florian/peggypad
Watch
1
0
Fork
You've already forked peggypad
 0
Code Issues Pull requests Projects Releases 11 Packages Wiki Activity Actions

Use participants and notes as transcription hints; clearer notes label #9

Merged
florian merged 1 commit from transcription-hints into main 2026-10-05 15:43:10 +02:00
Conversation 0 Commits 1 Files changed 9 +113 -8
florian commented 2026-10-05 15:43:06 +02:00
Owner
Copy link

Names and terms from a recording's Participants and Notes go to the
speech-to-text service as spelling hints: Mistral's context_bias,
Deepgram's keyterm, ElevenLabs' keyterms (one form field or query
parameter per term), or an OpenAI-style prompt when not diarizing.
Pieces are split at commas, semicolons, colons, line breaks and
parentheses; only term-like ones (at most 5 words, 50 characters) are
kept, up to 100. Hints are best effort: a 400/422 with hints is retried
once without them. The transcribe confirmation says when hints are sent.

The summary request now tells the model to use the notes to interpret
the transcript and spell names and terms, but to report only what was
said, instead of calling them "background only". The Notes placeholder
suggests agenda, topics, names and terms.

Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com

Names and terms from a recording's Participants and Notes go to the speech-to-text service as spelling hints: Mistral's context_bias, Deepgram's keyterm, ElevenLabs' keyterms (one form field or query parameter per term), or an OpenAI-style prompt when not diarizing. Pieces are split at commas, semicolons, colons, line breaks and parentheses; only term-like ones (at most 5 words, 50 characters) are kept, up to 100. Hints are best effort: a 400/422 with hints is retried once without them. The transcribe confirmation says when hints are sent. The summary request now tells the model to use the notes to interpret the transcript and spell names and terms, but to report only what was said, instead of calling them "background only". The Notes placeholder suggests agenda, topics, names and terms. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Names and terms from a recording's Participants and Notes go to the
speech-to-text service as spelling hints: Mistral's context_bias,
Deepgram's keyterm, ElevenLabs' keyterms (one form field or query
parameter per term), or an OpenAI-style prompt when not diarizing.
Pieces are split at commas, semicolons, colons, line breaks and
parentheses; only term-like ones (at most 5 words, 50 characters) are
kept, up to 100. Hints are best effort: a 400/422 with hints is retried
once without them. The transcribe confirmation says when hints are sent.

The summary request now tells the model to use the notes to interpret
the transcript and spell names and terms, but to report only what was
said, instead of calling them "background only". The Notes placeholder
suggests agenda, topics, names and terms.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
florian deleted branch transcription-hints 2026-10-05 15:43:10 +02:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
florian/peggypad!9
No description provided.