Show HN: Ghost Pepper – Local hold-to-talk speech-to-text for macOS

Posted by MattHart88 1 day ago

Show HN: Ghost Pepper – Local hold-to-talk speech-to-text for macOS(github.com)

I built this because I wanted to see how far I could get with a voice-to-text app that used 100% local models so no data left my computer. I've been using a ton for coding and emails. Experimenting with using it as a voice interface for my other agents too. 100% open-source MIT license, would love feedback, PRs, and ideas on where to take it.

442 points | 193 commentspage 7

eddie-wang 13 hours ago|

[dead]

Sukhbat 14 hours ago||

[flagged]

Ecko123 22 hours ago||

[dead]

lostathome 23 hours ago||

[dead]

romeroej 22 hours ago|

always mac. when windows? why can you just make things multios

SquareWheel 16 hours ago||

Windows has a native (cloud-based) dictation software built-in[1], so there's likely less demand for it. Nonetheless, there are still a handful of community options available to choose from.

[1] https://support.microsoft.com/en-us/windows/use-voice-typing...

cootsnuck 20 hours ago|||

Handy has Windows support. https://handy.computer/

patja 20 hours ago|||

I've been using Chirp which uses parakeet on Windows. Learned about it here:

https://news.ycombinator.com/item?id=45930659

Works great for me!

naikrovek 20 hours ago||

Because like all other modern Macs, the GPU in my Mac uses the same API as the GPU in your Mac.

Also, on a Mac with 32GB of RAM, 24GB of that (75%) is available to the GPU, and that makes the models run much faster. On my 64GB MacBook Pro, 48GB is available to the GPU. Have you priced an nvidia GPU with 48GB of RAM? It’s simply cheaper to do this on Macs.

Macs are just better for getting started with this kind of thing.

patja 20 hours ago||

Fair enough for GPU-intensive stuff like running Qwen locally. But do you really need a GPU for decent local TTS? I run parakeet just on CPU.