ElevenLabs Launches V4 Speech Models With More Expressive AI Voices

ElevenLabs has launched Eleven v4 and Eleven v4 Turbo, its latest text-to-speech models designed to make AI-generated voices more expressive, consistent and responsive. The company announced the models on September 28, 2026, positioning V4 as a major upgrade for creators, developers and businesses building voice-based applications.

One of the major improvements is greater control over how AI voices deliver speech. V4 can take the surrounding context into account when generating audio and respond to instructions about emotion, tone and delivery. ElevenLabs has also expanded its use of inline expression tags, allowing creators to guide changes in emotion and delivery within a single piece of text.

The company has also expanded language support from more than 70 languages to over 90. ElevenLabs says the new model shows particularly notable improvements in languages including Japanese, Brazilian Portuguese, Mandarin and Cantonese. The system also supports cross-language voice generation, allowing a recorded voice to speak another language while maintaining the identity of the original speaker.

Voice cloning has received an upgrade as well. ElevenLabs says its technology can create a high-fidelity voice clone from approximately 10 seconds of audio, while V4 is designed to maintain a speaker’s identity more consistently across longer passages. The V4 Turbo model focuses instead on speed, with lower latency aimed at real-time conversations and AI voice agents.

The launch comes as competition in AI-generated speech continues to grow, with companies including Google, OpenAI, Deepgram, Cartesia and Mistral developing increasingly capable voice systems. ElevenLabs has also expanded beyond speech into areas such as music, dubbing, transcription and AI agents. The company says more than 55% of its business now comes from large enterprise customers, highlighting the growing role of voice AI in commercial applications.