Deepgram
Usage-based with a free tier and annual commitment plans.
Documentation Provided
Overview
Deepgram offers AI-powered voice technology, including Speech-to-Text (STT), Text-to-Speech (TTS), and a Voice Agent API, designed to process and understand spoken language. For healthcare, Deepgram provides its Nova-3 Medical model, an STT solution specifically trained on medical terminology, including symptoms, diagnoses, treatments, medications, and clinical jargon.
- Who it is for: Deepgram’s medical AI tools are intended for healthcare professionals across various specialties and care settings. This includes clinicians, administrative staff, and developers building healthcare applications.
- How it fits a clinical or practice workflow: Deepgram’s technology integrates into clinical workflows to automate and enhance tasks such as clinical documentation, prescription workflows, and patient engagement. It can facilitate real-time transcription of patient encounters, generate structured clinical notes for EHR integration, and power AI voice agents for tasks like appointment scheduling, patient intake, and prescription refill requests. The aim is to reduce manual entry and administrative burdens, allowing providers to focus more on patient care.
- Notable Capabilities:
- High Accuracy for Medical Terminology: The Nova-3 Medical model is trained to accurately recognize specialized medical terms, acronyms, and Latin-derived terminology, even in challenging audio environments.
- Real-time Processing: It offers real-time streaming transcription, which is crucial for live clinical workflows and conversational AI agents.
- Customization: The model can be fine-tuned for specific medical specialties and workflows using Keyterm Prompting to boost accuracy for up to 100 custom terms.
- Deployment Flexibility: Deepgram supports various deployment options, including managed cloud, on-premises, and Virtual Private Cloud (VPC) configurations, to meet security and data residency requirements.
- Voice Agent API: This unifies STT, TTS, and large language model (LLM) orchestration into a single API for building conversational AI agents.
Reviewed by Pouyan Golshani, MD — Interventional Radiologist
Key Features
- Conversation-aware speech platform (Flux)
- Real-time Speech-to-Text (STT)
- Expressive Text-to-Speech (TTS)
- Unified Voice Agent API
- LLM orchestration
- Audio Intelligence
- Cloud or self-hosted deployment
- Custom models
- API-first approach
Use Cases
- Contact centers
- Speech analytics
- Conversational AI
- Medical transcription
- Voice agents
- Podcast transcription
What Physicians Need to Know
Healthcare API Support (FHIR/HL7)
Deepgram's solutions can integrate with EHR systems through FHIR APIs or HL7 via lightweight adapters. They support writing clinical observations back from an app using SMART scopes and Observation.Create operations. Deepgram's Nova-3 Medical model is designed for healthcare voice agent deployments and can be used to build applications that integrate with EHR systems.
HIPAA-Compliant Infrastructure
Deepgram maintains HIPAA compliance and is considered a Business Associate under US HIPAA legislation. They can provide a Business Associate Agreement (BAA) upon request for eligible healthcare customers. Deepgram's infrastructure includes robust security measures such as data encryption at rest (AES-256) and in transit (TLS 1.3), role-based access control, and mandatory multi-factor authentication. They offer flexible deployment options including managed cloud, private cloud (VPC), and on-premises to meet specific security and compliance requirements.
Clinical NLP Capabilities
Deepgram's Nova-3 Medical model is specifically trained on healthcare terminology, including symptoms, diagnoses, treatments, medications, and clinical jargon. It offers high accuracy in transcribing complex medical terminology, even in noisy clinical environments. The platform supports features like speaker diarization (identifying who is speaking) and smart formatting (punctuation, capitalization, paragraphs). Deepgram's Audio Intelligence API can also provide real-time actionable insights, including summarization, sentiment analysis, topic detection, and intent recognition.
De-Identification Tools
Deepgram offers redaction capabilities for Protected Health Information (PHI), Personally Identifiable Information (PII), and Payment Card Industry (PCI) data. You can redact entire groups of entities (e.g., `phi`, `pii`, `pci`) or select specific entity types from over 50 supported options. For streaming connections, Deepgram maintains context across audio chunks to detect entities spanning multiple segments. Redaction is available for English only.
