Vendor profile · Developer speech-to-text API
Deepgram review
Fast, developer-friendly speech-to-text API with medical model options — transcription, not documentation.
Short answer
What is Deepgram and who should use it?
Deepgram is the strongest pure developer experience in speech-to-text: excellent docs, self-serve keys, low latency, transparent per-minute pricing. It transcribes audio; it does not produce a clinical note, so you are still building the documentation layer yourself.
Cite as: Deepgram vendor assessment, Compare Healthcare API, last reviewed 2026-09-01.
Overview
If the requirement is genuinely speech-to-text, Deepgram is an easy recommendation. Documentation is public and complete, keys are self-serve, streaming latency is competitive, and pricing is published per minute so you can model costs before writing code.
The gap is scope. A clinical note is not a transcript: it requires structuring, summarisation, template adherence, clinical safety review and chart write-back. Choosing an ASR API means owning all of that, including the prompt engineering, evaluation and clinical safety work that a scribe API absorbs.
Integration surface
What an engineering team is actually buying: the API posture, how you authenticate, what you send and what comes back.
| API posture | Public developer API |
|---|---|
| What the API does | Speech-to-text API with a medical-tuned model line for clinical vocabulary, offered as batch and real-time streaming transcription. |
| Authentication | API key |
| Transport | REST for pre-recorded audio and WebSocket streaming for live audio |
| Response payload | Transcripts with word timings, confidence and diarisation. No clinical note structure — the documentation layer is yours to build. |
| SDKs / clients | Official client libraries across common server languages |
| Try without a contract | Yes — self-serve key and published pricing |
Official developer resources
- Deepgram developer docsdevelopers.deepgram.com
- Medical transcription solutiondeepgram.com
Scores by criterion
Rubric v1.0. Each score is out of 10; the weight is the share of the total.
| Criterion | Weight | Score |
|---|---|---|
| API & SDK maturityCan an engineering team ship against this without a services contract? | 22% | 9.69.6 |
| Clinical accuracy & output qualityDoes the transcript and the resulting note hold up on real clinical audio? | 20% | 7.87.8 |
| EHR & FHIR interoperabilityHow much integration work stands between the output and a chart? | 16% | 3.23.2 |
| Compliance & security postureCan this survive your customer's security review? | 16% | 8.38.3 |
| Latency & streaming behaviourIs it fast enough for the interaction you are building? | 10% | 9.49.4 |
| Specialty & template coverageWill it produce the note your users actually write? | 8% | 5.45.4 |
| Commercial terms & transparencyCan you price your own product on top of it? | 8% | 9.29.2 |
Disclosed facts
Recorded from public documentation. Anything a vendor does not publish is recorded as not disclosed, never estimated.
| Deployment model | Cloud API, streaming and batch; self-hosted options for enterprise |
|---|---|
| Pricing model | Published per-minute pricing |
| BAA availability | Available on qualifying plans |
| Certifications | SOC 2; HIPAA support — confirm current scope |
| Training on PHI | Documented in the data processing terms |
| Streaming support | Yes — low latency |
| FHIR / EHR write-back | No |
| White-label rights | Yes — it is infrastructure |
| Self-serve access | Yes |
Strengths
Where it scores well
- Best-in-class developer documentation and self-serve onboarding.
- Published per-minute pricing, so unit economics are modellable immediately.
- Low-latency streaming with strong concurrency characteristics.
- HIPAA BAA available on appropriate plans.
Limitations
Where it loses points
- Transcription only — no clinical note generation, templates or structured clinical output.
- You own the entire documentation layer above the transcript, including clinical evaluation.
- No EHR or FHIR integration surface.
Best fit
Teams that want raw transcription and have the capacity to build and validate the clinical documentation layer themselves.
Evidence & sources
Every factual claim on this page traces to one of the primary references below. Each entry records what it supports and its evidence tier, so documentation can be told apart from judgement.
U.S. Department of Health & Human Services · Regulation · Tier A — primary documentation
Supports: The contractual clauses a documentation vendor's BAA must contain.
U.S. Department of Health & Human Services · Regulation · Tier A — primary documentation
Supports: Administrative, physical and technical safeguards a vendor handling recorded encounter audio must implement.
HL7 International · Standard · Tier A — primary documentation
Supports: Resource definitions (DocumentReference, Composition, Encounter, Condition, MedicationRequest) that clinical documentation output must map onto.
HL7 International · Standard · Tier A — primary documentation
Supports: The canonical target resource for writing a generated clinical note back to a chart.
AICPA · Standard · Tier A — primary documentation
Supports: What a SOC 2 Type II report does and does not attest to during a vendor security review.
NIST speech recognition evaluation literature · Methodology · Tier B — published methodology
Supports: Why a headline WER figure without a stated corpus, audio condition and reference-transcript protocol is not comparable across vendors.
Source tiers are defined on the methodology page. Outbound links are unaffiliated and carry no commercial relationship.
Frequently asked questions
- Does Deepgram have an API?
- Public developer API. Speech-to-text API with a medical-tuned model line for clinical vocabulary, offered as batch and real-time streaming transcription. An engineer can obtain access and make a call without a contract.
- How do you authenticate with the Deepgram API?
- API key. Transport: REST for pre-recorded audio and WebSocket streaming for live audio.
- What does the Deepgram API return?
- Transcripts with word timings, confidence and diarisation. No clinical note structure — the documentation layer is yours to build.
- Is Deepgram a good fit for embedding in my software?
- Teams that want raw transcription and have the capacity to build and validate the clinical documentation layer themselves. Teams that need a finished clinical note without building an LLM documentation pipeline.
- What are Deepgram's main limitations?
- Transcription only — no clinical note generation, templates or structured clinical output. You own the entire documentation layer above the transcript, including clinical evaluation. No EHR or FHIR integration surface.
- How was Deepgram scored?
- Against seven weighted criteria under rubric v1.0. Deepgram scores 7.6 of 10 overall. Methodology.
- How much does Deepgram disclose publicly?
- 94 of 8 disclosure checks in our Vendor Transparency Index are met by published documentation. Transparency index.