Finally found the local speech-to-text model I have been looking for!
In the latest release of VS Code, I noticed they included a local model (Nemotron) for giving voice commands to your AI agent. After researching it a bit, I discovered the backend uses a very small model that runs entirely locally without heavy lifting.
I learned that Sherpa-ONNX can run these models, so I asked my AI agent to build a simple tool I can trigger with a keyboard shortcut to dictate text directly into my computer. Right now, I am amazed by the results-it is fast and accurate.
After some polish, I plan to release this as a standalone tool. I know everyone "could" build this themselves, but that's still labor you have to put in. Hopefully, releasing it will save a few hours of your time.
Check out the quick demo below to see its speed and accuracy in action.
Demo uses the parakeet-tdt-0.6b-v3-onnx model but nemotron works well too.