Eleven v4：我们最具表现力的文本转语音 AI 模型

同一句话会因说话方式不同而产生很大变化。“我需要你保持冷静”听起来应该有所不同，具体取决于说话的人：可能是医生温柔地安抚受惊的病人，也可能是游戏角色在投入战斗前对小队大声喊出这句话。A line of text can change significantly depending on how it’s spoken. “I need you to stay calm" should sound different depending on who's saying it, whether that's a doctor delivering it gently to a frightened patient, or a character in a game shouting to his squad before dropping into battle.

今天，我们正式推出 Eleven v4——迄今情感表现力最强的文本转语音模型，以及它的低延迟版本 Eleven v4 Turbo。Today we're launching Eleven v4, our most emotive text-to-speech model yet, and its low-latency variant, Eleven v4 Turbo.

Eleven v4 在 Artificial Analysis1 的排名中位列第一，并在与竞品模型进行的盲测一对一对比中，受到约 75% 的听众青睐2。它旨在理解语气、语速、情绪、角色和上下文。生成的语音可以听起来戏剧化、温柔、急切、诙谐或自然随意，同时保持说话者的身份特征。更自然的多说话者互动，让对话听起来更有来有往，而不是像把一句句台词拼在一起。说话者会根据对话上下文作出回应，从而呈现更自然的对话和角色互动。Ranked #1 by Artificial Analysis¹, and preferred by ~75% of listeners in blind head-to-head tests over competing models², Eleven v4 was designed to interpret tone, pacing, emotion, character, and context. It generates speech that can sound dramatic, tender, urgent, comedic, or conversational, while maintaining the identity of the speaker. More natural multi-speaker dynamics make conversations feel responsive, rather than like separate lines assembled together. Speakers respond to the context of the conversation, producing more natural dialogue and character interactions.

专为性能打造的新架构A new architecture built for performance

Eleven v4 基于全新架构打造，是我们情感表现力最强的文本转语音模型，并经 Artificial Analysis¹ 排名第一。Eleven v4 Turbo 将同样的技术应用于智能体等低延迟场景。其推理延迟中位数约为 100 毫秒，响应速度甚至快于两个人交谈时的平均停顿时间。Built on an entirely new architecture, Eleven v4 is our most emotive Text to Speech model ranked #1 by Artificial Analysis¹. Eleven v4 Turbo brings that same technology to low-latency use cases like agents. With a median inference latency of ~100ms, it can respond faster than the average pause between two people talking.

两种模型都能生成富有情感而非机械感的语音。整体音频保真度也更高，输出更加清晰、自然。Both models can generate speech that feels emotive, rather than mechanical. Underlying audio fidelity is also higher across the board, with cleaner, more natural-sounding output. 

早期几代文本转语音模型能够朗读文本，而 Eleven v4 则能为语音赋予更深层的情感，听起来更加自然。该模型可以理解预期的语气、语速节奏、文字风格以及文本中的上下文，从而生成富有感染力的语音。Earlier generations of text-to-speech models can read text aloud, but Eleven v4 provides an emotional depth to speech that feels far more natural. The model can interpret the intended tone, how the speech should be paced, the character of the writing, and context from the text to deliver emotionally resonant speech.

用户还可以精细控制输出，并用自然语言描述一句话应该如何表达。他们还可以添加具体指示，说明某些短语该怎么说、应传达什么情绪，甚至指定音效，例如使用 [laughs]、[said angrily in French accent]、[light rain] 或 [phone buzzing] 这样的内联标签。Eleven v4 对这些音频标签和指示提示的遵循比以往的模型更准确，因此你描述的表达方式就是最终呈现的效果。这让你更容易指导旁白、塑造角色或角色对话，以及处理其他需要关注表达方式的场景。Users also have fine-grained control over outputs, and can describe how a line should be delivered in natural language. They can also add specific instructions on how to say certain phrases, what emotion they should convey, and even sound effects, using inline tags like [laughs], [said angrily in French accent], [light rain], or [phone buzzing]. Eleven v4 follows these audio tags and direction prompts more accurately than prior models, so the delivery you describe is what you get. This makes it easier to direct narration, flesh out characters or their dialogue, and other places where delivery matters. 

ElevenLabs 开发者文档涵盖了完整的标签语法，以及如何通过 API 应用这些标签。对国际音标（IPA）音素的支持也得到了显著改进，因此自定义发音更加可靠。The ElevenLabs developer documentation covers the full tag syntax and how to apply these tags through the API. Support for the International Phonetic Alphabet (IPA) phonemes has also been significantly improved, so custom pronunciations behave more reliably.

