← Back to All Tools
Video & Audio Free Tier Available

Deepgram

Deepgram is a speech AI infrastructure platform for developers building applications that listen, transcribe, speak or hold automated voice conversations. It provides speech-to-text models, text-to-speech services and a Voice Agent API that combines speech processing with conversational logic.

Teams use Deepgram for use cases such as live transcription, recorded audio analysis, telephone automation and voice-enabled software. Its API-focused approach gives developers control over how speech capabilities are integrated into their own applications.

Deepgram offers a pay-as-you-go entry model with a one-time introductory credit rather than a recurring free usage allowance. Costs depend on which services, models and optional features are used, and Enterprise customers can discuss dedicated deployment requirements.

Visit Deepgram Compare Deepgram

Our tools are evaluated against our Scoring Methodology.
Spot an error? Request a Correction.

Research-based profile using Deepgram's official pricing, product and self-hosting documentation reviewed on 5–6 November 2026. Published technical specifications are vendor-reported; no independent accuracy, latency or reliability benchmarks are claimed.

Best For

  • Software developers adding speech recognition or generated voice to applications.
  • Contact-centre platforms analysing recorded or live customer conversations.
  • Enterprises needing configurable speech APIs and potential private infrastructure deployment.

Not Ideal For

  • Consumers seeking a finished meeting-notes or podcast-editing application.
  • Users expecting a fresh allocation of free API credits every month.
  • Teams unable to estimate audio-channel, add-on and connection-duration costs.

How It Works

An application sends recorded or streaming audio to Deepgram's speech recognition API. Deepgram returns recognised text using the selected model and options. Developers can add capabilities such as speaker diarization, formatting or supported redaction where appropriate.

For speech output, an application sends text to a Deepgram text-to-speech service and receives generated audio. The Voice Agent API supports an integrated conversational process involving speech recognition, language-model reasoning, tool use and speech generation.

Deepgram charges according to service-specific usage units, including processed audio duration, text characters or voice-agent connection time. Additional options and multiple audio channels can increase charges, so developers should review billing calculations and processing settings before scaling production workloads.

Key Features

Speech Recognition

  • Streaming and pre-recorded speech-to-text APIs.
  • Flux and Nova model options for supported use cases.
  • Optional speaker diarization, formatting, redaction and related features.

Speech Generation and Voice Agents

  • Text-to-speech through supported Aura and Flux models.
  • Voice Agent API for conversational application development.
  • Integration with external application logic and supported model configurations.

Developer and Enterprise Infrastructure

  • APIs and documentation for custom software integration.
  • Audio analysis features and usage-based billing.
  • Enterprise self-hosted deployment options using supported GPU infrastructure.

Practical Workflows

  1. Real-time voice application: Stream incoming audio to Deepgram, process the recognised speech within an application's conversation logic, and return a generated spoken response through the selected voice service.
  2. Recorded call analysis: Transcribe customer calls, use supported options to distinguish speakers or redact selected information, and generate structured summaries for an authorised support or quality-assurance workflow.
  3. Enterprise speech deployment: Evaluate Deepgram's self-hosted container requirements, provision compatible GPU infrastructure and integrate speech services into an approved private environment.

Strengths & Limitations

Strengths

  • Covers transcription, generated speech and voice-agent development.
  • Supports streaming and recorded audio workflows.
  • Offers a one-time introductory credit with no monthly platform minimum.
  • Provides enterprise deployment and data-handling options.

Limitations

  • Most use cases require software development and API integration.
  • The introductory $200 credit does not renew monthly.
  • Multichannel audio and optional recognition features can increase usage bills.
  • Enterprise self-hosted deployment requires commercial arrangements and suitable GPU infrastructure.

Pricing

Deepgram's Pay As You Go plan has no monthly platform minimum and includes a one-time $200 introductory credit that the vendor states does not expire. Usage is billed after credits are consumed. Growth starts from $4,000 per year in prepaid credits, with eligible discounted rates, while Enterprise offers negotiated commercial and deployment terms.

For speech recognition, published Pay As You Go examples include $0.0077 per minute for Flux STT English and $0.0077 per minute for Nova-3 monolingual streaming, with a lower published rate for selected prerecorded transcription. Audio is generally billed by processed duration, with multichannel recordings charged according to the combined duration across channels. Optional diarization, redaction and similar features may add separate charges.

Text-to-speech uses character-based billing, and the Voice Agent API uses connection-time billing. Deepgram's prices vary by model and configuration, so comparisons should use a complete estimated workload rather than treating one speech-to-text rate as the price of an entire voice-agent application.

How It Compares

Choose Deepgram when the priority is integrating speech recognition, speech generation or voice-agent services into existing applications. It is particularly relevant when a team wants to compare speech-model rates and retain flexibility over application architecture.

Consider Cartesia for another developer-oriented speech and managed-agent stack, AssemblyAI for transcription and speech analysis APIs, or Bland AI for a more integrated business telephony automation platform. The relevant decision depends on speech features, integration effort, data-handling requirements and total usage costs.

Visit Deepgram

Browse more Video & Audio tools →