Workflow: create one ElevenLabs voice profile per market, store seed audio and metadata, and call the API with voice_id + an explicit version tag to render ads. Use SSML (pauses, emphasis, prosody, lang tags) to match local pacing and legal reads—ElevenLabs supports SSML-style controls via their API. Versioning best practices: semantic names (voice_v1, voice_v1.1), keep original seeds/prompts, maintain a changelog and QA checklist, batch-render and A/B test final cuts across devices.

See ElevenLabs docs for API/SSML details.