ElevenLabs v4 Brings More Expression Control and 90+ Languages to AI Voice
The new speech model adds more control over tone, pacing and emotion, while Eleven v4 Turbo targets real-time voice applications.
ElevenLabs has launched ElevenLabs v4, a new speech synthesis model designed to give creators more control over how AI-generated voices sound. The model supports more than 90 languages and is joined by Eleven v4 Turbo, a lower-latency version aimed at real-time voice applications.
ElevenLabs says ElevenLabs v4 is built to interpret tone, pacing, emotion, character and context instead of treating speech as a simple conversion from text to audio. The company says the model can produce different styles of delivery while maintaining the identity of the speaker.
ElevenLabs v4 Adds More Control Over Voice Delivery
The model lets users describe how a line should be performed and supports inline controls for things such as emotion, reactions and sound effects. That gives creators more control over the delivery of a sentence without having to rewrite the underlying text.
ElevenLabs also says v4 improves multi-speaker dialogue and long-form generation. The company says request stitching, which chains multiple generations together for longer content, is more reliable in ElevenLabs Studio and the ElevenLabs Reader App.
One Voice Can Speak Across 90+ Languages
ElevenLabs v4 supports more than 90 languages, including Hindi, Urdu, Punjabi, Bengali, Tamil, Telugu and other languages across South Asia. ElevenLabs says a voice recorded in one language can speak another supported language while retaining the original speaker's identity and adopting a native accent.
That multilingual focus connects with the broader push toward AI voice interfaces. Saganote has also covered Claude Voice Mode gaining support for more languages and actions and Google Translate Live adding background and earpiece translation, showing how speech systems are increasingly being used across different languages and interaction types.
Eleven v4 Turbo Targets Real-Time Voice Apps
ElevenLabs is also launching Eleven v4 Turbo for applications where response speed matters. The company says the model has median inference latency of about 100 milliseconds and is designed for AI assistants, customer-support agents and interactive characters.
The focus on real-time speech puts ElevenLabs in the same broader category as other AI voice systems. Saganote recently covered Grok Voice Transcribe 2.0 and its speech-to-text improvements, although that system addresses speech recognition rather than text-to-speech generation.
ElevenLabs says both Eleven v4 and Eleven v4 Turbo are available through ElevenAgents, ElevenCreative and the ElevenAPI. The company's official Eleven v4 announcement provides the launch details and performance claims.