Block 的 Dessa 是一个只用人类对话训练的语音到语音模型——全双工、在 MacBook Pro M4 Max 或更好的机器上本地运行、训练只用了一套 NVIDIA GB200 NVL72——目前仅候补名单与安全研究者早期访问，尚无公开可用

2026 年 10 月 23 日（星期五，已按日历核算；日期由 Block 自己的公告 X 帖标注——@blocks 帖子的日期标签写着「6:50 PM · Oct 23, 2026」），Block 在 block.xyz/dessa 发布了 Dessa，该页面自身不带日期标记（申请表内嵌隐私声明的「Effective Date: October 21, 2026」是内嵌表单 chrome）。按页面原文：「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最自然的方式，所以我们造了 Dessa——有史以来最真实的全双工语音模型。和它说话的感觉就像在和一个人说话……充满情绪、细腻、停顿和自然的节奏。）」训练主张：「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.（我们从零开始、只用人类对话训练了 Dessa。是声音！没有文本，没有预训练模型。它仅仅靠听就学会了进行实时对话。）」——首创主张带 Block 自己的对冲：「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.（据我们所知，Dessa 是世界上第一个只靠人类语音训练、就能进行逼真实时对话的 AI。）」对当前语音 AI 的批评，Block 的话：「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.（和现在的语音 AI 交互可能真的很糟。它把你的声音变成文本，扔掉了大量光靠文本承载不了的信息。世界上最强的 AI 听不出你的感受。我们认为这很荒谬，体验更差，而且有潜在危险。）」机制：「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.（音频进、音频出。Dessa 是一个真正的语音到语音模型，它知道的一切都来自人类语音。它是全双工的，所以能像你一样同时听和说。）」；「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.（它从真实对话中学习，所以知道什么时候插话、什么时候搭腔（「yeah!」「got it!」「ooh」）来表示它跟着你。）」；「Inference runs faster than real time, so there's no awkward wait. And it runs locally on a MacBook Pro M4 Max or better.（推理跑得比实时还快，没有尴尬的等待。而且它在 MacBook Pro M4 Max 或更好的机器上本地运行。）」；以及「It natively hears and speaks emotion, sarcasm, laughter, a frustrated sigh, a voice cracking with sadness.（它原生地听懂并说出情绪、讽刺、笑声、沮丧的叹息、悲伤到哽咽的声音。）」Block 的效率说法：「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).（大模型需要海量的文本。Dessa 用少几百倍的词量学会了语言——公平地说，大模型还得学 C++ 和什么鬼 Angular。）」算力一句：「we figured out how to train Dessa on only one NVIDIA GB200 NVL72（我们搞明白了怎么只用一套 NVIDIA GB200 NVL72 训练 Dessa）」——页面给「one」加了加粗；GB200 NVL72 是机架级系统，TechCrunch 的「a single NVIDIA GPU（一块 NVIDIA GPU）」是该媒体的压缩，按信源各留各的。样本：「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.（下面每一段音频都是真人与 Dessa 对话的现场录音。没有人在幕后操控。说什么、什么时候说、怎么说——包括它自己的音色和个性——都由 Dessa 自己决定。字幕是事后加上的。）」——其中一段演示是「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.（在这段演示里，Dessa 在与第三方 AI 模型 GPT-Live-1 对话。两个声音都是 AI 生成的。GPT-Live-1 的开发者与 Block 无关，也没有审阅或为 Dessa 背书。）」实验室：「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.（我们是 Block 内部一个（非常）小的 AI 研究实验室，由多伦多深度学习创业公司 Dessa 的前成员创立。我们自 2012 年起做深度学习模型，做生成式语音模型已经八年多。）」并且「directly to Jack Dorsey, Block's cofounder, Chairman and Block Head（直接向 Block 联合创始人、董事长兼 Block Head Jack Dorsey 汇报）」。可用性：「Join the waitlist」加申请访问表单，并按「Calling all safety researchers!」一节：「We're giving a small group of AI safety researchers early access to find the risks we haven't anticipated.（我们正在给一小群 AI 安全研究者早期访问权限，让他们找出我们没有预料到的风险。）」——署名「— Ragavan Thurairatnam, Dessa lead」——而「More recordings will be shared on X soon.（更多录音将很快在 X 上分享。）」是公告正文唯一的前瞻时间句（内嵌隐私声明 chrome 另有一句「Launch links are coming soon.（上线链接即将推出。）」）。按 TechCrunch（Ivan Mehta，published_time 2026-10-24T16:00:00+00:00）：「Jack Dorsey's Block released a new voice model out of the company's research lab that runs locally and sounds strangely human with its expressions like laughs, pauses, and cries.（Jack Dorsey 的 Block 发布了一个出自公司研究实验室、本地运行的声音模型，它的笑、停顿和哭腔听起来出奇地像人。）」——附该媒体的可用性判断：「However, there is no public access to Dessa at the moment, and Block has put out a call for safety researchers to test the model.（不过，Dessa 目前没有公开访问渠道，Block 已发出邀请安全研究者测试该模型的呼吁。）」以及其业务背景：「Square already uses voice AI to handle ordering.（Square 已经在用语音 AI 处理点单。）」