What's the best speech-to-text API for transcribing therapy sessions?

One that gets psychiatric medication names and clinical terms right, not just a low word error rate. Universal-3.5 Pro with Medical Mode reaches a 3.2% Missed Entity Rate and supports speaker separation for clinician/client turns.

Can it generate therapy progress notes automatically?

AssemblyAI provides the accurate transcript; pair it with the LLM Gateway to summarize into SOAP or progress-note formats. The accuracy of the source transcript is what keeps the generated note trustworthy.

Does it separate the therapist from the client with speaker diarization?

Yes, with speaker diarization — on pre-recorded and realtime — so notes attribute statements correctly.

Is it appropriate for protected mental-health data (PHI)?

AssemblyAI is a business associate under HIPAA and signs a standard BAA, with PHI redaction across audio and transcripts and SOC 2 Type 2.

How accurate is Medical Mode on psychiatric medication names?

Medical Mode cuts entity errors by about 87% versus the base model, which disproportionately helps drug names and dosages.

How do I enable Medical Mode for behavioral-health transcription?

Add domain: "medical-v1" to your request — $0.15/hr, no model change.

Behavioral health

Therapy notes that start from an accurate transcript

Behavioral-health sessions are long, emotional, and full of medication names and clinical terms that general speech-to-text fumbles. Universal-3.5 Pro with Medical Mode captures the vocabulary — SSRIs, diagnoses, therapeutic modalities — so your progress-note layer works from a clean source. One parameter, async or realtime.

Try Medical Mode free on your audio Read the docs

The same medical accuracy clinicians trust, tuned for the language of behavioral health

Missed Entity Rate 3.2%

Missed Entity Rate on clinical conversations with Medical Mode.

vs. Universal-3 Pro ~20%

Fewer missed medical entities than Universal-3 Pro alone.

vs. base model 87%

Fewer medical entity errors than the base model.

Benchmarked providers #1

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

See the medical benchmarks

Activation

Add it to your therapy-scribe pipeline in one line

If you already transcribe with AssemblyAI, adding Medical Mode is a one-line change — async or realtime, no model swap and no re-integration.

import assemblyai as aai

aai.settings.api_key = "YOUR_API_KEY"

# Behavioral health: long sessions, psych meds, modalities
config = aai.TranscriptionConfig(
    speech_models=["universal-3-5-pro", "universal-2"],
    domain="medical-v1",
    keyterms_prompt=["sertraline", "CBT", "generalized anxiety disorder", "SSRI"],
)

transcript = aai.Transcriber().transcribe("./therapy-session.wav", config)
print(transcript.text)

Gets psych vocabulary right

Medication names, diagnoses, and modality terms (CBT, DBT, EMDR) land accurately so progress notes don't inherit transcription errors.

Universal-3.5 Pro

Handles long, conversational sessions

50-minute sessions, overlapping speech, and emotional cadence transcribed reliably, with optional diarization to separate clinician and client.

Speaker Diarization

Private by design

PHI redaction across audio and transcripts, BAA available, SOC 2 Type 2.

Medical Mode

Teams building behavioral-health scribes on AssemblyAI

"In the medical context, accuracy is highly important. The biggest impact has been enabling our technical team to focus on workflow-specific features rather than a general speech-to-text pipeline."

Jackson Bierfeldt · Cofounder & CTO, JotPsych

Common questions

: One that gets psychiatric medication names and clinical terms right, not just a low word error rate. Universal-3.5 Pro with Medical Mode reaches a 3.2% Missed Entity Rate and supports speaker separation for clinician/client turns.
: AssemblyAI provides the accurate transcript; pair it with the LLM Gateway to summarize into SOAP or progress-note formats. The accuracy of the source transcript is what keeps the generated note trustworthy.
: Yes, with speaker diarization — on pre-recorded and realtime — so notes attribute statements correctly.
: AssemblyAI is a business associate under HIPAA and signs a standard BAA, with PHI redaction across audio and transcripts and SOC 2 Type 2.
: Medical Mode cuts entity errors by about 87% versus the base model, which disproportionately helps drug names and dosages.
: Add domain: "medical-v1" to your request — $0.15/hr, no model change.

Try Medical Mode free on your audio

Run your own session audio through Medical Mode in the playground — one parameter, no model swap. Free to start, no credit card required.

Try Medical Mode free