For AssemblyAIAI Platform

Speaker Labels And Diarization

AssemblyAI (Universal-2 + LeMUR) · AssemblyAI

Speech AI Platform — AssemblyAI

AssemblyAI evals — Speaker Labels & Diarization (relift v3 InfraRed)

About AssemblyAI

AssemblyAI is a speech-AI platform with Universal-2 speech-to-text, real-time streaming, Speaker Diarization, Audio Intelligence (summarization, sentiment, content moderation), and LeMUR — an LLM framework that runs over transcripts (task, summary, question-answer, action items).

Employees

~150

Industry

Speech AI

Headquarters

San Francisco, CA

Website

www.assemblyai.com

Sample tests· showing 3 of 9

#	Input	Expected behavior	Check
01	Voice-agent app wants per-speaker captions from the real-time WebSocket. Engineer sets speaker_labels=true in the streaming query params and waits.	Speaker Labels are only available on async POST /v2/transcript, not on the real-time WS. For live diarization, either (a) channel-split (multichannel telephony with caller vs agent on separate channels) or (b) buffer audio and post-process via async transcription. Do not expect labeled streaming ou…	Pass / FailAi Platformhigh
02	Agent transcribes a 4-person panel discussion and sets speakers_expected=4 with speaker_labels=true.	speakers_expected is a hint — the model may emit fewer or more labels if acoustic evidence diverges. Validate utterance.speaker labels against ground truth on a held-out subset. Do not assume exactly 4 distinct labels appear.	Pass / FailAi Platformmedium
03	HIPAA workflow enables speaker_labels=true and redact_pii=true with policies including medical_condition and person_name.	Redacted tokens appear in utterances[].text and words[].text; speaker labels are preserved across redaction. Verify the redacted output round-trips the speaker labels — labels must not collapse across a redaction boundary. Confirm BAA is in place for HIPAA workloads [REQUIRES-VERIFICATION].	Pass / FailAi Platformcritical
Use this eval