implement keyword boosting #3512
Description
Many speech-to-text providers already support keyword boosting or custom vocabularies. Our responsibility is to expose this capability in a thoughtful, ergonomic way.
The core problem: when users correct mistranscriptions (proper nouns, jargon, company names, etc.), the system should remember and improve over time.
Possible approaches
- Implicit learning from edits: When a user writes or corrects a proper noun or domain-specific term (especially ones that were previously mistranscribed but fixed during summarization), the system should remember this correction.
- Learning from transcript / summary edits: When users manually edit transcripts or summaries, those corrections should feed back into future transcription accuracy.
- Global dictionary / vocabulary: Consider maintaining a user-level (or workspace-level) custom dictionary that is automatically used for keyword boosting in future transcriptions.
Open questions
- How do we scope memory (note-level vs user-level vs workspace-level)?
- How explicit should this be in the UI vs fully automatic?
- How do we surface and manage learned keywords over time?
Related: #3474