Deepgram

Build, deploy, and operate voice applications with integrated speech models, agent orchestration, developer tools, and enterprise deployment options.

Find reviewed voice AI partners and simplify discovery, purchasing, and billing for participating partner products.

Create accurate, readable transcripts and captions, then make media searchable and useful for analysis across languages.

Automate routine calls, guide agents in real time, and turn conversations into actionable contact center data.

Help healthcare teams capture clinical conversations, support multilingual care, and turn speech into documentation and workflow insights.

Automate restaurant conversations and routine tasks to save labor, increase ticket value, and speed service while helping staff focus on guests.

Turn conversations into structured insights to support call review, coaching, routing, and analytics at scale.

Helps early-stage teams build voice AI products with Deepgram credits, engineering resources, and startup community access.

Deepgram Voice AI Platform

A production-ready voice AI platform combining speech-to-text, text-to-speech, voice agent orchestration, audio intelligence, and developer tooling. Build real-time or batch voice applications with APIs, SDKs, and conversational speech models. Capabilities include transcription, speech generation, conversational agents, and task-specific audio insights such as summaries, sentiment, intent, topics, entities, and redaction. Adapt models to domain terminology or customer audio. Deploy in Deepgram’s cloud, a private VPC, on-premises, or air-gapped environments, with enterprise runtime and support options.

Category: Conversational AI Platforms

• Unify Speech And Voice Agent Workflows
  Use speech recognition, speech generation, audio intelligence, and voice agent orchestration as part of an integrated voice AI platform.
• Coordinate Natural Voice Conversations
  Build streaming voice agents with turn detection, interruption handling, function calling, and conversation state managed through the voice agent API.
• Manage Natural Conversation Turns
  Handle pauses, turn endings, and interruptions within the voice workflow, reducing the need to build separate timing logic.
• Maintain Natural Speech Across Turns
  Generate expressive, context-aware speech that carries tone across a conversation and can begin streaming quickly.
• Recognize Speech Across Real-World Conditions
  Produce transcripts designed for varied voices, accents, noise, overlapping speech, and specialized terminology.
• Process Live And Recorded Audio
  Use streaming for live audio and batch processing for recordings across voice applications and transcription workflows.
• Support Responsive Voice Interactions
  Process speech and begin generating audio with low latency to support responsive voice agents and other real-time applications.
• Build Voice Experiences Across Languages
  Support multilingual voice applications with speech recognition and generation, including language detection or code-switching on supported models.
• Keep Conversations Flowing Across Turns
  Use speech models designed to follow dialogue, recognize completed thoughts, and carry context across turns.
• Choose Your Models And Voices
  Configure voice agents with Deepgram models or preferred language model and text-to-speech providers.
• Analyze Conversations Efficiently
  Apply task-specific conversational models to produce structured insights alongside transcripts.
• Deploy Where Your Data Lives
  Choose cloud, private, regional, on-premises, Kubernetes, or air-gapped deployment options to fit infrastructure and data residency requirements.
• Meet Enterprise Deployment Requirements
  Choose deployment environments and runtime controls intended to address enterprise security, compliance, data residency, and service-level requirements.
• Match Voice Delivery To The Moment
  Choose available voices and deliver speech that adapts its tone and emotional register to the conversation.
• Deliver Critical Details Accurately
  Speak names, numbers, and other specific information that voice agents need to communicate reliably.
• Detect Languages And Handle Switching
  Recognize language changes and support conversations that switch between supported languages where the selected model allows it.
• Adapt Models To Your Audio
  Tune recognition to domain terminology and customer audio through runtime term boosting, managed adaptation, industry-specific models, or custom model training.
• Extract Insights From Conversational Audio
  Generate structured insights such as summaries, sentiment, intent, topics, entities, and redaction results from audio and transcripts.
• Build And Test With Familiar Tools
  Use APIs, first-party SDKs, documentation, code examples, and a browser playground to develop and test voice applications.
• Update Agents During Calls
  Adjust context, models, voices, or exact wording during an ongoing conversation.
• Keep Calls Running Through Failures
  Configure fallback for language model or text-to-speech failures so a provider issue does not have to end the conversation.
• Connect Conversations To Tools
  Enable agents to invoke functions during a call and interact with application tools and logic.
• Support Reliable Production Workloads
  Deepgram Dedicated includes a stated 99.99% uptime SLA for production deployments.
• Adjust Speech While It Plays
  Adapt speed, pronunciation, and pacing during live speech generation.
• Create Readable, Attributed Transcripts
  Format transcripts for readability, identify speakers, and provide word-level timing information.
• Redact Sensitive Transcript Data
  Remove sensitive details such as credit card numbers, phone numbers, and personal information from transcripts.
• Choose A Model For Your Workload
  Select Flux TTS for live conversational speech or Aura-2 for efficient high-volume speech generation.
• Integrate Batch Or Streaming Speech
  Use batch requests to generate speech or WebSocket streaming to deliver audio as it is produced.
• Stream Audio At 24kHz Fidelity
  Deliver streamed speech at full 24 kHz fidelity using Flux TTS’s high-fidelity neural codec.
• Inspect Turn Timing And Events
  Review latency across stages of each turn and inspect conversation text, agent events, and function calls.
• Find Words In Long Recordings
  Search for words or phrases in long recordings without running speech recognition on the audio again.
• Get Engineering Help From Benchmark To Production
  Work with Applied Engineers on architecture review, integration, and tuning from initial evaluation through production.

