Find Federico on MastodonFind John on MastodonFind Alex on MastodonFind Club MacStories on Mastodon

MassReplaceIt

MACSTORIES RECOMMENDS

Great apps, accessories, gear, and media recommended by the MacStories team.

MassReplaceIt

I’ve written about my exploration of Whisper by OpenAI for converting spoken audio into text before. The transcriptions are good. In fact, they’re better than any other transcription app I’ve tried, but it still makes too many mistakes to create a publishable transcript. As I wrote in the Monthly Log, it’s this last 10% of cleanup and other steps necessary to create a presentable and usable transcript that is a bigger hurdle than the speed of running Whisper on a Mac’s CPUs, which is slow.

What’s nice about computers, though, is that the mistakes they make are consistent. For example, if Whisper trips over my name or calls Mastodon ‘Master Don,’ it’s likely that it will do so again in the future. As I created a few test transcripts, attempting to reduce the production time as much as possible, I began to notice Whisper’s most common mistakes and make a list of them.

This story is for Club MacStories, Club MacStories+, and Club Premier members only.

Join the Club and get access now.

Already a member? Sign in