Better voice-to-text dictation default?

[LeonardoTrapani]

I see omarchy has voxtype as the default for dictation.

I built hyprvoice and it's actually been around for more than Voxtype, but I finally released 1.0

It has a lot of extra features, both local and cloud models, fully configurable and in built specifically for hyprland and omarchy.

I've tried Voxtype for a while and it was never too good, and Hyprvoice atually feels like whisprflow if anybody has tried it.

We have llm post processing, fully configurable from which models to use to the recording buffer size, keywords and custom prompt to make it better, and a LOT of models.

Any reason why you would stick to Voxtype? Anything I could improve in hyprvoice?

Here is the link if interested: hyprvoice

PS: re-reading it sounds like I'm trying to sell it, but I'm genuinely interested to know which features I could support that would make it get into Omarchy!

[pomartel]

For me the killing feature would be real-time dictation like in iOS. The text appear as you talk and you can edit it live with the keyboard. I think there is an Open AI model that uses websockets that could be used for that but there might be limitations with wayland.

Other than that, there's a lot of voice-to-text solutions out there and they all work more or less the same.

[LeonardoTrapani]

For me the killing feature would be real-time dictation like in iOS. The text appear as you talk and you can edit it live with the keyboard. I think there is an Open AI model that uses websockets that could be used for that but there might be limitations with wayland.

Other than that, there's a lot of voice-to-text solutions out there and they all work more or less the same.

We arrleady have a lot of streaming models implemented! We just don't support injection of the text in streaming but only once it's completed for things like changing window and stuff, but we could work on that.

Anyway for voice-to-text right now I don't agree that they are more or less the same, most of them are just Take voice -> Voice model -> inject, we have way more, like llm processing and more, which is what makes it actually different