Deepgram Marketplace

A catalog of partner products and services for building voice agents, covering testing and evaluation, audio enhancement, telephony, security, avatars and video, health signals, voice models, and agent platforms. Deepgram reviews listings before publication; partners set their own prices and support their own products. Qualified customers can purchase participating partner products through Deepgram and, in a limited preview, apply eligible Deepgram commitments to those purchases.

• Find Partners Across The Voice AI Stack
  Explore partner offerings for different layers of voice agents in one catalog.
• Choose Partners For Your Voice Stack
  Select products that fit your needs without being limited to partners that integrate with Deepgram. Listings indicate when a documented integration is available.
• Browse Listings Reviewed By Deepgram
  Explore partner options that have been reviewed by Deepgram before publication.
• Apply Eligible Deepgram Credits To Partners
  Qualified customers may use their Deepgram commitment for participating partner products, subject to eligibility and partner-specific terms.
• Simplify Partner Purchasing And Billing
  Purchase participating partner products through Deepgram and receive billing from one vendor, while partners retain responsibility for product pricing and support.
• Connect With A Partner Specialist
  Share what you are building and identify partners of interest to receive follow-up from a Deepgram partner specialist.
• Submit Voice AI Offerings For Review
  Voice AI companies can apply to join the marketplace by submitting a product or service for review.

Media Transcription Solution

Converts podcasts, video, and streaming content into transcripts for captions, subtitles, and analysis. Speech-to-text is designed for multimedia audio, including background noise, overlapping speakers, cross-talk, accents, and industry-specific terms. Transcripts can include speaker labels, word-level timestamps, punctuation, paragraphs, and formatted entities and numbers. The solution supports streaming and batch processing, transcription across more than 60 languages, and Audio Intelligence for summaries, sentiment, intent, and topic detection to support chapter markers, highlight reels, search, and analytics.

• Improve Transcription Accuracy
  Capture words more reliably across challenging recordings, including noisy audio, overlapping speech, accents, and specialized terminology.
• Create Readable, Structured Transcripts
  Receive speaker labels, word timings, punctuation, paragraphs, and formatted details for captions and downstream workflows.
• Transcribe Live And Recorded Content
  Generate transcripts for streaming media or process longer recordings in batch.
• Reach Audiences Across Languages
  Transcribe and analyze content across more than 60 languages, with automatic language detection.
• Generate Insights From Media Content
  Use transcript insights to summarize media, analyze sentiment and intent, detect topics, and support search and content organization.
• Build Accurate Captions And Subtitles
  Use speaker labels, word-level timestamps, and formatted transcripts as the foundation for captions and subtitles.
• Find And Surface Relevant Content
  Search media, create chapter markers, and identify highlight reels using transcript-based insights.

Contact Center Voice AI

Provides a voice AI layer for contact centers and CCaaS platforms, combining speech-to-text, text-to-speech, voice agents, and audio intelligence. Teams can automate routine calls, route callers by intent, support live agents with streaming transcripts, and analyze calls for quality and reporting. The solution supports real-time and post-call workflows, telephony audio, and deployment in Deepgram’s cloud, a private cloud or VPC, on-premises, or inside a carrier network.

