Numbers: Choose the Eleven Multilingual v2 TTS model for your agent. It handles number normalization better than Flash v2.5 (recommended for phone numbers, dates, currencies). (elevenlabs.io)
Names: Attach a Pronunciation Dictionary to the agent (Voice tab). Use alias entries for any model; for phoneme-level control (IPA/CMU, English-only), use a model that supports phonemes (docs list Turbo v2 and Flash v2). (elevenlabs.io)
Practical setup:
In Agents Platform, set the TTS model to eleven_multilingual_v2 when number accuracy matters. (elevenlabs.io)
Add a Pronunciation Dictionary and define key names as alias or phoneme entries (phonemes: English-only; supported on Turbo v2 and Flash v2 per docs). (elevenlabs.io)
If you need low latency with better number reading on Flash v2.5, pre-normalize text with your LLM, or (on Enterprise) enable apply_text_normalization for v2.5 models. (elevenlabs.io)
For short/number-only prompts via API, set language_code so the normalizer applies the correct language rules. (help.elevenlabs.io)
Related references:
Using the Studio Pronunciations Editor (phoneme support note): Eleven Flash v2, Eleven Turbo v2, Eleven English v1. (help.elevenlabs.io)
upper basin
teal scaffold