Can it keep up with fast emergency-department speech?

Yes — Universal-3.6 Pro Realtime streams transcripts during the encounter while holding a 3.2% Missed Entity Rate.

How does it handle overlapping voices and noise?

Universal-3.6 Pro Realtime is built for real-world audio; add speaker_labels to separate speakers.

Does it capture critical medications and doses?

Yes — that entity class is exactly what Medical Mode improves; add specifics via keyterms.

Can it transcribe a triage handoff or shift-change report?

Yes. Handoffs are dense with medications, doses, lab values, and scoring instruments (HEART, GCS, trop) — the entity classes Medical Mode is tuned for. Stream the report live or transcribe a recording, and pass your department's protocol vocabulary as keyterms.

How does far-field capture behave in a curtained trauma bay?

Set voice_focus to far-field on the streaming session to isolate the primary speaker and suppress the bay next door. For noisy, overlapping rooms, the max_accuracy mode trades a little latency for accuracy — the right default when the audio is hard rather than the turn-taking is fast.

Does it hold up on code and critical-medication capture under pressure?

Medication names and dosages are the entity class Medical Mode improves most, and keyterms let you weight your code-cart formulary explicitly. Note that no ASR should be the only record of a code — treat the transcript as a draft a clinician signs off on.

Is emergency-department PHI protected and BAA-eligible?

AssemblyAI is a business associate under HIPAA and signs a standard BAA; PHI redaction; SOC 2 Type 2.

How much does real-time ED transcription cost?

$0.15/hr Medical Mode add-on on the realtime model.

How do I enable Medical Mode for emergency medicine?

Add domain: "medical-v1" to the realtime session.

Emergency medicine

Fast-paced ED audio, captured accurately

Emergency-department transcription converts fast, overlapping ED speech into an accurate clinical record in real time. AssemblyAI's Medical Mode, a one-parameter add-on to Universal-3.5 Pro, holds a 3.2% Missed Entity Rate on clinical conversations — the lowest across benchmarked providers — on trauma-bay, triage, and shift-change audio. BAA available.

Emergency documentation is the hardest case: rapid speech, overlapping voices, critical drug names and doses, no time to correct. Universal-3.5 Pro with Medical Mode holds a 3.2% Missed Entity Rate under that pressure, in real time, so the record keeps pace with the room.

Try Medical Mode free on your audio Read the docs

Speed and accuracy at the same time — where ED documentation usually forces a trade-off

Missed Entity Rate 3.2%

Missed Entity Rate on clinical conversations with Medical Mode.

Fewer missed entities ~20%

Fewer missed medical entities than the base model without Medical Mode.

Fewer entity errors 87%

Fewer medical entity errors than the base model.

Benchmarked providers #1

Lowest Missed Entity Rate vs. Deepgram, Speechmatics, AWS, and Google.

Benchmarks

Emergency department transcription accuracy vs. other providers

Missed Entity Rate is the share of clinical entities — medication names, dosages, diagnoses — that a model fails to transcribe correctly. Lower is better.

Missed Entity Rate and Word Error Rate on clinical conversation audio Lower is better. Medical Mode enabled via `domain: "medical-v1"`.
Provider	Missed Entity Rate	Word Error Rate
AssemblyAI Universal-3.5 Pro w/ Medical Mode	3.2%	5.3%
Speechmatics Enhanced Medical	3.6%	5.5%
Deepgram Nova-3 Medical	4.7%	6.1%
AWS Transcribe Medical	8.7%	5.9%
Google Medical Conversation	24.4%	12.9%

See the full benchmarks

Activation

Real-time capture under pressure

Stream the room's audio and keep the record in step with a fast ED — one WebSocket, one parameter, no model swap.

from assemblyai.streaming.v3 import StreamingClient, StreamingParameters

client = StreamingClient(api_key="YOUR_API_KEY")

# Emergency medicine: rapid speech, critical meds/doses
client.connect(StreamingParameters(
    speech_model="universal-3-6-pro",
    domain="medical-v1",
    keyterms_prompt=["epinephrine", "1 mg", "tachycardia", "STEMI"],
))

Worked example

Triage handoffs and shift-change documentation

ED speech is fast, overlapping, and often two conversations deep. Medical Mode holds the medication names, dosages, lab values, and scoring instruments a handoff turns on — captured live, in a curtained bay, at far-field distance.

Shift-change handoff -> structured summary

Bed 4, 68M, chest pain onset 0340, EMS gave 324 ASA
and 0.4 SL nitro x2. Trop 0.04 -> repeat pending at
0600. EKG: NSR, no ST elevation. HEART score 4.
Cards consulted, awaiting callback.

Bed 7, 34F, MVC restrained driver. FAST negative.
CT C-spine and abd/pelvis pending. 2 units T&C held.

Captured under trauma-bay noise, overlapping speech,
and a curtained bay - drug names, dosages, lab values,
and scores all have to land on the first pass.

Critical-med accuracy

Emergency drugs and doses (epinephrine, 1mg) captured where errors are most consequential.

Universal-3.5 Pro

Low latency, no trade-off

Streaming transcription keeps documentation in step with a fast room.

Universal-3.6 Pro Realtime

Robust on chaotic audio

Overlapping speech and trauma-bay noise handled without fabricated text.

Medical Mode

Common questions

: Yes — Universal-3.6 Pro Realtime streams transcripts during the encounter while holding a 3.2% Missed Entity Rate.
: Universal-3.6 Pro Realtime is built for real-world audio; add speaker_labels to separate speakers.
: Yes — that entity class is exactly what Medical Mode improves; add specifics via keyterms.
: Yes. Handoffs are dense with medications, doses, lab values, and scoring instruments (HEART, GCS, trop) — the entity classes Medical Mode is tuned for. Stream the report live or transcribe a recording, and pass your department's protocol vocabulary as keyterms.
: Set voice_focus to far-field on the streaming session to isolate the primary speaker and suppress the bay next door. For noisy, overlapping rooms, the max_accuracy mode trades a little latency for accuracy — the right default when the audio is hard rather than the turn-taking is fast.
: Medication names and dosages are the entity class Medical Mode improves most, and keyterms let you weight your code-cart formulary explicitly. Note that no ASR should be the only record of a code — treat the transcript as a draft a clinician signs off on.
: AssemblyAI is a business associate under HIPAA and signs a standard BAA; PHI redaction; SOC 2 Type 2.
: $0.15/hr Medical Mode add-on on the realtime model.
: Add domain: "medical-v1" to the realtime session.