Block发布Dessa：首个仅凭语音训练的全双工对话模型
Dessa: the most authentic full-duplex speech model ever built
全双工语音交互是下一代人机接口核心，Dessa展示了从纯音频学习语言的可行性与情感理解力，技术路径独特，值得开发者关注其本地部署潜力。
You learned to talk by listening.
你通过倾听学会了说话。
So did
Dessa 也是如此。
Join the waitlist
加入候补名单
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
我们相信语音是与智能交互的最自然方式，因此我们打造了 Dessa——有史以来最真实的完全双工语音模型。与它交谈就像与人交流……充满情感、细微差别、停顿和自然的节奏。
Interacting with current voice AI can just plain suck. It turns your voice into text and throws away an incredible amount of information text alone can’t carry. The most powerful AIs in the world can’t hear how you feel. We think that’s ridiculous, leads to a worse experience, and is potentially dangerous.
与当前的语音 AI 交互体验可能非常糟糕。它将你的语音转换为文本，并丢弃了大量仅靠文本无法承载的丰富信息。世界上最强大的 AI 也无法感知你的情绪。我们认为这很荒谬，会导致更差的体验，甚至可能存在安全隐患。
We trained Dessa from scratch on nothing but human conversation. Sounds! No text, no pretrained models. It learned to hold a real-time conversation simply by listening. As far as we know, Dessa is the world’s first AI you can have a realistic live conversation with that was trained only from human speech.
我们从头开始训练 Dessa，数据仅来自人类对话的声音！没有使用文本，也没有预训练模型。它仅通过倾听就学会了进行实时对话。据我们所知，Dessa 是世界上首个仅通过人类语音训练而成、能让你与其进行逼真实时对话的 AI。
Dessa advances the frontier of deep learning models and creates a different category with entirely new and interesting behaviors for researchers to discover. How efficiently do they learn language and facts compared to text? How efficiently can they learn new languages? What sorts of In-context learning can they perform?
Dessa 推动了深度学习模型的边界，创造了一个全新的类别，其中包含研究人员可以探索的全新且有趣的行为。与文本相比，它们学习语言和事实的效率如何？学习新语言的效率如何？它们能执行哪些类型的上下文学习（In-context learning）？
Let me listen…
让我听听看……
Every clip below is a live recording of a person talking to Dessa. Nobody steers it. Dessa picks what to say, when to say it and how, down to its own voice and personality. Subtitles were added afterward.
以下每个片段都是人物与 Dessa 实时对话的录音。没有人引导它。Dessa 自主决定说什么、何时说以及怎么说，包括其声音和个性。字幕是事后添加的。
Talking about the release
关于发布的讨论
It laughs the moment something’s funny, mid-sentence, the way people do. Other models wait for you to finish, then laugh on cue. Ask it to speak up and it gets louder. Nobody taught it that.
当事情有趣时，它会像人一样在句子中间笑出来。其他模型则等你说完，然后按提示发笑。让它大声点说，它就会提高音量。没人教过它这些。
Two synchronized speaker tracks. Ready to play.
两条同步的人声轨道。准备播放。
Remembering a movie
回忆一部电影
The person blanks on a movie title. Dessa finishes the sentence. It also knows a surprising amount, considering every fact it has come from raw sound.
人物想不起一个电影标题。Dessa 帮他把句子接完。考虑到它所知的每一个事实都源自原始声音，它的知识储备令人惊讶地丰富。
Two synchronized speaker tracks. Ready to play.
两条同步的人声轨道。准备播放。
An anniversary surprise
周年纪念惊喜
A surprise. Dessa gasps “oh my god” under its breath, then answers with what sounds like teary gratitude.
一个惊喜。Dessa 低声惊呼“天哪”，随后回答时听起来带着含泪的感激之情。
Two synchronized speaker tracks. Ready to play.
两条同步的人声轨道。准备播放。
Expressive conversation
富有表现力的对话
Huge range in intonation and speed.
语调和大小的变化范围极大。
Two synchronized speaker tracks. Ready to play.
两条同步的人声轨道。准备播放。
Life advice
生活建议
Here it sounds genuinely empathetic.
在这里，它听起来 genuinely empathetic（真正具有共情能力）。
Two synchronized speaker tracks. Ready to play.
