Voice notes.

Inside any round (creator or guest), open the note composer and hit hold to record. The mic asks for permission once. While you're holding, a live level meter shows the mic is picking up your voice — emerald bars dance in the spectrum. Release to stop.

Limits: 500ms minimum, 2 minutes maximum.

Once the bytes upload, the row appears in the comment list with "transcribing…" while our worker hands the audio to OpenAI's Whisper (gpt-4o-mini-transcribe). The transcript shows under the audio player within a few seconds of release.

Click the hone button on your own voice note. Cue rewrites the casual transcript in mixing-engineer language ("the piano hurts my ears" → "the piano's bite is harsh; soften the upper-mids"). Cached forever; you can toggle show original any time. See the Hone article.

Replies support voice too. Hit reply on a comment, hold to record, release, save. Same transcription path. Voice replies inherit the parent's anchor like text replies — they don't drop their own marker.