Medical Terminology Support
Deepgram's Nova-3 Medical model is purpose-built for medical vocabulary recognition, including pharmaceutical names, clinical acronyms, and Latin-derived disease terminology. They offer 'Keyterm Prompting,' allowing developers to add up to 100 domain-specific terms per API request at inference time without model retraining, which significantly improves accuracy for low-frequency medication names or specialty jargon.
Sandbox/Testing Environment
Deepgram provides an API Playground for developers to explore and test the API in an isolated environment without writing code. New accounts receive $200 in free credit, equivalent to approximately 43,000 minutes (over 700 hours) of transcription, which does not expire, allowing for extensive prototyping.
SDK Languages
Deepgram provides official SDKs for JavaScript/TypeScript, Python, .NET, Go, Java, and Rust.
Rate Limits & Pricing
Deepgram offers a 'Pay-As-You-Go' plan with no minimums or expiration, starting with a $200 free credit. The Nova-3 model for monolingual streaming costs $0.29/hour, and multilingual is $0.35/hour. Deepgram uses true per-second billing, avoiding rounding up audio time. They also have a 'Growth' plan for higher usage (starting at $4,000/year commitment) with discounted rates and higher concurrency limits, and custom 'Enterprise' pricing for large-scale deployments with negotiated rates and dedicated support. Add-on features like summarization, topic detection, and sentiment analysis are billed separately, often on a per-token basis. The Voice Agent API bundles STT, TTS, and orchestration at $4.50/hour.
Certification Program
Deepgram is SOC 2 Type 2 certified. They also adhere to PCI compliance.
Physician Tip
Deepgram's Nova-3 Medical model is highly accurate for transcribing clinical conversations, even in noisy environments, and supports specialized medical terminology. Physicians can leverage 'Keyterm Prompting' to ensure accurate recognition of specific drug names, jargon, or institution-specific terms without needing model retraining. The HIPAA-compliant infrastructure and de-identification tools for PHI can help maintain patient privacy. Consider using Deepgram to automate documentation, create AI voice agents for patient interactions (scheduling, refills), and gain insights from patient calls to reduce administrative burden and improve patient care.
Deepgram offers flexible deployment options including managed cloud, private cloud (VPC), and on-premises, compatible with major cloud providers like AWS, GCP, Oracle, and Azure. Integration with Electronic Health Record (EHR) systems is possible via FHIR APIs or HL7 adapters. For generative AI orchestration, Deepgram can be integrated with services like Amazon Bedrock. They also have partnerships with companies like Five9 for contact center AI solutions. Deepgram's SDKs support various languages, facilitating integration into diverse development stacks.
Details
| Category | Developer Tools & APIs, Documentation & Scribing, General Productivity (AI Assistants) |
| Pricing |
Usage-based with a free tier and annual commitment plans.
|
| Free Trial | Unknown |
| Deployment | Deepgram offers flexible deployment options including multi-tenant cloud, single-tenant isolation, customer-managed environments (VPC, private cloud), and self-hosted (on-premise or in your own cloud instance like AWS, GCP, Oracle, Azure, or bare-metal servers, requiring NVIDIA GPUs). [1, 2, 5, 9, 14, 27, 36, 39] |
| API Available | Unknown |
| Languages | Deepgram supports numerous languages. For Speech-to-Text (Nova-2 and Nova-3 Multilingual), this includes English, Spanish, French, German, Hindi, Italian, Japanese, Dutch, Russian, Portuguese, Bulgarian, Catalan, Chinese (Mandarin, Simplified & Traditional), Czech, Danish, Estonian, Finnish, Flemish, German (Switzerland), Greek, Hungarian, Indonesian, Korean, Latvian, Lithuanian, Malay, Norwegian, Polish, Romanian, Slovak, Swedish, Thai, Turkish, Ukrainian, and Vietnamese. [7, 16, 32, 34] Deepgram's Aura Text-to-Speech supports English (American, British, Australian, Irish, Filipino accents), Spanish (Mexican, Peninsular, Colombian, Latin American accents), German, French, Dutch, Italian, and Japanese. [7] |
| Training | Unknown |
