Use SSML <phoneme> tags in your input text. ElevenLabs supports CMU Arpabet and IPA. Example (CMU): <phoneme alphabet="cmu-arpabet" ph="M AE1 D IH0 S AH0 N">Madison</phoneme>. Phoneme tags currently work only on the models: Eleven English v1, Eleven Turbo v2, and Eleven Flash v2. (elevenlabs.io)
How to do it (API or Studio)
Pick a supported model
Use model_id eleven_monolingual_v1 (English v1), eleven_turbo_v2, or eleven_flash_v2. Other models will ignore phoneme tags. (elevenlabs.io)
Wrap each target word with a phoneme tag
Tags apply to single words only; create one tag per word. Include correct stress markers for multisyllabic words (e.g., AE1). (elevenlabs.io)
Send your text (with phoneme tags) to the Text-to-Speech endpoint
POST /v1/text-to-speech/:voice_id with your text containing the <phoneme> tag and model_id set to one of the supported models. Example field usage is shown in the API reference; phoneme tags are accepted by the API on these models. (elevenlabs.io)
Prefer CMU Arpabet for consistency
Both IPA and CMU work, but CMU is generally more predictable today. (help.elevenlabs.io)
Fallbacks if your chosen model doesn’t support phoneme tags
Use alias rules or a pronunciation dictionary (.PLS or TXT) to substitute pronunciations globally across a project. (elevenlabs.io)
References
Controls (Prompting best practices: phoneme tags, model support, examples, single-word scope, stress) (elevenlabs.io)
Help Center: Do pauses and SSML phoneme tags work with the API? (API support + supported models) (help.elevenlabs.io)
Help Center: Force a certain pronunciation (IPA/CMU support; CMU recommended; supported models) (help.elevenlabs.io)
API reference: Create speech (endpoint to send text with tags and choose model) (elevenlabs.io)
pine marsh
stuck flume