Twilio Changelog | Sep. 09, 2026
ElevenLabs Voices for Say is now Public Beta
ElevenLabs voices are now available in Public Beta for <Say>, powered by the Flash v2 and Flash v2.5 models for fast, natural-sounding text-to-speech. Get started by specifying an ElevenLabs voice via the REST API or in a TwiML Bin.
Text-to-speech in <Say> now supports ElevenLabs voices, currently available as a Public Beta feature. This integration uses ElevenLabs' Flash v2 and Flash v2.5 models, which are optimized for low-latency generation without sacrificing voice quality, well-suited for real-time conversational applications.
To use an ElevenLabs voice, specify the voice parameter using the ElevenLabs provider prefix followed by the specific Voice ID (for example, voice="ElevenLabs.<Voice ID>"). You can configure this in two ways:
TwiML via Voice API: Set it in your backend code when generating voice calls (inbound or outbound) or when responding to webhooks using the <Say> verb.
TwiML Bins: Set it directly through TwiML Bin configuration in the Twilio Console.
Example:
<?xml version="1.0" encoding="UTF-8"?>
<Response>
<Say language="es-ES" voice="ElevenLabs.UgBBYS2sOqTuMpoF3BR0">Hola, ¿cómo te llamas?</Say>
</Response>
The system automatically selects the underlying model based on your language attribute, optimizing for the best balance of speed and quality:
English (en-US, en-GB, etc.): automatically uses the Flash 2.0 model.
Non-English (all other locales): automatically uses the Flash 2.5 model.
Note: custom model selection is not currently supported.
For more information, please refer to the docs: