Voice transcription and sign language recognition

[pdufour]

Hi, thanks for open sourcing this library. I’ve developed two new features I wanted to run by the team.

FYI this post is not for feature requests, I’ve already fully developed them with working code and am looking for feedback :). I’ll also be at the Codex After Hours event tonight if you want to brainstorm IRL.

1. Voice transcription using local LLM - fully working PR here - Add local LLM voice model transcription pdufour/codex#1

My thoughts here are to have an option similar to the ollama option but instead be I.e. —voice-model: lets you download and use a STT model from HF. Use cases would include you have spotty internet ( i.e. airplane) or you maxed out your token usage.

2. Sign language recognition using local CNN model, PR - Add support for signing detection using yolo11 model pdufour/codex#2

Right now it’s a local model but a hosted version could co-exist. It works quite well at detecting letters using a YOLO model. Future enhancements include full sentence comprehension.

Looking for any and all feedback! Cheers!