ElevenLabs rolled out Expressive Mode for its ElevenAgents platform, giving voice bots the ability to shift tone mid-conversation. An agent handling an angry customer can soften its delivery; one walking someone through a complex process can pick up pace and clarity. That's the pitch, anyway.
Two pieces make it work. The first is Eleven v3 Conversational, a low-latency variant of the company's Eleven v3 TTS model built specifically for live dialogue. It tracks conversational context across turns and adjusts emotional delivery, supporting tags like [whispers], [sighs], and [laughs] that the LLM can inject for fine-grained control. Each tag affects roughly 4-5 words before reverting to normal delivery. Pricing matches existing ElevenLabs TTS in Agents: $0.08 per minute, per the docs.
The second component is a rebuilt turn-taking system powered by Scribe v2 Realtime, ElevenLabs' transcription engine. Instead of relying solely on silence detection, it reads prosodic cues to figure out whether "yeah" is a full response or a lead-in to more. The company claims this reduces the classic bot problem of interrupting mid-sentence, though real-world performance will depend heavily on use case and language.
Karthik Rajaram, ElevenLabs' India country head, called it "a shift from scripted voice automation to emotionally intelligent conversations," framing India as a key market. The company already counts Meesho, Cars24, and TVS Motors among its Indian enterprise clients.
Expressive Mode is live now for ElevenAgents users. Language support spans 70+ languages via Eleven v3, though the company notes expressiveness "may vary across voices and languages," a caveat worth testing before any production deployment.
The Bottom Line: ElevenLabs is betting that emotional nuance and better conversational timing, not just voice quality, will differentiate enterprise voice bots in 2026.
QUICK FACTS
- Eleven v3 Conversational priced at $0.08/minute (same as other ElevenLabs TTS in Agents)
- Expressive tags (
[whispers],[sighs],[laughs]) affect ~4-5 words each - Scribe v2 Realtime powers the new turn-taking system
- 70+ languages supported via Eleven v3 (expressiveness varies by language, company-reported)
- Expressive Mode enabled by default when selecting v3 Conversational as agent TTS model