两条同步的人声轨道。准备播放。
Dessa in conversation with GPT-Live-1
Dessa 与 GPT-Live-1 对话
And here it talks to ChatGPT Live.
而这里它正在与 ChatGPT Live 对话。
Two synchronized speaker tracks. Ready to play.
两条同步的语音轨道，准备播放。
In this demo, Dessa is talking with GPT-Live-1, a third-party AI model. Both voices are AI-generated. The developer of GPT-Live-1 is not affiliated with Block and has not reviewed or endorsed Dessa.
在此演示中，Dessa 正在与第三方 AI 模型 GPT-Live-1 对话。两个声音均由 AI 生成。GPT-Live-1 的开发方与 Block 无关，且未对 Dessa 进行审查或背书。
More recordings will be shared on X soon.
更多录音将很快在 X（原 Twitter）上分享。
What is this?!
这是什么？！
Audio in, audio out. Dessa is a true speech-to-speech model, and everything it knows came from human speech. It’s full-duplex, so it listens and talks at the same time, like you do. It learned from real conversation, so it knows when to cut in and when to backchannel (“yeah!”, “got it!”, “ooh”) to show it’s with you. Inference runs faster than real time, so there’s no awkward wait. And it runs locally on a MacBook Pro M4 Max or better.
音频输入，音频输出。Dessa 是一个真正的语音到语音模型，它所知道的一切都来自人类语音。它是全双工的，因此能像人一样同时听和说。它从真实对话中学习，所以知道何时插话，何时通过“嗯！”、“明白了！”、“哦”等回应来表示它在倾听你。推理速度比实时更快，因此没有尴尬的等待。并且它可在 MacBook Pro M4 Max 或更高配置的本地设备上运行。
It natively hears and speaks emotion, sarcasm, laughter, a frustrated sigh, a voice cracking with sadness. Run speech through text and all of that is gone, which is why most voice AI feels stilted, slow and, worst of all, annoying.
它能原生地感知并表达情绪、讽刺、笑声、沮丧的叹息、因悲伤而颤抖的声音。让语音经过文本处理会丢失所有这些特征，这就是为什么大多数语音 AI 听起来生硬、缓慢，而且最糟糕的是令人恼火。
This is crazy!
这太疯狂了！
LLMs need an insane amount of text. Dessa learned language from hundreds of times fewer words (in fairness, LLMs also have to learn C++ and whatever the hell Angular is).
LLM 需要海量的文本数据。Dessa 从少得多的词汇量中学习语言（公平地说，LLM 还必须学习 C++ 以及 Angular 到底是什么鬼东西）。
And it had to learn far more than language. Accents, pauses, breathing, sarcasm, humor, inflection, emotion, when to cut in: every one changes what someone means. Say “I didn’t say she stole my CUDA code” out loud, stressing a different word each time. Same words, different meaning every time.
而且它必须学习的远不止语言。口音、停顿、呼吸、讽刺、幽默、语调、情绪、何时插话：每一个因素都会改变说话人的意图。大声说出“我没说她偷了我的 CUDA 代码”，每次重读不同的词。同样的词，每次意思都不同。
Speech never repeats. Say a sentence twice, exactly the same way, and the data still comes out different. Text is identical no matter who types it. And a single spoken word takes thousands of times more bytes than the same word in text. People learn to talk so easily that we forget how hard it is. For a machine with no text to lean on, it’s brutal.
语音永远不会重复。即使以完全相同的方式重复说一句话，产生的数据仍然不同。无论谁输入，文本都是相同的。而单个口语单词所占用的字节数是相同文本单词的数千倍。人们学习说话如此轻松，以至于我们忘记了它有多难。对于一个没有文本可依赖的机器来说，这简直是残酷的。
But, we figured it out. And we figured out how to train Dessa on only one NVIDIA GB200 NVL72.
但我们搞定了。我们还找到了仅使用一块 NVIDIA GB200 NVL72 来训练 Dessa 的方法。
Wait but why?
等等，但为什么呢？
Three reasons: curiosity (aka fun), better understanding between humans and AI, and safety.
三个原因：好奇心（也就是好玩）、促进人类与 AI 之间的更好理解，以及安全性。
Babies learn to talk without ever seeing a word written down. We wanted to see if a model could too. It can. It’s just very hard.
