Local ASR Models

It doesn’t run locally right?

It does, but it’s resource intensive, and is very slow on low end phones.

SenseVoice and Moonshine were specifically made to solve that, but they’re newer and therefore not as well known.

2 Likes

will these be better or faster than the voice to text app?

are these AI?

The other one?

These are all AI ASR[1] models.

There’s also NVIDIA Parakeet, which is better than all of them, but is too heavy for most phones, and better suited for running on computers.


  1. Automatic speech recognition ↩︎

It sounds like there’s another that I missed, NVIDIA Canary 180M Flash.

Heavier than both SenseVoise and Moonshine, but more accurate for English and more powerful.