ElevenLabs launches Voice Library for AI speech

If you have listened to an eerily realistic AI audiobook lately, odds are ElevenLabs was behind the mic.

The audio research company has consolidated its platform into two main engines: ElevenCreative for media production, and ElevenAgents for conversational AI.

An all-in-one studio for speech, music, and SFX

ElevenLabs is no longer just a text-to-speech tool. Its Voice Library now hosts over 17,500 voices, spanning expressive and controllable speech across more than 90 languages.

On the creative side, users can clone existing voices, generate new ones from text prompts, or compose studio-grade music and custom sound effects directly inside its multi-track editor.

It even integrates third-party video and image models like Veo, Wan, Kling, and Seedance to bridge audio prompts directly into finished video clips.

Enterprise brands are already leaning in. NVIDIA uses the synthetic speech tech for multilingual marketing, while Duolingo taps it to power character voices.

Agents that talk, listen, and handle business

The other major half of the ecosystem is ElevenAgents, a platform built to deploy conversational voice bots across phone lines, chat, email, and WhatsApp.

To make phone calls feel natural, latency is critical. The platform features models like Eleven Flash, which brings conversational latency down to 75 milliseconds.

It also pairs with Eleven Scribe, a speech-to-text engine boasting 98 percent accuracy with speaker diarization and character-level timestamps.

To keep enterprise bots from going off the rails, the system includes built-in simulation testing, workflow logic, and compliance guardrails.

Why it matters

ElevenLabs is smartly moving away from being a single-feature voice generator to owning the entire audio production and enterprise interaction stack.

For creators and developers, having voice, music, sound effects, and transcription in one unified API eliminates the headache of stitching together multiple audio vendors.

For the wider tech landscape, pushing conversational latency to 75 milliseconds across 90 languages sets a new baseline for customer support agents. The days of robotic interactive voice response menus are basically numbered.

The lingering challenge remains moderation. With over 17,500 accessible voices and ultra-realistic cloning, enforcing guardrails against impersonation and intellectual property misuse will test how effectively the platform can police its own power.