ElevenLabs v4 Speech Models Expand to 90 Languages for Smarter Global Voice Tools

ElevenLabs has released its v4 and v4 Turbo speech models with support for more than 90 languages.

The update builds on the v3 model from last year and adds better expression control along with lower latency.

Voice Cloning Gets Faster and More Accessible

Users can now create a voice clone from just 10 seconds of audio thanks to a new architecture.

This change lowers the barrier for founders who need custom voices without long recording sessions.

The models also maintain speaker identity across long text passages and adjust tone based on context.

Lower Latency Opens Doors for Real-Time Agents

Voice agents benefit most from the reduced response times that make conversations feel natural.

Startups building customer support tools can now deploy more human-like interactions in multiple markets.

Quality gains stand out in languages like Japanese, Brazilian Portuguese, Mandarin, and Cantonese.

Over the next 12 months this could accelerate adoption of voice AI in customer service across Asia and Latin America.

Historically text-to-speech tools started robotic and have steadily moved toward human-like delivery with each generation.

Founders should watch how these models reshape competition in the voice agent space.