Buzz and Whisper: local transcription setup and tutorial

This tutorial introduces Buzz, an open source desktop application for audio and video transcription with Whisper. It covers installation on Windows, macOS and Linux, then explains how to download a model for offline use. The speaker presents local processing as a way to keep sensitive recordings on your computer and describes the application as free, without subscriptions or minute limits.

Model choice affects speed, resource use and transcription quality. The speaker finds Tiny and Base fast but prone to errors, describes Small and Medium as a practical compromise, and recommends trying Turbo on your own hardware. These are the speaker's assessments rather than accuracy guarantees.

The file walkthrough covers model selection, automatic or manual language selection, word-level timings, and automatic export to plain text or timestamped subtitles. The speaker reports that audio preprocessing worked with a file but failed with a link. Finished transcripts can be viewed as text, copied or downloaded.

Buzz also accepts YouTube links and supports live transcription. In the demonstrated link test, a 5.5-minute video took 10.5 minutes to process, with some errors in English names. Live transcription showed substantial lag with Large Turbo; a lighter model worked with a delay. Processing time depends on the device, and models must be downloaded before local use.