| Target Size | Deepgram serves a wide range of organizations, from developers and startups to growing applications and enterprises. [2, 37, 39] |
| Compliance | |
| BAA Available | Yes AI-estimated |
| HIPAA Compliant | Yes AI-estimated |
| FDA Status | Unknown AI-estimated — unknown |
| SOC 2 | Yes AI-estimated |
| GDPR | Yes AI-estimated |
| Integrations | |
| EHR | Not specified |
| Specialties | Family Medicine, Hospital Medicine, Internal Medicine |
Social Proof
| Customers | As of January 2026, more than 1,300 organizations were building on Deepgram. [38] As of October 2025, Deepgram serves over 400 enterprise customers and has over 200,000 developers actively using its APIs worldwide. [37] As of November 2022, Deepgram had 300+ customers and 15,000 users. [28, 33] As of April 2024, Deepgram had over 500 companies in its customer portfolio. [40] |
| Notable | NASATwilioCitiSpotifyTethrMarchexSaleskenConvergeOneDecagonSierraVapiGranolaFive9Sharpen. [82628334041] |
Support & Reliability
| Training Provided | Unknown |
What the Web Says
Deepgram is an AI-powered speech recognition platform offering speech-to-text (STT), text-to-speech (TTS), and voice agent capabilities through APIs. It is praised for its high accuracy, especially in real-time transcription and handling specialized terminology, including medical jargon. The platform is designed for developers and enterprises building voice-enabled applications, with flexible deployment options and strong security features like HIPAA compliance.
Overall: PositiveStrengths
- High accuracy in speech-to-text, even with accents, noisy environments, and specialized terminology (e.g., medical).
- Fast and low-latency real-time transcription, crucial for conversational AI and live applications.
- Easy API integration and well-organized documentation, making it developer-friendly.
- Comprehensive platform offering STT, TTS, and Voice Agents in a unified stack.
- Customizable models and fine-tuning for domain-specific vocabulary.
- Strong enterprise compliance, including SOC 2, HIPAA, GDPR, and PCI, with flexible deployment options (cloud, VPC, on-premise).
Limitations
- Pricing can be a concern as usage scales and for high volumes or add-ons.
- Fewer languages supported compared to some competitors (40+ vs 100+).
- TTS language support is still expanding and has fewer voice options than dedicated TTS platforms.
- May be more complex than needed for simple batch transcription tasks.
- Speaker diarization can struggle with multiple speakers or similar-sounding voices, especially in non-English long-form audio.
- Opting out of data training may forfeit a 50% discount.
Based on reviews from: G2, Reddit, PeerSpot, Gladia, Capterra, DIY AI, Agent Finder, Deepgram Blog, Deepgram Website, FeaturedCustomers
Last updated: 2026-08-30
Ratings & Reviews
No reviews yet. Be the first to review this tool!
Rate Deepgram
Press & Coverage
Business Wire
Introducing Nova-3 Medical: The Most Accurate Medical Transcription Model in the World
Deepgram launched Nova-3 Medical, a next-generation AI-powered speech-to-text model specifically engineered for the healthcare industry, claiming industry-leading transcription accuracy with a 63.6% reduction in errors compared to competitors. The model was announced at HIMSS25 and is designed for highly accurate, customizable, and secure voice AI products in clinical environments.
2025-03
MedCloudInsider
Deepgram Unveils Nova-3 Medical, Advanced Speech AI for Healthcare - MedCloudInsider
Deepgram announced Nova-3 Medical, a speech-to-text model for healthcare, at HIMSS25, touting it as the most accurate real-time medical transcription on the market with a 3.45% word error rate. The model is HIPAA compliant, offers on-premises and VPC deployment options, and supports integration with EHR systems.
2025-04
AI News
Deepgram Nova-3 Medical: AI speech model cuts healthcare transcription errors - AI News
Deepgram's Nova-3 Medical AI speech-to-text model, designed for healthcare, significantly reduces transcription errors with a 3.45% Word Error Rate and a 6.79% Keyword Error Rate. It aims to improve efficiency and accuracy in clinical documentation and is presented as a cost-effective solution for healthcare tech companies.
