The author explores the use of speech as an input mechanism for computers, acknowledging its maturity for drafting ideas and taking notes, but not so much for structured writing. They propose augmenting transcription tools with LLMs to enable real-time edits, creating a package for Emacs that corrects spoken words in real-time. They highlight the potential for future improvements in latency and transitioning to lightweight on-device alternatives. While the package is open-source, it currently relies on Deepgram and OpenAI’s services. The author intends to resolve minor bugs and enhance user experience in the future. The launch of Aqua Voice sparked their exploration into this project after months of delay.
https://lepisma.xyz/2024/09/12/emacs-dictation-mode/index.html