A clever hack by George Mandis - Cut Your OpenAI Transcription Costs in Half with This Simple Audio Hack
Want to cut costs on OpenAI’s audio transcription (gpt-4o) without sacrificing quality? George Mandis discovered a simple yet brilliant hack: speed up your audio with ffmpeg before uploading. Running files at 2x or 3x playback speed slashes the minutes (and dollars) charged, while maintaining near-identical transcript quality—ideal for webinars, interviews, and long recordings. Bonus: it also bypasses model limits that choke on longer real-time files. Smart, scrappy optimisation—exactly the kind of engineering mindset we need more of.