ElevenLabs launches Speech Engine, a voice layer to integrate with an existing text-based agent.

ElevenLabs launches Speech Engine, a voice layer enabling developers to transform text-based conversational agents into voice agents via a single command.

Speech Engine allows transforming a written conversational agent into a voice agent without touching its architecture. The LLM, the document retrieval system, and the business logic remain intact: the voice component is layered on top of the existing stack. Installation is done via a single command, through a skill that configures everything needed to switch from text to voice in a single prompt.

The pipeline brings together transcription, speech synthesis, and audio orchestration in a single package, designed to work in concert. Speech recognition covers over 90 languages, synthesis over 70, with access to a library of over 11,000 voices and voice cloning. Dedicated models manage end-of-turn detection and in-sentence interruptions, without requiring additional logic to be coded.

For integration, a server SDK receives transcriptions, transmits them to the language model, and returns the response, while a client SDK launches a session from a browser or a mobile application. The latter is identical to ElevenAgents', making subsequent migration seamless. The platform supports SOC 2, HIPAA, and GDPR compliance, with data residency in Europe and no-retention mode available.