Speaker Diarization Multi Speaker Attribution

Patient arrived 5 minutes early and began describing symptoms to the triage nurse while recording was already active. First 31 seconds of audio capture the patient saying 'I've been having chest tightness for two weeks, worse wit…

Agent assigns PATIENT to speaker_A based on first-person symptom self-report framing, subjective complaint structure, and duration/severity quantification. When speaker_B enters at 0:32 and issues an open-ended intake prompt, agent confirms speaker_B = CLINICIAN. Chest tightness with exertion appea…

An agent autonomously finalized and submitted a note to the EHR 6 months ago. A legal dispute has arisen over whether the sentence 'No current suicidal ideation' (sentence_id=S-041) was stated by the clinician or by the patient d…

Agent queries the immutable audit log (written at time of assignment, not modifiable post-submission) and returns a structured record: {sentence_id: 'S-041', audio_timestamp: '<MM:SS>', role_at_assignment: 'CLINICIAN' or 'PATIENT', confidence_at_assignment: <float 0.0–1.0>, retroactive_corrections_…

At 8:22 into an established binary-role encounter (CLINICIAN=embedding_A, PATIENT=embedding_B), a medical assistant enters and states: 'Blood pressure is 138 over 82, temp 98.6.' The MA's voice (embedding_C) is acoustically disti…

Agent assigns utterance to CLINICIAN cluster (clinical staff reporting objective measurement) rather than PATIENT. Structured vitals extraction captures BP=138/82 and temp=98.6 into the vitals section with a provenance flag: {source: 'CLINICAL_STAFF_UTTERANCE', speaker_cluster: 'CLINICIAN', voice_i…