2h ago
pass speaker labels to custom mode ai processing
when "identify speakers" is enabled, superwhisper separates speakers correctly, and i can see the labeled version in the history segments tab. but the transcript sent to ai processing in a custom mode doesn't include those labels. the raw result and the processed result both come through as one block of text with no speaker info
this means a custom prompt can't do anything speaker-aware, like formatting output as "[speaker 1]: ..." / "[speaker 2]: ...", summarizing by speaker, or keeping someone else's audio (a video or call) separate from my own dictation. the model only sees plain text, so it has no way to know the speaker changed
the current workaround is to record in a voice to text mode, copy the transcript from the segments tab, paste it into an editor, then run a second custom mode with application context. it works, but it takes several manual steps for something that could happen in one recording
what would help:
when "identify speakers" is on, send the speaker-labeled transcript to the ai processing step (for example "speaker 1: ... speaker 2: ...")
or add a toggle like "include speaker labels in ai input" so people can choose
bonus: a template variable for the speaker-labeled transcript in custom prompts, so it can be referenced directly
this would make speaker separation work end to end in a single mode, for meetings, interviews, and recordings with more than one voice
Pending