London-based AI voice startup ElevenLabs launched speech models v4 and v4 Turbo on September 28, with more expression controls, lower latency for voice agents, and support for more than 90 languages.

The company said the models can clone a voice from 10 seconds of audio, maintain a more consistent voice identity over longer passages, and begin generating audio while a large language model is still producing text.

ElevenLabs said the release expands language coverage from 70 in v3 to more than 90, with some of the largest quality gains in Japanese, Brazilian Portuguese, Mandarin, and Cantonese.

The company, which also sells an enterprise AI voice platform, said large companies account for more than 55% of its business. In February, it said it raised US$500 million at an US$11 billion valuation.

The launch comes as rivals such as Cartesia and Deepgram target voice agent and enterprise text-to-speech markets.

🔗 Source: TechCrunch

Recent ElevenLabs developments