Hello Eleven Labs Support Team,We have integrated your text-to-speech solution and are facing chal | ElevenLabs | Page 1

AI-generated Answer for aurelia_93589

Thanks for reaching out—here are the officially supported ways to control name/pronounciation in ElevenLabs, and a practical workflow that avoids building a full IPA dictionary from scratch.

What ElevenLabs offers to control pronunciation

SSML phoneme tags (IPA or CMU Arpabet) in your input text. Supported models for TTS are Eleven English v1, Eleven Flash v2, and Eleven Turbo v2. Example: <phoneme alphabet="cmu-arpabet" ph="M AE1 D IH0 S AH0 N">Madison</phoneme>. Note: phoneme tags apply to single words; for multi‑word names (first + last) you must tag each word. ElevenLabs recommends CMU for more predictable results. (elevenlabs.io)
Pronunciation Dictionaries (project‑level). Add rules as:
- phoneme entries (IPA/CMU) for exact pronunciations, or
- alias entries (string substitutions) when phonemes aren’t supported or you prefer “spelled‑as‑sounds.” Dictionaries can be created in Studio with the Pronunciations Editor, uploaded as .PLS files, or managed via API/SDK. They’re attached to a project and applied automatically during synthesis. (help.elevenlabs.io)
Model‑specific behavior:
- TTS/Studio: phoneme entries are honored by English v1, Flash v2, and Turbo v2; alias works with all models. (help.elevenlabs.io)
- Conversational AI agents: phoneme entries currently work with Turbo v2; alias works with all models. (elevenlabs.io)

Guidance on “Anglo‑accent” issue in cloned voices

Accent is determined by the voice, not the text. If a cloned voice was trained mainly on English/Anglo‑accent audio, it will tend to anglicize spellings—even “phonetic” ones. For best results, clone using source audio in the target language/accent, or select a library voice trained in that accent. (help.elevenlabs.io)
When using the API, set language_code (ISO 639‑1) so the normalizer applies the correct language rules for short/ambiguous inputs (helps reduce unintended English‑leaning normalization). (help.elevenlabs.io)

A lightweight, best‑practice workflow for proper names (no full phonetic dictionary required)

Pick model and strategy

If you’re on English v1 / Flash v2 / Turbo v2: prefer CMU phoneme tags for recurring names; they’re more consistent. Use alias only when convenient. (elevenlabs.io)
If you’re on a model that doesn’t support phonemes in your context (e.g., Multilingual v2 or Turbo v2.5 mentioned in prompting docs): use alias rules in a dictionary. (elevenlabs.io)

Build a minimal project dictionary

Start with your top 50–100 problematic names.
For each name, add:
- a phoneme entry (CMU preferred) where supported, and
- a fallback alias entry (e.g., Claughton → Cloffton) if needed. Remember case sensitivity—add both “Tomato” and “tomato” if applicable. (elevenlabs.io)
Create and test directly in Studio’s Pronunciations Editor; the dictionary auto‑attaches to the project and you can preview output immediately.

#Hello Eleven Labs Support Team,We have integrated your text-to-speech solution and are facing chal