Eleven v4 还提升了多位说话者之间对话的一致性和动态表现。通过一种全新的说话者身份捕捉方法，Eleven v4 能保留每种声音的独特特质。无论是智能体对话、有声书还是广告，它都能让语音保持一致。由于该模型能够理解整个场景的上下文，因此生成的对话自然流畅，说话者会针对刚刚说过的话作出回应，而不是把互不相关的台词拼接在一起。Eleven v4 also brings improvements in consistency and dynamic conversations between multiple speakers. Using a new method for capturing speakers’ identities, Eleven v4 preserves the unique qualities of each voice. It’s able to keep speech consistent through agent conversations, audiobooks, or ads. Because the model understands the context of a whole scene, it generates natural dialogue where speakers respond to what's just been said, rather than stitching together isolated lines.

适用于低延迟场景的 Turbo 模型A Turbo model for low-latency use cases

高质量语音模型生成语音的速度往往较慢，这意味着用户不得不在快速响应和富有表现力的语音代理之间做出选择。许多人选择了能力出众却语调单一的代理，而对于那些焦急地只想解决问题的客户来说，这样的代理听起来可能像机器人。High-quality voice models have tended to be slower to generate speech, meaning users have had to choose between fast or expressive voice agents. Many have opted for proficient, but monotone, agents that can seem robotic to a flustered customer just looking to resolve their issue. 

Eleven v4 Turbo 将速度与情感表现融为一体，首次语音输出时间的中位数约为 150 毫秒³，因此能够在众多行业部署强大的智能体：从能准确读出医学术语、温暖而令人安心的医疗健康智能体，到语速飞快、满口俚语、为玩家带来娱乐体验的游戏智能体。Eleven v4 Turbo combines speed and emotion with a median time to first speech of ~150ms³, making it possible to deploy powerful agents in countless industries, from warm, reassuring agents that can accurately pronounce medical terms in healthcare, to fast-talking, slang-wielding agents to entertain players in gaming.

Eleven v4 Turbo 模型专为 ElevenLabs 的对话式智能体平台 ElevenAgents 打造。其他智能体构建工具会将不同供应商提供的模型和软件拼接在一起，因此用户无法针对各自的使用场景优化或改进模型输出。ElevenLabs 的研究和工程团队将 Eleven v4 Turbo 与 ElevenAgents 作为一个整体系统进行协同优化，从而打造出表现力更强、更可靠、延迟更低的智能体。The Eleven v4 Turbo model is built to work with ElevenLabs’ conversational agents platform, ElevenAgents. Other agent builders stitch together models and software from different vendors, meaning users are left without a way to refine or improve the model outputs for their individual use cases. ElevenLabs’ research and engineering teams optimized Eleven v4 Turbo and ElevenAgents together as one system, delivering more expressive, reliable, and low latency agents.  

一种声音，万千语言One voice, myriad languages 

我们的沟通方式会因情境、地点甚至一天中的时段而异。要构建能够在不同语言中反映这些差异、富有表现力的语音，仍然是音频 AI 面临的最大挑战之一。The way we communicate varies by context, location, even the time of day. Building expressive speech that mirrors that, across languages, remains one of the hardest challenges in audio AI. 

Eleven v4 在这些方面都更进一步，而且改进不只体现在模型对短语的发音上。Eleven v4 对不同语言中的节奏、情感和表达方式把握得更好，帮助创作者生成符合语言习惯和语境的自然语音。比如，在日本，你和一位素不相识的长者交谈时的说话方式，就和在意大利时大不相同。Eleven v4 is a step up in all these areas, and the improvements extend beyond just how the model pronounces phrases. Eleven v4 better captures rhythm, emotion, and delivery across languages, helping creators produce speech that feels natural to the language and context. The way you’d speak to an elderly stranger in Japan is quite different from the way you’d speak to them in Italy, for example. 

Eleven v4 和 Eleven v4 Turbo 均支持 90 多种语言。现在，以一种语言录制的声音也能流利地说其他任何语言，在保留原声音色身份的同时，采用母语者的口音。与以往相比，口音贴合度明显更高，因此在生成过程中，声音不再逐渐偏回源语言的口音。Both Eleven v4 and Eleven v4 Turbo support more than 90 languages. Now, a voice recorded in one language also speaks any other fluently, adopting the accent of a native speaker while retaining the identity of the original. That accent adherence is noticeably stronger than before, so the voice no longer drifts back toward its source accent over the course of a generation.

