Add support for signing detection using yolo11 model

[pdufour]

Description

This adds the ability to detect letters you are signing via ASL in realtime via a YOLO model trained on an ASL image dataset and convert it into text.

i.e. any of these

are converted into their letter equivalents when you type /signletter and do one of the signs

The architecture to the similar to the voice transcription one. It uses hf_hub crate to download models, so that way if they are already cached you can re-use the same cached models.

Everything is contained within the app, meaning you do not need to spin up any other programs. This offers a lot of benefits like easier installs for users and more control of what models are allowed.

Example of what it captures:

Video demo

564455988-1f583bb0-ad3b-45a1-a912-ceb0c894d1af.mp4

Specific Changes

• It adds two commands

/sign-model - allows you to select from different onnx models trained on ASL sign letters - there are some I uploaded here - https://huggingface.co/pdufour/asl-yolo-models-onnx/tree/main. They are downloaded on demand
/signletter - opens webcam and allows you sign letters, the Yolo model that runs locally automatically detects when your signs are and transcribes it into text.

Future Enhancements

• Support all of ASL and just not just lettering
• You could let the user decide between using a local model and a remote hosted one.
• There are also other model architectures that might offer better performance like this transformer architecture model https://www.nature.com/articles/s41598-025-31558-1

Test Plan

Sign letter

• Open codex
• Run /sign-model to select from a Yolo model - the larger ones take more time to download
• Run signletter
• Make a sign and see the it outputs the letter you expecte