Deepgram API

Converts live audio streams to text via WebSocket (WSS) protocol with ultra-low latency processing. Deepgram's Flux models process audio chunks incrementally, detecting natural speech boundaries and returning partial transcripts in real-time without waiting for audio completion. Supports 150-225 concurrent WebSocket connections depending on tier, enabling high-throughput voice applications.

Processes pre-recorded audio files via REST API with automatic speaker identification and segmentation. Nova-3 models analyze complete audio files to detect multiple speakers, assign speaker labels, and return structured transcripts with speaker turns and timing information. Handles background noise, crosstalk, and far-field audio through deep learning-based noise robustness.

Deepgram offers custom model training for organizations with proprietary speech patterns, accents, or domain-specific audio characteristics. Custom models are trained on customer-provided datasets and deployed as dedicated endpoints. Enables organizations to achieve higher accuracy on edge-case audio (heavy accents, background noise, specialized vocabulary) that generic models struggle with.

Automatically applies formatting rules to transcripts to improve readability without manual post-processing. Converts numbers to digits, adds punctuation, capitalizes proper nouns, and formats currency/dates according to locale. Smart formatting operates on raw transcription output, transforming 'one thousand two hundred thirty four dollars' to '$1,234' and 'the meeting is on january fifteenth' to 'The meeting is on January 15th'.

Flux Multilingual model supports 10 languages (English, Spanish, German, French, Hindi, Russian, Portuguese, Japanese, Italian, Dutch) within a single WebSocket stream, automatically detecting language switches mid-conversation. Enables applications to handle multilingual users without requiring separate connections or language pre-specification. Language detection happens continuously throughout the stream.

Deepgram enforces concurrent connection limits that vary by API type and subscription tier. WebSocket STT supports 150 (free/pay-as-you-go) or 225 (Growth tier) concurrent connections; REST STT/TTS limited to 50 concurrent; Voice Agent API limited to 45 (free) or 60 (Growth) concurrent; Audio Intelligence limited to 10 concurrent regardless of tier. Developers must manage connection pooling and queuing to respect these limits.

Deepgram offers free tier with $200 credit that never expires, no credit card required to sign up. Free tier includes access to all public models (Flux, Nova-3) and all endpoints (STT, TTS, Voice Agent, Audio Intelligence) at full concurrency limits (150 WebSocket STT, 50 REST, etc.). Developers can build and test production applications without payment until credit is exhausted.

Deepgram offers two pricing models: pay-as-you-go (per-minute consumption) and Growth tier (pre-paid annual credits with 10-20% discount). Pay-as-you-go pricing ranges from $0.0048/min (Nova-3 Monolingual) to $0.0078/min (Flux Multilingual) for STT. Growth tier offers same models at discounted rates ($0.0042-$0.0068/min) with pre-paid annual commitment. Pricing is per-minute of audio processed, not per request.

Deepgram CLI is a command-line tool with 28 built-in commands for transcription, synthesis, and management tasks. Includes integrated MCP (Model Context Protocol) server, enabling AI agents to call Deepgram APIs directly without custom integration code. CLI supports both interactive and scripted usage, with output formatting options (JSON, text, etc.).

Deepgram provides official SDKs for Python, JavaScript, Go, .NET, and Java. SDKs abstract HTTP/WebSocket complexity, handle authentication, manage connection pooling, and provide language-idiomatic APIs. Feature parity across SDKs is claimed but not verified — specific version numbers and supported features per SDK not documented.

Automatically identifies spoken language from audio without pre-configuration, supporting 45+ languages in Nova-3 Multilingual model or 10 languages in Flux Multilingual for real-time. Detection happens during initial audio processing; language is returned in response metadata and used to optimize transcription accuracy for that language's phonetics and vocabulary.

Flux models detect natural speech boundaries and turn-taking in conversations, automatically identifying when a speaker has finished talking and when another speaker begins. Built-in interruption handling allows overlapping speech to be processed without requiring explicit silence detection thresholds. Enables voice agents to know when to stop listening and trigger response generation without timeout-based heuristics.

Voice Agent API combines speech-to-text, LLM integration, and text-to-speech in a single WebSocket connection, orchestrating the full conversational loop. Audio input flows to Flux STT model, transcript is sent to configured LLM (provider UNKNOWN), LLM response is streamed to TTS model, and synthesized audio is returned to client — all within one persistent connection without intermediate API calls.

Converts text to natural-sounding audio via REST or WebSocket API. Supports streaming text input (partial text can be sent before full response is available), enabling real-time audio generation as LLM generates response tokens. Multiple voices and languages available (specific count and list not documented). Synthesized audio is returned as audio stream (format UNKNOWN).

Audio Intelligence endpoint analyzes transcribed speech to detect emotional tone and sentiment (positive, negative, neutral). Processes audio or transcript to extract sentiment signals, returning sentiment labels and confidence scores. Operates as post-processing on transcription output or as standalone analysis on pre-transcribed text.

Audio Intelligence endpoint automatically identifies topics and themes discussed in audio conversations. Analyzes transcribed speech to extract key topics, categorize conversation content, and return topic labels with relevance scores. Enables automatic routing, content classification, and conversation summarization without manual tagging.

Audio Intelligence endpoint generates abstractive summaries of audio conversations, condensing key points and action items from transcribed speech. Summarization operates on full transcript or speaker segments, extracting essential information and generating concise natural language summaries without manual review.

Allows developers to provide domain-specific keywords or phrases that the STT model should prioritize during transcription. Keyterm prompting biases the model's decoding toward specified terms, improving accuracy for technical jargon, product names, or domain-specific vocabulary that might otherwise be misrecognized. Implemented as optional parameter in transcription requests.