Intel / article

Webinar Recap Give Your Text Chatbot A Voice That Sounds Human

Source: elevenlabs.io
Author: ElevenLabs editorial sitemap
Date: Published Oct 9, 2026

Context

An existing chatbot can gain a spoken interface while keeping its response logic. In ElevenLabs' workshop recap, the voice service transcribes speech, passes conversation history to the application's , and turns its streamed reply into audio. Supplying the earlier text conversation when voice starts lets the exchange continue from the same .

Speech synthesis can begin before the model finishes writing, reducing the wait for a reply. The separate external-agent integration guide explains how more complex applications can preserve their own task state.

Key takeaways · AI-distilled

ElevenLabs identifies turn-taking as a core voice problem: standard voice activity detection only detects silence, so agents interrupt and treat natural hesitations as finished sentences. They recommend a dedicated turn-taking model separate from the LLM.
Their suggested way to add voice to an existing chat is a dual WebSocket architecture, with the server passing full conversation history to its own LLM on each transcript and a session-start context update carrying over the earlier text conversation.
Voice favors faster models over deep reasoning ones, since reasoning pauses register as unnatural hesitation. The team also recommends WebRTC for audio because of built-in echo and noise cancellation, and detecting language from speech instead of asking users to pick.

Terms in this piece · Glossary

LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.

Read the source elevenlabs.io