对于配音和本地化而言，这意味着你为品牌选定的声音——无论是与你合作的名人演员的声音，还是最符合公司风格的语调——用 Eleven v4 支持的任何语言呈现出来，效果都会很出色。For dubbing and localization, this means the voice of the brand you’ve decided on — whether that’s a celebrity actor you’re working with or a tone that best fits your company’s style — will sound great in any language supported by Eleven v4.

更真实、更一致的声音克隆More authentic, consistent voice cloning

在 Eleven v4 和 Eleven v4 Turbo 中，声音克隆更加真实、强大且稳定，与原始音源的说话人相似度显著提升。即时语音克隆现在仅凭 10 秒音频就能高保真地捕捉声音。Voice cloning is more authentic, powerful, and consistent across Eleven v4 and Eleven v4 Turbo, with significantly better speaker similarity to the original source voice. Instant Voice Clones can now capture voices with high fidelity using just 10 seconds of audio. 

Eleven v4 在不同生成内容、对话、旁白和重新生成的台词中，也能更可靠地保持说话者身份一致。这意味着在长篇项目中，角色、旁白者和克隆语音在整个制作过程中都能保持一致。Eleven v4 还支持专业语音克隆（PVC），适用于对克隆保真度要求最高的场景。Eleven v4 also preserves speaker identity more reliably across generations, dialogue, narration, and regenerated lines. This means for long-form projects, the characters, narrators, and cloned voices stay consistent throughout a production. Eleven v4 also adds support for Professional Voice Clones (PVC), for the highest-fidelity cloning use cases.

请求拼接——将多次生成串联起来，创作更长篇幅的内容——在 Eleven v4 中也可靠得多，从而提升了在 ElevenLabs Studio 和 ElevenLabs Reader App 中的使用体验。Request stitching — chaining generations together for longer-form content — is also significantly more reliable in Eleven v4, which improves the experience of working in ElevenLabs Studio and the ElevenLabs Reader App.

亲自听听其中的差别。Hear the difference for yourself  

Eleven v4 和 Eleven v4 Turbo 集我们在富有表现力的语音生成领域的最新研究成果于一身。它们专为注重表达方式、与措辞同样重要的内容打造，适用于有声书、角色演绎、旁白、影视配音以及对话式智能体的本地化。Eleven v4 and Eleven v4 Turbo are the culmination of our latest research in expressive speech generation. They're built for content where delivery matters as much as the words themselves, whether that’s audiobooks, character performances, voiceovers, dubbing, or localizing conversational agents.

现在，ElevenAgents、ElevenCreative 和 ElevenAPI 均已提供这两款模型。立即创建免费账户，开始使用 Eleven v4 或 Eleven v4 Turbo 进行生成。Both models are available now in ElevenAgents, ElevenCreative, and via ElevenAPI. Create a free account to start generating with Eleven v4 or Eleven v4 Turbo today.

1. Artificial Analysis，服务商语音竞技场排行榜，2026年9月
   ^{Artificial Analysis, Provider Voice Arena Leaderboard, Sept 2026}
2. 基于 2026 年 9 月与 Cartesia Sonic 3.6、Inworld TTS-2、Google Gemini 3.8 Flash-Lite TTS 和 Google Gemini 3.8 Flash TTS 进行的盲测用户偏好对比。每组对比中，评审以盲测方式听取 Eleven v4 和一款竞争产品朗读的同一句台词，并判断哪一个表达力更强、听起来自然度更高；平局按半票计算。
   ^{Based on blind head-to-head user preference testing against Cartesia Sonic 3.6, Inworld TTS-2, Google Gemini 3.8 Flash-Lite TTS, and Google Gemini 3.8 Flash TTS, September 2026. For each pair, graders head the same line from Eleven v4 and one competitor, presented blind, and judged which was more expressive and which sounded more natural; ties counted as half.}
3. 从请求到听到语音的中位时间。于 2026 年 9 月使用相同脚本和默认设置，与 Cartesia Sonic 3.6、xAI TTS、Google Gemini Flash-Lite TTS 和 OpenAI GPT-4o mini TTS 进行比较测量；所有系统均测量并扣除了网络延迟。Eleven v4 Turbo 使用 WebSocket 流式传输。
   ^{Median time from request to audible speech. Measured September 2026 with identical scripts and default settings against Cartesia Sonic 3.6, xAI TTS, Google Gemini Flash-Lite TTS, and OpenAI GPT-4o mini TTS; network latency measured and removed for all systems. Eleven v4 Turbo over WebSocket streaming.}