Granola does dictation, Wispr Flow does meetings

So, the meeting notetaker now does dictation, and the dictation app now does meetings. Between them, Granola and Wispr Flow are valued at three and a half billion dollars, and each spent a chunk of this year building the other one's product.

We've had both features in talat for a while: meetings since March, and dictation since June. It goes to show how far AI coding assistants have come: these days even a large, well-funded company can ship a new feature.

What it is

Granola launched what they're calling "voice typing" on 29 October. Hold a key, talk, and the words appear where your cursor is, in any app. It's free on every plan with no limit, it's in beta, and it's Mac only for now, with Windows "coming soon".

Your voice doesn't actually go to Granola. It goes to AssemblyAI, a speech-to-text company, which transcribes it and tidies it up before Granola pastes it in, meaning a second company handles your voice every time you press the key. Granola is open about this, and says its providers don't train on your dictations. For the record, I've used AssemblyAI a lot in the past and think they're great. If you need bleeding-edge cloud transcription, they're well worth a look.

Two things worth knowing

When you start dictating, Granola picks up the text around your cursor and other text visible nearby, and sends it to AssemblyAI to help spell names correctly. It's a good idea, it skips password fields, and it's on by default. You can turn it off in settings.

For half a minute after it types, Granola keeps an eye on whatever you dictated into, and if you fix a name it got wrong, it sends the correction and the words around it to Granola to learn from. Also on by default; turning off the screen reading turns this off too.

The function key turf war

Inside Granola is a list of rival dictation apps, Wispr Flow among them, each with a link straight to that app's settings so you can switch off its shortcut. Wispr, meanwhile, has a help article explaining that Granola's voice typing "can interfere with Wispr Flow dictation", and suggesting you switch Granola's off.

talat isn't on Granola's list, and Wispr hasn't written a help article about us. We're not taking that personally.

And Wispr's meetings

Wispr Notetaker came out in August, and I missed the boat on writing about it at the time. Like Granola, it records from your computer rather than sending a bot into the call, and like Wispr's dictation, it all happens in the cloud. Instead of learning voices, it gets names by reading the "who's speaking" indicator in Zoom, Teams or Meet. Clever, and brittle: when we tried it, it kept up through one short call and lost track early in another.

Wispr's CEO said at launch: "Nobody reads meeting transcripts. But everyone acts on them." He's right, which is why speaker identification matters. An action point is only useful against the right person.

How talat compares

talat does both, and both run on your Mac or Windows PC. Your voice is turned into text by the same on-device model that transcribes your meetings, and nothing is uploaded anywhere. Tidying up is optional and can run on your machine too. In meetings, talat tells you who said what by learning voices, also on your machine.

If you already use Granola, voice typing is free and worth a try. If you'd rather nothing you say left your computer, in a meeting or in a text box, that's what talat is for. You can try it free for ten hours without an account, and there's a talat and Granola comparison if you want the detail.