Concepts
Languages and glossaries
Set a language per course, and teach Stenny the terms of each course.
Languages
Set language for each course in config.toml. Stenny is used with English, French and German.
| Setting | Behavior | Use it when |
|---|---|---|
"fr" | One fixed language. | The course is always in one language. Most reliable. |
"auto" | One language per lecture, detected from five samples across the recording, chosen among languages. | The language changes from lecture to lecture. |
"mixed" | Detected about every 25 seconds, chosen among languages. | A lecture switches language, or quotes another one. |
[courses."International Economics"] language = "mixed" languages = ["fr", "en", "de"]
With a fixed language, Whisper tends to translate passages in another language instead of transcribing them. Use "mixed" for lectures that switch.
To force a language for a single file, use stenny add recording.m4a --course ie --lang en.
Glossaries
Each course has a glossary note in _Glossaries, like Data Analytics (glossary).md. Whisper gets its terms as a hint, so names and technical terms come out right.
The glossary note
A glossary has three sections. You edit the first and the last.
## Terms - heteroscedasticity - Gauss–Markov ## Terms (added automatically) - ordinary least squares ## Corrections - hetero skedasticity → heteroscedasticity
- Terms are the course's names and technical terms. Whisper's prompt takes them first, then the 15 most used automatic terms, then the current lecture's slide terms, as many as fit.
- Corrections fix mishearings that keep coming back. They're applied word for word to every new clean transcript, never to the raw output. Stenny never adds corrections by itself.
Terms added automatically
The AI step names each lecture's technical terms. A term is kept only if it's on the slides or was transcribed the same way at least twice, so a misheard word can't get in. Kept terms go under Terms (added automatically), sorted by how many of the course's last five lectures use them. Stenny only ever rewrites that section.
A lecture's own terms help from the next lecture on, because Whisper has already run by the time they're found. To keep an automatic term at the top, move it up to Terms.
To apply glossary changes to a lecture that's already done, run stenny refresh.