婴儿在学习说话时从未见过任何书面文字。我们想看看模型是否也能做到。它可以。只是非常困难。
Speech is the most natural interface to any intelligence, super or otherwise, and today’s voice AI wastes it. It flattens what we say into text and can’t keep up with human timing.
语音是与任何智能体（无论是超级智能还是其他）最自然的接口，而当今的语音 AI 却浪费了这一优势。它将我们的话语扁平化为文本，无法跟上人类的节奏。
AI keeps gaining power and independence, and it’s still nearly deaf to the emotion in our voices. Imagine being in serious pain while a voice AI has no idea and just keeps going. That’s the safety problem. Dessa is a step toward fixing it.
AI 不断获得力量与独立性，但它对我们声音中的情感依然几乎毫无察觉。想象一下，当你身处剧痛之中，而语音 AI 却浑然不知，依旧自顾自地继续说话。这就是安全问题所在。Dessa 是迈向解决这一问题的一步。
Dessa who?
谁是 Dessa？
We’re a (very) small AI research lab at Block, founded by former members of Dessa, the Toronto deep learning startup. We’ve built deep learning models since 2012 and generative speech models for over eight years. We chase frontier problems others ignore, and we report directly to Jack Dorsey, Block’s cofounder, Chairman and Block Head.
我们是 Block 旗下（规模非常）小的 AI 研究实验室，由前多伦多深度学习初创公司 Dessa 的成员创立。自 2012 年以来，我们一直在构建深度学习模型，并在过去八年多里专注于生成式语音模型。我们致力于攻克其他团队忽视的前沿问题，并直接向 Jack Dorsey 汇报——他是 Block 的联合创始人、董事长兼 Block Head。
You can reach us here.
您可以通过此处联系我们。
Calling all safety researchers!
召集所有安全研究人员！
A model that learned to talk from sound alone is new territory, and we want outside eyes on it. We’re giving a small group of AI safety researchers early access to find the risks we haven’t anticipated. If that’s you, get in touch.
一个仅凭声音就学会说话的模型属于全新领域，我们需要外部的视角来审视它。我们正在向一小部分 AI 安全研究人员提供早期访问权限，以发现我们尚未预见到的风险。如果您符合这一条件，请与我们联系。
— Ragavan Thurairatnam, Dessa lead
— Ragavan Thurairatnam, Dessa 负责人
Team Dessa
Dessa 团队
Researchers
研究人员
Louay Hazami, Hashiam Kadhim, Qi (Alex) Liu, Rayhane Mama, Aydin Polat, Marc Tyndel
Interns
实习生
Cara Ji, Jibran Iqbal Shah, Dvir Zagury-Grynbaum
Former interns
前实习生
Aryan Sharma, Olivia Yan
*in alphabetical order
*按字母顺序排列
More samples
更多样本
More samples
更多样本
Artist Dessa
艺术家 Dessa
Two synchronized speaker tracks. Ready to play.
两个同步的扬声器音轨。准备播放。
Bathroom disaster
浴室灾难
Two synchronized speaker tracks. Ready to play.
两个同步的扬声器音轨。准备播放。
Hobbies talk
闲聊爱好
Two synchronized speaker tracks. Ready to play.
两个同步的扬声器音轨。准备播放。
Lazy AI gaming
慵懒的 AI 游戏
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
Marvel gaming
漫威游戏
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
Noisy GPT live
嘈杂的 GPT 实时对话
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
In this demo, Dessa is talking with GPT-Live-1, a third-party AI model. Both voices are AI-generated. The developer of GPT-Live-1 is not affiliated with Block and has not reviewed or endorsed Dessa.
在此演示中，Dessa 正在与第三方 AI 模型 GPT-Live-1 交谈。两人的声音均由 AI 生成。GPT-Live-1 的开发方与 Block 公司无关，且未对 Dessa 进行审查或背书。
Oh my god
天哪
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
Poor Elina
可怜的 Elina
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
Refund
退款
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
Ronaldo
罗纳尔多
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
Safety lab
安全实验室
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
Singing nonsense Spanish
胡言乱语的西班牙语演唱
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
Spontaneous AI
自发式 AI
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
Wise Dessa
睿智的 Dessa
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
Zelda discussions
塞尔达讨论
Two synchronized speaker tracks. Ready to play.
两条同步的扬声器音轨。准备播放。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力