This is a quick demonstration of using Orukeet as a speech recognition model inside Emacs. I'm recording this in real time on a Lenovo P52 laptop using only CPU. I can replay just one part of my recording. I can switch languages in the middle of recording too. Par exemple, maintenant je parle français. Seems promising.

Links/notes:

• oruk/orukeet · Hugging Face
• file:///home/sacha/vendor/orukeet/orukeet-server.py - I generated this with Google Gemini for a quick POC. It keeps the model in memory for slightly faster processing per segment.
• Run Orukeet locally: Python speech recognition | oruk
  • https://oruk.ai/examples/orukeet-local.py - Python script you can run from the command line to transcribe files.
• file:///home/sacha/vendor/orukeet/orukeet-local.py
• sachac/whisper.el - my fork of natrys/whisper.el
• sacha-whisper-continue - This lets me cue transcriptions of previous segments while continuing to record new ones. I like it because I tend to pause while thinking out loud. I can define the chunk that will get transcribed so that I don't end up with incomplete chunks or lost context.