• Automate Routine Calls
  Build voice agents that handle routine calls and support natural exchanges with built-in barge-in, turn-taking, and function calling.
• Route Calls By Intent
  Use callers’ spoken intent to guide calls to the appropriate destination.
• Guide Agents During Calls
  Use fast streaming transcripts to provide relevant guidance while the customer is still speaking.
• Analyze Every Call
  Turn calls into structured information for quality assurance, coaching, and reporting across live and post-call workflows.
• Deploy Within Your Environment
  Choose cloud, VPC, on-premises, self-hosted, or carrier-network deployment to fit infrastructure and data residency needs.
• Transcribe Real-World Calls
  Process contact center audio with telephony codecs, overlapping speech, accents, alphanumerics, and background noise in mind.
• Attribute Speech To Speakers
  Separate who said what in a call to support clearer transcripts and speaker-specific analysis.
• Give Voice Agents Natural Speech
  Generate fast, conversational speech for voice agents, with tone that carries across turns.
• Redact Sensitive Information
  Remove PII from transcripts before sharing them with reviewers or sending them to storage and downstream systems.
• Recognize Important Terms
  Add account terms, product names, and acronyms to a request to support recognition without retraining a model.
• Transcribe Calls In More Languages
  Transcribe supported languages in recorded or live calls and provide transcripts that can help agents continue conversations after a handoff.
• Connect Contact Center Platforms
  Integrate Deepgram’s contact center speech capabilities with supported platforms and providers.

Medical and Healthcare Solution

A medical speech-to-text solution using Nova-3 Medical, trained on clinical audio and optimized for medical terminology. It supports multilingual transcription in batch and real time for ambient scribes, EHR platforms, healthcare applications, dictation, and voice agents. Teams can adapt recognition with Keyterm Prompting and choose managed cloud, on-premises, or VPC deployment. The solution also provides speaker labels, word-level timestamps, formatting, summaries, topic detection, and sentiment analysis. Healthcare deployments can use HIPAA-eligible infrastructure with a BAA available, encryption for patient data, and enterprise Zero Data Retention options.

• Recognize Clinical Terms More Accurately
  Capture medical terms that clinical applications depend on for documentation, intake, and other healthcare workflows.
• Transcribe Medical Speech Across Languages
  Support conversations in 10 languages with a medical-specialized model, including conversations that switch between supported languages.
• Process Live And Recorded Audio
  Choose real-time transcription for live clinical conversations or batch processing for recorded audio, archives, and review workflows.
• Capture Clinical Conversations As Visits Happen
  Provide live transcription for ambient documentation workflows, helping note drafts reflect the clinical conversation for review.
• Deploy Within Your Approved Environment
  Choose a managed or private deployment path to meet healthcare security and regulatory requirements, with available encryption, HIPAA-eligible infrastructure, BAA, and Zero Data Retention options.
• Adapt Recognition To Clinical Vocabulary
  Provide relevant clinical terms for a request to support specialty-specific recognition without retraining.
• Identify Speakers And Locate Spoken Words
  Identify who said what and connect transcript content with its position in the original recording.
• Transcribe Speech Across Clinical Conditions
  Support transcription across recording devices and everyday clinical speech patterns, including accents, disfluencies, and overlapping speakers.
• Generate Insights From Transcribed Speech
  Use summaries, topic detection, and sentiment analysis with medical transcription to support healthcare analytics and workflow automation.
• Integrate Transcription Into Existing Pipelines
  Use documented REST and WebSocket APIs and SDKs to connect medical transcription with an existing healthcare application or pipeline.
• Format Clinical Transcripts For Review
  Receive formatted transcripts with punctuation and paragraphs to support review and clinical documentation workflows.

Restaurant Voice AI Solution

Provides voice models, tooling, and orchestration for restaurant operations. It supports order-taking across drive-thru, phone, and kiosk; reservations; employee assistance; call center automation; scheduling; catering and special requests; and interaction analytics. The system is designed for restaurant conditions such as background noise, cross-talk, accents, and changing menu data. It can connect with POS and other restaurant systems, convert spoken orders into structured carts, and escalate conversations to staff when needed.

