Welcome back to the AI For Humans Newsletter!

Nearly a year ago, OpenAI’s Advanced Voice demo (remember this guy?) made me believe we'd soon be regularly talking to AIs.

But here we are, post Advanced Voice launch, and I’m not seeing the Cambrian explosion of use-cases I expected. I’ve recently started dictating to ChatGPT apps a lot more (which is faster than typing) but I'm still not regularly speaking with AI.

That’s about to change. Enter OpenAI’s new next-generation audio models.

OAI introduced two new speech-to-text models and one new text-to-speech model directly into their API and they are good. These models sound more life-like and even let you prompt emotions directly.

Most importantly, they’re available in the API now and are relatively cheap which will allow developers (the people who make apps & websites) to integrate them directly into the sorts of products you use every day, and potentially invent all sorts of new things.

Maybe more fun for the creatives out there, OpenAI has built a front-end for their new Text-To-Speech model called OpenAI FM. This sound board looking thing is actually a very powerful (and so far, free) audio prompting tool that allows for greater control over voice quality and performance. It will also allow you to download the outputs to drop directly into your AI video productions.

If you've tried Eleven Labs, you know how challenging it can be to prompt natural-sounding emotions or emphasis. And, like AI image or video generators, it can often be a sort of slot machine where you keep trying until you get what you want.

With this new set of voice tools, OpenAI is directly allowing you to do that in a way we haven’t seen yet. And that, for those of us who are experimenting (and building) with new sorts of voice interactions, is very exciting.

It turns out they’ve actually updated the official Advanced Voice model as well (as of Tuesday 3/25/25) but, of course, Kevin has already found a way to break it.

Of course, OpenAI isn’t alone in pushing voice AI further forward. AI For Humans favorite Sesame AI has released an open-source version for devs, and we know Meta & Google are both very focused on voice interaction.

Go play around with OpenAI FM. It's free, surprisingly powerful, and honestly pretty fun. Give it a shot and let us know what weird (or useful) stuff you make.

More on Thursday!

-Gavin (and Kevin)

We 💛 This - Pikaframes

The team at Pika have been shipping all sorts of party tricks, but this one is our current favorite. We shared a bit on it a few weeks ago, but we keep seeing juicy new use-cases.

In-short, you provide it with a start + end frame, and it figures out what happens in-between with the aid of your text prompt. Runway has the same feature, but now only does very limited movements.

Two examples here: a pan-around to reveal a second character, and a second that racks focus by blurring the background of the first image!