ElevenLabs releases v4 speech models with support for over 90 languages
ElevenLabs has released v4 and v4 Turbo speech models. The company says the new generation supports more than 90 languages and gives users finer control over delivery, while Turbo targets shorter delays in voice-agent conversations. Audio can begin before the underlying language model completes its answer. The claimed quality and speed improvements are company assertions rather than independent comparative results.
Artificial Intelligence··Night
Two speech models arrive
ElevenLabs has released two speech models, v4 and v4 Turbo. The company says the new generation supports more than 90 languages and gives users finer control over emotion and delivery; the previous generation supported 70 languages. The standard v4 model is geared toward prepared audio, while Turbo targets voice-agent conversations where a short delay matters. The language count describes announced support, rather than an independent assessment of speech quality in each language.[1], [2]
Delivery tags can be sequenced
Inline expression tags introduced with v3 can be stacked and ordered in v4. ElevenLabs says the new model uses information about the speaker and earlier lines to keep delivery and voice identity steadier over long passages. A company example reported by TechCrunch describes cloning a voice from a 10-second recording. That example and the claimed improvement in consistency were not presented as independently measured comparative results.[1]
Turbo starts speaking before text finishes
v4 Turbo can begin generating audio before the underlying language model completes its reply. The streaming design is aimed at voice-agent applications with short exchanges between participants. ElevenLabs presents lower latency and smoother turn-taking as goals for the Turbo version. The reporting does not provide an independent speed comparison against rival systems under common conditions. The two-model release separates prepared speech from real-time conversational use, which places different demands on delivery and response time.[1]
