Localising a training deck into five languages without five recording sessions

The reason most training libraries exist in one language is not that teams do not want translations. It is that with recorded narration, each language is a full production: find a speaker, brief them, record, edit, review, publish. Multiply by five and it becomes a project nobody sponsors.

Generated narration changes the shape of that, but not in the way people expect — and there is one part that trips up every first attempt.

What actually needs translating

The script. This is the narration, and it is text. Translate it, and the audio follows.

The slide content. Here is the part everyone forgets. If the conversion renders each slide as an image — and any fidelity-preserving conversion does — then the words on the slide are pixels. A Spanish audio track over English slides is not a Spanish video. It is an English video with Spanish voiceover, which is useful for some audiences and useless for others.

So a genuine localisation needs one translated deck per language, not just a translated script. That is the real workload, and it is worth knowing before you promise five languages by Friday.

Two levels of localisation, pick deliberately

Voiceover only. Original-language slides, localised narration. Fast, cheap, and appropriate when the audience is comfortable reading the source language but prefers to hear their own — common in multinational engineering and finance teams where the working language is English.

Full localisation. Translated deck plus translated script. This is what “Spanish version” usually means when someone asks for it, and it is the right target when the video is mandatory training or the audience is not expected to read the source language.

Sequencing a five-language rollout

1. Finalise the source deck first. Every language inherits the source. Translating a deck that is still being edited means re-translating, and re-translating means re-rendering.

2. Lock the script in the source language. A reviewed, final script is much easier to translate than a draft — and translation errors compound when the source is ambiguous.

3. Translate the deck, then the script. Do the deck first: the script has to describe what is now on the slide, and if the slide changed in translation, the script must follow it.

4. Check length drift. Most languages run longer than English — German and Spanish are noticeably longer, sometimes 20–30%. Slides that fit in English may overflow in translation. This is the most common defect in localised decks and it is invisible until you render.

5. Choose voices per language, once. Pick a voice per language and stick with it across the whole library. Inconsistent voices between modules read as different products.

6. Render in batches. Each language is a render job, and rendering does not parallelise for free. Five languages is five jobs; submit them with time to spare rather than the morning you need them.

Things that will go wrong

Number and date formats. A translated deck that still writes 10/28/2026 or uses comma decimal separators inconsistently is immediately jarring.

Proper nouns that should not be translated. Product names, feature names and team names usually stay in the source language. Put this in the brief for whoever translates, or you will get “Human Resources” translated and product names half-translated.

Pronunciation of the untranslated terms. The terms you left in English are exactly the ones text-to-speech will mispronounce in a non-English voice. Check these by language — a name that sounds fine in English TTS can be mangled in a Spanish one.

What this buys you

The difference is not that translation becomes free — it does not, someone still has to translate the deck and the script. The difference is that translation becomes the only cost. There is no studio, no speaker, no session, no re-recording when the source changes. A five-language library stops being a capital project and becomes a maintenance cycle, which is the only version of it most organisations actually sustain.