Eleven v4

Eleven v4 is a text-to-speech and voice generation model developed by ElevenLabs. Built upon an entirely new architecture, the model interprets context, tone, pacing, emotion, and character to produce speech that mirrors natural human performance. It reads scripts dynamically, maintaining consistent speaker identity across long-form narration, multi-speaker dialogue, and continuous generation tasks.

The model lineup features two core variants sharing the same underlying research. Eleven v4 is tuned for produced content where maximum fidelity and nuance are required, such as audiobooks and voiceovers. Meanwhile, Eleven v4 Turbo provides a low-latency option optimized for real-time voice agents and interactive dialogue systems, achieving a median inference latency of approximately 100 milliseconds.

Both variants support over 90 languages with advanced pronunciation control through the International Phonetic Alphabet (IPA). The system introduces improved Instant Voice Clones—which require only 10 seconds of source audio—alongside higher-fidelity Professional Voice Clones. Additionally, enhanced inline audio tags allow users to direct specific delivery styles, pauses, and emotional reactions directly within text inputs.