ElevenLabs Launches v4 Speech Models With 90+ Language Support

ElevenLabs has launched two new speech models, ElevenLabs v4 and v4 Turbo, offering more expression control, lower latency for voice agents, and support for more than 90 languages. The announcement follows the release of v3 last year and a teaser at an event in Warsaw earlier this year.

The v4 generation uses a new architecture that enables better control and faster cloning. Users can clone a voice with just 10 seconds of audio. The model handles voice identity better over longer chunks of text and keeps context in mind to change expressions. Inline tags, introduced in v3, are expanded in v4, allowing users to stack multiple tags.

Language support has grown from 70 languages in the previous version to over 90. The biggest quality jump was observed in Japanese, Brazilian Portuguese, Mandarin, and Cantonese. For voice agents, the lower latency allows audio generation to start as soon as the underlying LLM begins generating answers. The model can also handle confrontations, escalations, and holds differently for better issue resolution.

More than 55% of ElevenLabs' business comes from large companies. The company raised $500 million earlier this year from Sequoia, valuing it at $11 billion, and rumors suggest a follow-up round could value it at $22 billion. Annualized revenue run rate has climbed from roughly $330 million at the start of the year to over $600 million. ElevenLabs has over 800 employees and has hired aggressively in India, Europe, and Brazil.

CEO Mati Staniszewski said the company is aiming for an IPO in the next years but didn't commit to a timeline. Competitors in speech models include Cartesia, Deepgram, Fish Audio, Boson, WellSaid Labs, Google, and OpenAI. The launch was reported by techcrunch.com.