Fix Whisper's misspelled names with a glossary

My product name and coworkers' names, spelled a different way in every transcript.

Beginner to intermediateRead 5 minDoing: about an hour6 steps

What you'll have at the end

The same recording transcribed twice, once with no glossary and once with one, with a side-by-side check proving every name now matches how it's really spelled.

Before you start

You need

You can already send an audio file to a transcription endpoint and get back its plain-text transcript.

Not covered

A word the model never caught at all because of background noise or a mumbled recording; a glossary only changes which spelling it picks for a word it already heard clearly.

Leans on

6 steps

Where to go next

Long silence at the start throws off Whisper's transcript

when the WHOLE transcript reads garbled from the first few words, not one name misspelled among many, dead air at the start of the recording is the more likely cause, and that needs a different fix than a glossary.

Checked 15 Aug 2026

Part of the Multimodal AI cookbook