Start with one approved script per language
Decide which languages your audience needs and prepare a separate script for each. A literal translation can change sentence length and timing. Have a fluent reader check names, terminology, and tone before generating audio.
Language selection is not a promise of automatic translation. Supply the intended language in the script itself. Keep product names and abbreviations consistent across versions, and note words that may need a different spelling for clear pronunciation.
Choose a compatible voice
Open Text to Speech and inspect the available voices. Choose a voice that supports the language of your prepared text. Voices and models can expose different controls, so do not assume one voice's supported languages or expressive options apply to another.
Start with a short sentence containing the most difficult name or phrase. Listen to it with a fluent speaker when possible. Check the account allowance before generating additional versions.
Keep versions easy to compare
Store each script with a clear language label. For example, keep an English introduction separate from a Chinese introduction rather than treating one audio file as both versions. This is a workflow suggestion, not an automatic translation feature.
For video projects, compare the length of each narration with its matching scene. Revise the text or adjust the scene in your video editor when necessary. Do not assume every translated sentence will fit the timing of the original.
Review each download
Listen to the downloaded narration, including the opening and ending. Check pronunciation, meaning, and pauses. Keep the approved text with its audio file so future corrections can be traced to the right version.
Read the text-to-speech workflow, video narration guide, or Getting Started. Existing voice design and voice cloning tools have their own requirements; use only recordings you have permission to use.
Common questions
Can every voice speak every language?
Check the capabilities shown for the selected voice. Language and model support can vary.
Who should check the translated narration?
A fluent reviewer should check both the script and the generated speech. A successful generation does not establish translation accuracy.