Currently, the transcript appears as a solid block of text, which leads to attribution errors, for example, when I say something about myself, the note says "Pt states __".

To improve note accuracy, could you introduce speaker diarization and voice recognition? Specifically:

Speaker Separation: Label speakers dynamically (e.g., Speaker 1, Speaker 2) with the ability to rename speaker tags, similar to platforms like Otter.

Voice Profile Learning: Allow Heidi to learn/recognize the clinician's voice profile to automatically differentiate clinician vs. patient dialogue and prevent cross-attribution errors.

Currently, Heidi's transcript doesn't distinguish between speakers, which can lead to misattribution errors (e.g., attributing the mother's medical history to the patient). The request is for Heidi to assign voices to speakers (using letters or numbers) and track who is saying what, creating a script-like transcript. Over time, the model should learn the provider's voice to better distinguish it from others in the room. This would decrease misattribution errors and reduce the need for editing, preventing critical errors in medical records.