff member's family can all catch the errors that matter, which are usually names, tone, and something that reads as unintentionally rude rather than fine points of grammar.

Review the audio as well as the text. A correct translation can still be voiced with the wrong stress or a mispronounced name, and that is only audible once it has been recorded. Listening to the rendered audio before launch catches things no amount of proofreading will.

Voicing the translations

A human narrator per language is the best result and the one that scales worst, because each additional language is a new casting decision, a new session, and a new set of retakes. Synthetic narration is what makes a long language list affordable at all, and it has become good enough that most visitors do not comment on it.

The sensible split most venues land on is a human voice for the main language and anything emotionally weighted, and synthetic voices for the rest, with a native speaker checking the rendered audio of each. That keeps the languages a visitor is most likely to hear at the highest quality while letting the list be long.