ElevenLabs Launches ElevenLabs Realtime, a Conversational Voice AI Model

At a glance

ElevenLabs has released ElevenLabs Realtime, a new speech-to-speech model for building voice agents. It comes in two versions, Mini for fast conversation and Max for complex tasks, and supports more than 90 languages, including Korean.

Key points

1. 1

   To build a voice agent (an AI program that talks with people), developers used to choose and combine separate parts such as speech recognition and turn-taking. Now they can choose a single model.

2. 2

   The Mini model suits fast conversation, such as AI assistants and educational tutors. The Max model suits customer support and multi-step work that involves finding information and using tools.

3. 3

   According to the company, the Max model scored 98.8% accuracy on a voice reasoning test and began responding 0.57 seconds after a question.

4. 4

   Developers already using the OpenAI Realtime API can try it by changing only the connection address and model name. However, WebRTC is not supported at launch.

Why it matters

Because it supports Korean, it makes it easier for Korean companies and developers to build voice customer service that talks with customers naturally.