2025-03
Amazon Web Services (AWS) Blog
Modernizing Healthcare with Generative Voice AI Using Deepgram on AWS
This article details how Deepgram Nova-3 Medical, an advanced speech-to-text AI model trained on healthcare terminology, integrates with AWS to provide highly accurate, real-time, and HIPAA-compliant transcription. The solution aims to transform patient and medical professional interactions by building custom applications for clinical documentation, pharmacy refills, and more.
2025-08
Healthcare Markets & Technology
Deepgram's healthcare gambit: when the best voice AI isn't built for healthcare
This essay analyzes Deepgram's competitive positioning in healthcare voice AI following its $143.2 million Series C raise, exploring its horizontal platform approach versus healthcare-specific competitors. It discusses Deepgram's technical architecture, primary healthcare use cases like ambient documentation, and the challenges of a platform-first strategy in a specialized vertical.
2026-02
IBM Newsroom
Deepgram and IBM Introduce Advanced Voice Capabilities for Enterprise AI
IBM and Deepgram announced a collaboration to integrate Deepgram's speech-to-text and text-to-speech capabilities into IBM's watsonx Orchestrate generative AI solution. This partnership aims to provide highly performant, enterprise-grade transcription and real-time captioning for various fields, including healthcare and finance.
2026-02
Deepgram Blog
Deepgram Delivers Private Voice AI to Regulated Industries with On-Premises Deployments Powered by Fortanix Confidential AI and NVIDIA Confidential Computing
Deepgram partnered with Fortanix and NVIDIA to enable enterprises in regulated industries, such as healthcare, to run voice AI in their own environments with enhanced security. This solution provides on-premise deployments with protection for sensitive audio and AI model weights during processing.
2026-06
Deepgram Blog
AI Voice Agents in Healthcare: 7 Production Use Cases - Deepgram
This article discusses the deployment of AI voice agents in healthcare for tasks like scheduling, prescription refills, and triage, emphasizing the importance of domain-specific STT models like Deepgram's Nova-3 Medical for accurate medical vocabulary recognition. It highlights the challenges of general-purpose ASR models with medical terminology.
2026-05
Frequently Asked Questions
Deepgram is HIPAA compliant and can act as a Business Associate, offering a Business Associate Agreement (BAA) to eligible healthcare customers. They encrypt sensitive information, including PHI, both in transit and at rest, using industry-standard algorithms like AES-256 and TLS. Deepgram also implements robust security measures such as hardened systems, secured environments, role-based access control, and two-factor authentication to protect customer data.
Deepgram provides flexible deployment options including managed cloud, private cloud, and self-hosted (on-premise or within your own VPC) solutions. While cloud-hosted processing requires a BAA, self-hosted deployments keep audio within your infrastructure, shifting the full security burden to your organization but potentially reducing the scope of subcontractor BAAs. Deepgram also offers regional endpoints, such as an EU-specific endpoint, to help with data residency requirements.
While Deepgram's Nova-3 Medical model is specifically trained on healthcare terminology and aims for high accuracy in clinical environments, general ASR models can struggle with specialized medical vocabulary, acronyms, and diverse accents. Deepgram addresses this with 'Keyterm Prompting,' allowing developers to add up to 100 clinical terms at inference time without retraining, which can significantly improve accuracy for specific jargon like drug names. However, even with advanced models, human review is still recommended for critical medical documentation due to the potential for clinically significant errors.
Deepgram offers a 'Pay-As-You-Go' plan with a free credit, and then charges per hour for speech-to-text, with rates varying by model and whether the audio is streaming or pre-recorded. For enterprise customers with high usage or specific requirements like ePHI handling, enterprise sales agreements are available, which may include custom concurrency limits and volume discounts. Audio intelligence features like sentiment analysis and summarization are typically priced per token rather than per minute.
Alternatives to Deepgram include Speechmatics, AssemblyAI, Google Cloud Speech-to-Text, and AWS Transcribe Medical. Speechmatics, for instance, offers a dedicated Medical Model trained on clinical terminology and supports over 55 languages, often with lower word error rates in noisy settings. AWS Transcribe Medical is also a strong option for HIPAA compliance and signed BAAs. Each alternative has varying strengths in areas like multilingual support, on-premise deployment, and integrated audio intelligence features.
Suggest an Edit →
| Last Verified: 2026-08-31 | First Added: 2026-08-30