Eleven v4 and Turbo: dialogue, emotion and the silence between lines

Not every conversation needs the same voice model

One voice tells a story; another needs to answer immediately on the phone. Both are voice work, but their needs differ. ElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28, 2026, in ElevenCreative, ElevenAgents and through the API. It claims better dialogue, emotion control and greater voice consistency when regenerating lines. The documentation lists more than 90 languages, including Polish.

V4 is for prepared material: narration, dialogue and longer recordings. Turbo is meant to respond quickly, for example as a voice agent — an application conversing with a user. We would therefore choose a different starting point for an event teaser than for a live conversational assistant.

A moment of silence says something too

The documentation describes audio tags directing delivery: a whisper, emotion or pause. If you know the SSML break notation, you cannot use it here — v4 does not support it. Pauses are described through audio tags, punctuation and text structure.

Think of a short announcement. A quiet opening brings us closer to the scene, silence leaves room for the name and the last line adds energy. The words need not change. We change their delivery, and with it the invitation’s character.

Two voices on opposite sides of a door

For the fictional Cookie Sessions, here is our idea: the first voice speaks softly, like someone outside a closed venue. The second answers from inside. A brief exchange, silence, the event name and a poster. This is a recording plan; we have not produced the two-voice teaser yet.

We would keep the voice at the door close to the listener and restrained. The one inside could have a freer pace. Plan ambience and the sound transition separately in the edit, so changing one line does not require rebuilding everything. You could reverse the roles too and explore two interpretations: an intimate invitation and a stronger entrance.

V4 suits this idea. Its integration documentation points to Text to Dialogue API, while v4 Turbo has a streaming option for live conversations. Before choosing a performance, listen to the Polish pronunciation, the name Cookie Sessions, transitions between speakers and silence before the title. Ideally, do that alongside the poster: then you can hear whether voice and image belong to one scene.