ElevenLabs releases new v4 speech model supporting 90 languages
AI voice company ElevenLabs has launched its new v4 speech model, enhancing expression control and expanding language support to over 90 languages. The model also allows for faster voice cloning.
ElevenLabs announced on Monday the release of its v4 and v4 Turbo speech models, which offer improved expression control, reduced latency for voice agents, and support for more than 90 languages.
The new v4 architecture enables greater control and faster voice cloning, with the company stating users can clone a voice from just 10 seconds of audio. Creatively, the model better handles voice identity over longer text segments and considers textual context to adjust expressions. This builds on features introduced in v3, expanding tag-based control for more nuanced delivery.
The company has increased language support from 70 languages in the previous version to 90 in v4. ElevenLabs noted significant quality improvements in Japanese, Brazilian Portuguese, Mandarin, and Cantonese.
The model is designed for voice agents due to its lower latency, allowing for more fluid conversations. It can also handle aspects like customer frustration and escalations more effectively for improved issue resolution, generating audio as soon as the underlying language model provides output.