customers
All customer stories
Top Voice AI companies are building with Assembly.
resources
Latest Release
Text-to-Speech API
Our first text-to-speech model, AssemblyAI-TTS-1, is built for voice agents, on live calls. It sounds natural, responds fast, and keeps callers on the line.
resources
In healthcare, transcription accuracy isn't a product feature — it's a patient safety requirement. A misheard drug name or dosage is a clinical risk, not a UX annoyance. Here's how to build ambient scribes, medical intake, and clinical voice agents on infrastructure built for it.
Medical error rate · Lower is better
Medical entity error rate across providers, lower is better. Source: AssemblyAI benchmarks.
Fewer entity errors
87%
Medical Mode reduces medical entity errors ~87% versus the base model — the headline accuracy gain on clinical terminology.
Source: AssemblyAI benchmarks
Medical error rate
3.2%
Medical Mode Missed Entity Rate across specialties — oncology, cardiology, primary care — without retraining.
Source: AssemblyAI benchmarks
Compliance
BAA
Business Associate Agreement available at no additional cost, alongside SOC 2 Type 2, ISO 27001:2022, and PCI DSS v4.0.
Source: BAA FAQ
Self-hosted
48
Concurrent sessions per container for teams that need complete data isolation (Docker + NVIDIA GPU).
Source: Self-hosted deployment
Customer story · Report 3
“We've integrated the newest models from AssemblyAI for pre-recorded audio ASR in our ambient product, and it's been excellent. We're now exploring Universal-3.5 Pro for async and realtime speech-to-text capabilities for new use cases. What's been just as important is the reliability of the platform itself—both technically and in terms of partnership.”
— Gautam Pradeep, Tech Lead, Commure
A different standard
Healthcare audio is categorically harder: far-field ambient capture across the room, overlapping speakers (provider, patient, staff, family), dense terminology (drug names, procedures, dosages), and the highest-consequence failure mode of any voice AI application. Generic voice platforms weren't built for this — and "good enough" transcription creates clinical documentation that's actively dangerous.
Medical Mode optimizes transcription for medication names, procedures, conditions, and dosages — 3.2% MER, ahead of every general and medical model tested. $0.15/hr add-on.
Medical ModeSpeaker role identification (NURSE, PATIENT, PROVIDER) plus SpeakerRevision for provider/patient/family disambiguation in multi-party encounters.
Universal-3.6 Pro RealtimeAutomatic PHI redaction (text and audio), keyterms prompting for patient-specific drug lists, and data training opted out by default. Streaming PHI redaction is nearing release.
Medical ModeCompliance
AssemblyAI offers a Business Associate Agreement (BAA) for organizations subject to HIPAA, included at no additional cost. SOC 2 Type 2, ISO 27001:2022, and PCI DSS v4.0 certified, with data training opted out by default.
For complete data isolation, a self-hosted deployment runs as Docker containers with NVIDIA GPU support — up to 48 concurrent sessions per container.
Certifications & controls
The paths
Both run on the same Universal-3.6 Pro Realtime foundation with Medical Mode — add LLM Gateway and the Text-to-Speech API for a front-office agent, or stream the exam room into an ambient scribe.
Recommended
Patient intake, triage, scheduling, front-office
$0.45 /hr
+ $0.15/hr Medical Mode · Text-to-Speech $15 per 1M characters
Bring your own stack
The ambient scribe architecture
$0.45 /hr
+ $0.15/hr Medical Mode add-on
Proof
Specialized infrastructure lets clinical teams focus on workflow, not a general-purpose speech pipeline.
Healthcare readiness
JotPsych (behavioral health documentation). “In the medical context, accuracy is highly important, and there can be multiple people present — separating them is key. The biggest impact AssemblyAI has had is enabling our team to focus on workflow features rather than a general speech-to-text pipeline.”
Careship. 36% improvement in WER, turning qualitative research into better healthcare experiences across Europe.
Works across specialties. Oncology, cardiology, primary care — no retraining.
Structured, EHR-ready output. Diarized, PHI-redacted transcripts feed SOAP-note generation.
Get your API key, or talk to us about a BAA and self-hosted deployment.
AssemblyAI's Medical Mode posts a 3.2% Missed Entity Rate — the lowest across benchmarked providers (Deepgram, Speechmatics, AWS, Google) — and is purpose-built for medication names, procedures, conditions, and dosages. It activates with a single parameter (domain: "medical-v1") on both pre-recorded and streaming models and works across specialties without retraining.
Stream exam-room audio to Universal-3.6 Pro Realtime with Medical Mode, speaker diarization, and PHI redaction enabled, then pipe the diarized, structured transcript into your SOAP-note generator and EHR. voice_focus (far_field) handles ambient room capture and SpeakerRevision disambiguates provider, patient, and family. Pricing is $0.45/hr base plus a $0.15/hr Medical Mode add-on.
Yes. Build intake and triage agents on Universal-3.6 Pro Realtime with Medical Mode, LLM Gateway, and the Text-to-Speech API. Start from the LiveKit or Pipecat reference implementation, which wires all three together on one API key and one bill, and let your orchestrator handle tool calls for EHR lookups and scheduling. Text-to-Speech starts speaking in about 90 ms and reads dates, times, and confirmation codes exactly as written.
On medical entity accuracy, Medical Mode posts a 3.2% Missed Entity Rate versus 4.7% for Deepgram Nova-3 Medical — and leads the field against Speechmatics Enhanced Medical (3.6%), AWS Transcribe Medical (8.7%), and Google Medical Conversation (24.4%) as well. See the full benchmark table for methodology.
AssemblyAI is considered a business associate under HIPAA and offers a standard Business Associate Agreement (BAA) — at no additional cost — for customers processing protected health information (PHI). It also provides PHI redaction across audio and transcripts, SOC 2 Type 2, ISO 27001:2022, and PCI DSS v4.0, plus a self-hosted deployment option for complete data isolation.
Medical Mode is a $0.15/hr add-on on top of the model you're using — on Universal-3.6 Pro Realtime streaming ($0.45/hr), that's $0.60/hr in total. Voice agents that talk back also use the Text-to-Speech API, billed at $15 per 1M input characters.
