ElevenLabs 发布 Eleven v4 与 v4 Turbo 语音模型

ElevenLabs has released Eleven v4 and Eleven v4 Turbo, dividing its latest text-to-speech generation between produced audio and real-time voice agents. Eleven v4 is the quality-focused model, while Turbo targets live calls and agent loops with about 100 ms median inference latency and 150 ms median time to first speech. Both are available through ElevenLabs products and the ElevenAPI, with Turbo also offered through ElevenAgents.

Built on a new architecture, Eleven v4 reads a script with awareness of who is speaking, what happened before, and how each line should land. It supports more than 90 languages, multiple speakers, sound effects, and a broad emotional range. Creators can place directions such as [laughs], [whispers], and [door slams] directly in a script. ElevenLabs says the model follows tag sequences more reliably than v3. SSML break tags are disabled, so pauses and delivery are controlled through natural-language audio tags.

Context stitching is designed to keep pacing and delivery consistent across long-form projects, while each generation supports up to 10,000 characters. Speaker stability is intended to prevent vocal drift when a line is regenerated, and Professional Voice Clones return after being unavailable in v3. All 17,500-plus voices in the company’s library work with v4, though older Instant and Professional Voice Clones need retraining.

Eleven v4 Turbo adds bidirectional streaming, allowing developers to send text as a language model produces it and receive audio before a sentence is finished. It retains v4’s expressive controls and supports the same Professional Voice Clone across a call. In ElevenLabs’ comparison, Turbo reached speech in 150 ms, versus 262 ms for Cartesia Sonic 3.6 and 814 ms for OpenAI GPT-4o mini TTS.

The release expands an ElevenLabs platform that also includes voice cloning, voice design, pronunciation dictionaries, and MP3, WAV, PCM, and telephony-ready µ-law output. Developers can use streaming or non-streaming endpoints through REST, TypeScript, and Python tools, then switch models with a single model ID. Both models are included on every plan, including a free tier with 10,000 monthly credits, while paid plans start at $6 per month. Every cloned voice requires verified owner consent, and generated audio is covered by the company’s AI Speech Classifier.

Source