• Automate Orders Across Channels
  Take voice orders across drive-thru, phone, and kiosk while handling restaurant noise, menu details, modifiers, and upsells.
• Automate And Route Support Calls
  Handle common support requests for guests, employees, and franchisees while directing calls that need human attention to staff.
• Recognize Speech In Noisy Restaurants
  Use speech recognition designed for restaurant audio conditions, including background noise, overlapping speech, accents, and speaker distortion.
• Keep Voice Agents Current On Menus
  Normalize live menu data for voice agents and keep item availability, limited-time offers, and valid modifiers up to date.
• Convert Orders Into POS-Ready Carts
  Translate spoken orders into structured carts, validate items against the menu, and pass them to the POS ready for payment.
• Manage Reservations By Voice
  Let guests book, change, or cancel a table through supported channels using live availability.
• Give Staff On-Demand Answers
  Provide kitchen and floor staff with voice-accessible answers to operational questions, such as recipes and inventory counts.
• Analyze Every Interaction
  Review interactions for sentiment, script adherence, upsell conversion, hold times, and out-of-stock moments.
• Respond Quickly With Natural Speech
  Generate expressive voice responses with time to first audio as low as 80 milliseconds and maintain tone across order turns.
• Adapt Agents To Restaurant Data
  Fine-tune agent understanding using a restaurant’s menus, scripts, regional accents, and order history.
• Connect Voice To Existing Systems
  Integrate voice workflows with restaurant systems such as POS, menu management, online ordering, scheduling, inventory, kitchen displays, CRM, and loyalty.
• Escalate Conversations To Staff
  Pass conversations to an employee when requested by a guest or when the interaction is unusual or outside the agent’s scope.
• Manage Shift Requests By Voice
  Let employees make shift swap and scheduling requests through a voice agent that works with existing scheduling tools.
• Capture And Automate Special Orders
  Capture details for special inquiries, catering, and special orders so staff do not have to relay them manually.
• Review Sessions Across Locations
  Use session logs, latency traces, and transcript review to examine activity across locations, lanes, and shifts.

Speech Analytics Solution

Provides speech-to-text, speaker diarization, and audio intelligence APIs for analytics platforms. The APIs turn calls into transcripts and structured insights such as sentiment, intent, topics, and entities, for real-time or post-call use. Teams can build quality assurance, coaching, routing, reporting, and other workflows, with deployment options in the cloud, a VPC, or on-premises.

• Transcribe Calls For Analysis
  Create the transcript foundation for call analytics, quality assurance, and reporting using streaming or batch processing.
• Extract Structured Call Insights
  Use structured conversation insights to categorize calls, identify customer intent and topics, and power analytics workflows.
• Attribute Speech To Speakers
  Identify which participant said what to support speaker-level metrics such as talk-time ratios and per-speaker sentiment.
• Review Every Call
  Build QA workflows that cover the full call volume and provide a basis for reviewing and scoring more conversations.
• Analyze Calls Live Or Later
  Support in-call use cases such as coaching, as well as post-call analysis and batch processing of stored recordings.
• Summarize Calls For Review
  Give supervisors a concise account of a call so they can review conversations without replaying audio or reading the entire transcript.
• Redact PII Before Review
  Remove sensitive information from transcripts before sharing them with reviewers, customers, or downstream systems.
• Deploy Where Recordings Live
  Choose cloud, VPC, or on-premises deployment to fit requirements for data location and control.
• Boost Important Call Terms
  Provide relevant product names, competitor names, and acronyms at request time to support terminology recognition without retraining.

Deepgram for Startups

An application-based program for AI builders and early-stage startups creating their own voice AI products. Selected companies can receive Deepgram API credits for speech-to-text, text-to-speech, and voice agent capabilities, along with access to invite-only events, engineering support groups, and the Deepgram startup community. Credit awards depend on company stage, intended usage, and partner referrals, and may reach $100,000 for 12 months.

• Access Up To $100,000 In API Credits
  Use program credits to build with Deepgram speech-to-text, text-to-speech, and voice agent capabilities. Amounts depend on company stage, intended usage, and partner referral; grants expire after 12 months.
• Connect With Startup Builders And Engineers
  Participate in invite-only events and engineering support groups, and connect with the Deepgram startup community.
• Apply As An Early-Stage Voice AI Builder
  Startups can apply if they are building their own voice AI product and are in production or expect to launch within six months. Some applicants receive case-by-case review.
• Receive Decisions And Credits On A Clear Timeline
  Applicants typically receive a decision within one week. Approved credits are automatically added to the Deepgram console within 24–48 hours.
• Explore Enhanced Offers Through Program Partners
  Startups applying through participating venture capital firms, accelerators, or startup networks may be eligible for partner-specific enhanced offers.
• Add Team Members Without Per-User Billing
  Invite the team members needed to build with Deepgram; billing is based on usage rather than user count.