评论
打印
So, Geoff Fowler, my colleague on the tech team at The Post, my understanding is that you have something that you want to play for me.
杰夫·福勒，我在《华盛顿邮报》技术团队的同事，据我所知，你有些东西要放给我听。
Yeah, that's right.
是啊，没错。
Hi there. It's Martine. Ooh.
嗨，你好啊。是玛蒂娜。噢！
I'm here with tech columnist Geoffrey Fowler to talk about advancements in AI audio.
今天，我要和科技专栏作家杰弗里·福勒谈论人工智能音频的进步。
What?!
什么？
In the last few months, AI audio technology has gotten so good that it's...a little scary.
过去几个月里，人工智能音频技术蓬勃发展，以至于它……有点吓人。
You can use AI to generate a podcast about any topic you want.
你可以使用人工智能来生成任何你想要的话题的播客。
No. Oh, my God, Geoff, no. That sounds just like me.
不是吧，哦，天哪，杰夫。听起来跟我一模一样啊。
You can use it to make any voice say anything.
你可以用它让任何声音说出任何东西。
Aah! Who -- That -- That's -- Obviously, I didn't -- I didn't say that, right? Like, that's not from an episode of "Post Reports."
啊！谁--那是--显然，我没有--那不是我说的话，对吧？那可不是《华盛顿邮报》的内容。
That's correct. So, I made that with software from a company called ElevenLabs.
是的。我是用一家名为ElevenLabs的公司的软件制作了这个音频。
And I should say here that The Post also uses ElevenLabs to make AI read aloud stories on our website.
我要说一下，《华盛顿邮报》也使用ElevenLabs让AI朗读我们网站上的文章。
But I took it a step further -- I used less than five minutes of audio of your voice hosting "Post Reports" to make an AI Martine voice that can say literally anything, and it can speak in literally any language.
不过我更进了一步--我用了你主持《华盛顿邮报》不到五分钟的声音，制作了一个人工智能马蒂娜的声音，它可以说任何话，而且它可以说任何语言。
I'm Martine Powers. Je suis Martine Powers. ...Martine Powers. How about Japanese?
我是玛蒂娜·鲍尔斯。我是玛蒂娜·鲍尔斯。...玛蒂娜·鲍尔斯。来句日语怎么样？
The AI is sending you a message, Martine. I'm coming for your job.
AI在向你传达一个信息，玛蒂娜。我可要抢走你的饭碗了。
Okay, just kidding.
好吧，开个玩笑。
I -- This is like -- This is like 10% funny, and this is 90%, like, egregiously scary and unsettling.
我--这就像--这10%是好笑，90%是可怕和令人不安啊。
What? I know my own voice pretty well, and that sounds quite -- just like me.
什么？我很熟悉自己的声音，它听起来--就跟我一样。
Yeah, that's what I want to talk with you about today, because it's not just your voice that AI is able to mimic.
是的，这就是我今天想和你们谈的，因为人工智能够模仿的不仅仅是你的声音。
It's gotten really good at doing everything that goes into making a podcast. It's good at mimicking a casual conversation.
它真的很擅长制作播客所需的一切。它擅长模仿闲谈对话。
It can even re-create some of the nuances of speech that make a voice sound human.
它甚至可以重新创造一些声音上的细微差别，让它听起来更像人类。
Ah! I just -- I know we have to get on with this interview, but I just -- I can't get over this.
阿!我只是--我知道我们这次采访还得继续，但我--我无法接受。
I mean, it's amazing, but it's also upsetting.
这太神奇了，但也让人不安。
I think that's the theme of our conversation today -- amazing, but also a little bit upsetting.
这就是我们今天谈话的主题--神奇，但也有点令人不安。
From the newsroom of The Washington Post, this is "Post Reports."
这里是《华盛顿邮报》新闻编辑部的《华盛顿邮报报道》。
Today, many of us are grappling with both fascination and fear about the potential of AI.
如今，我们中的许多人都在努力应对对人工智能潜力的迷恋和恐惧。
Just this morning, a group of 10,000 artists, actors, and musicians, including Julianne Moore, Kevin Bacon, Rosie O'Donnell -- They signed an open letter condemning the unlicensed use of their work to develop AI.
就在今天上午，包括朱莉安·摩尔、凯文·培根、罗西·奥唐奈在内的1万名艺术家、演员和音乐家签署了一封公开信，谴责未经许可使用他们的作品开发人工智能。
At the same time, though, tech companies are still moving full speed ahead on AI.
不过，与此同时，科技公司仍在全速推进人工智能。
That includes making the most human-sounding artificial voices possible.
包括尽可能地制造出听起来最像人的人工声音。
And so I hopped on a Zoom call with Geoffrey in San Francisco.
我在旧金山与杰弗里进行了一次Zoom通话。
I wanted to talk to him about how tech companies are trying to capture the human voice with AI and about the gap that still exists between a machine-generated voice and a real one like mine.
我想和他谈谈科技公司如何试图用人工智能捕捉人类的声音，以及机器生成的声音和像我这样的真实声音之间仍然存在的差距。
