Dub a 45-second ad into Spanish and German: which steps Sume covers

Sume covers five of the six dubbing steps for a voiceover-led ad: detach the audio, transcribe it, speak the translated script, render the new audio onto the footage and burn captions. The translation step is yours, and for a 45-second ad into Spanish and German the Sume jobs come to about 70 cents.

This recipe suits narration over footage. It does not move the mouth of an on-camera speaker, so a talking head will keep the lip movement it was filmed with.

The six steps and who does them

The receipt for two languages

Assume the 45-second English read is about 675 characters and each translation lands near 800 characters. That length is an assumption; translation changes it, so count your actual strings. Steps 1 and 2 run once. Steps 4 to 6 run once per language.

Where dubs go wrong

• Voice and language disagree. If the voice's primary language differs from your language field, Sume returns 409 tts_voice_language_mismatch before it creates a job or charges. Pick a Spanish voice for Spanish.
• The dub runs long. Translated text is rarely the same length as the source, so compare each language's audio duration with the 45-second slot. If it overruns, tighten the copy or raise speed toward the 1.5 ceiling.
• Captions burn the wrong wording. Send script_text with the translated script so the caption text is yours, and let speech-to-text supply only timings. Use a new caption job on the dubbed video, not a restyle of the English one.
• Hangul styles. If a target language is Korean, pick a Hangul caption style, because slam rejects Hangul with a 400.

Keep the sentence order

Request sentence segmentation on the English STT and keep the sentence count in the translation. When both languages have the same number of sentences, you can compare each dubbed sentence's length with its source slot, and re-render only the lines that overrun, at a cent or two each.

Microsoft's MAI-Voice-2.1 page lists 23 languages for both its Standard and Flash models (read 2026-10-05). The language list matters less than this loop, which is how you find out whether a dub in a given language fits your footage.

What this dub will not do

Be clear with your team before you start: there is no lip-sync step in this list, no voice that matches the original speaker unless you supply a voice you are licensed to use, and no judgement about whether the translation sells. A native speaker should read both scripts before the first render, and the 70 cents buys the machine steps, not the review.

Budget the retakes. A bad German line costs 1 to 4 cents to speak again, but a re-render and a re-caption cost 30 cents, so settle the text before the expensive steps and run TTS first, listen, then render.