Not every AI job needs a subscription or an API bill. A recent laptop can run good open models on its own, offline, for free. We now do two big jobs this way, and only send the hard thinking to a paid AI.
Job 1: transcribe hours of audio and video
- What: a free speech-to-text model (OpenAI's open Whisper, run locally) turns every video or audio file into a text file and a subtitle file.
- Languages: it handles English and Arabic well, and many others.
- Speed we measured: about 6 seconds for a 30-second reel on a recent Apple laptop, so roughly 12 minutes per hour of audio.
- Cost: zero. Nothing leaves your computer, which also helps with privacy.