Vivid Multi-Character Dialogue Built with Eleven V3 Audio Tags
Key point
ElevenLabs' **Eleven V3** uses audio tags to bring natural multi-character dialogue to life with a single model.
Details
ElevenLabs' Eleven V3 supports Multi-character dialogue, a feature that lets a single model perform the roles of multiple characters through audio tags. Users can insert tags such as [interrupting], [overlapping], and [laughs] directly into a script to create natural conversational flow.
Complex scenes that previously required multiple voice actors, recordings, and timing adjustments can now be handled with a single script. The key tag features are as follows.
- Turn control:
[interrupting],[overlapping],[cuts in] - Emotional shifts:
[excited],[annoyed],[flustered],[casual] - Rhythm control:
[fast-paced],[hesitates],[pause],[drawn out] - Identity switching:
[childlike tone],[deep voice],[pirate voice],[robotic tone]
This feature aims not merely for simple line-swapping, but for an orchestrated performance in which voices interact, react, and overlap with one another. This provides storytellers, game writers, and interactive designers with an environment for producing complex scenes while lowering production costs.
However, since this is currently in the Research Preview stage, using an Instant Voice Clone (IVC) or a designed voice is more advantageous for optimizing the V3 feature than using Professional Voice Clones (PVC). PVC optimization is planned for a future update.