ElevenLabs Releases v4 and v4 Turbo Voice Models with 10-Second Voice Cloning
ElevenLabs has officially launched its next-generation v4 and v4 Turbo voice models built on a brand-new architecture. The updated models feature 10-second voice cloning, support for over 90 languages, lower latency streaming for conversational AI agents, and enhanced emotional control. As enterprise demand for real-time AI voice agents accelerates, reducing streaming latency and expanding language support are critical for natural customer support interactions. This release solidifies ElevenLabs' position in AI audio synthesis as its annual recurring revenue surpasses $600 million amid preparation for a future IPO. The v4 architecture allows models to begin audio generation simultaneously as an upstream LLM streams its response, helping manage complex call scenarios like disputes or wait times. Additionally, users can now stack multiple inline tags in sequence to fine-tune voice tone and emotion dynamically across long passages.
## BACKGROUND
ElevenLabs is a prominent synthetic voice company offering realistic text-to-speech (TTS) and audio generation technology. In conversational AI, streaming TTS models process real-time text outputs from large language models to produce human-like speech with minimal response delay. Inline tags are embedded text markers used by TTS engines to direct emotion, pacing, and tone mid-sentence.