Deepgram offers Voice AI APIs for speech-to-text, text-to-speech, voice agents, and audio intelligence in real-time and batch, cloud and self-hosted configurations.
Deepgram offers Voice AI APIs for speech-to-text, text-to-speech, voice agents, and audio intelligence in real-time and batch, cloud and self-hosted configurations.
Deepgram's speech-to-text API delivers ultra-accurate, low-latency transcription at enterprise scale, powered by Flux and Nova-3 models.
Deepgram delivers sub-200ms streaming text-to-speech built for voice agents with domain-specific accuracy and secure, scalable deployment across cloud and on-prem environments.
A single, unified conversational AI API for building enterprise-ready, cost-effective voice AI agents, combining speech-to-text, LLM orchestration, and text-to-speech in real time.
Extract vital insights and understanding from conversational audio at scale, powered by efficient task-specific language models for summarization, sentiment analysis, intent rec...
“Before Deepgram, Randall-Reilly found that when it audited its speech-to-text query results from other vendors, only 80% of the search results produced confirmable matches. Using Deepgram they were able to increase audit-confirmed results to levels exceeding 95% accuracy.”
BIGVU integrated the Deepgram API into its production video app to generate synchronized captions using batch transcription. Deepgram reports that BIGVU measured a 58% reduction in transcription latency after…
Case studyCreditas deployed Deepgram Nova speech-to-text in AWS Mumbai for Hindi-English collections calls, voicebots and call audits. Deepgram reports full call-audit coverage, a 93% reduction in compliance risk and a 27%…
Case studySharpen used Deepgram to replace an older transcription solution supporting contact-center coaching and compliance. Deepgram reports that testing existing calls showed improved transcription quality, with custom models…
Case studyAs reported by Deepgram. Ask for audit reports before you buy.
SubprocessorsAutoNotes, Bond, Caddy, Char, Chasi AI, Clueso, Conveo, Denki, Ember AI, Illume Labs, Juno, Level AI, Motives, Ontora, Phenom, Pickle, Plivo, Quotain, Regal, RingCentral, Inc., screenpipe, Sierra, Ezra AI Labs, Live Aware Labs, Jamworks, Eve Legal, ReflexAI, Nxtcure Labs
Claim the profile to keep facts current, answer buyers and submit evidence for the next ranking review.
Claim DeepgramDeepgram has released updated Nova-3 monolingual models, expanding language coverage in speech-to-text. New batch models are available for Kannada and Telugu. New streaming models are available for Greek, Kannada, and Telugu. These models are live for all users and do not require changes to existing API requests.
The Flux speech-to-text model now supports numeral formatting, which converts spoken numbers into numerical format (e.g., "nine hundred" to "900"). This feature is available for the English Flux model and the Flux Multilingual model (covering English, Spanish, French, German, Russian, Portuguese, Italian, and Dutch). Users can enable this by setting the 'numerals=true' query parameter during the initial WebSocket connection.
Deepgram released improved Nova-3 monolingual models for eleven languages, including German, French, Turkish, and Ukrainian, among others. The updates, which apply to both batch and streaming workloads for specific languages, enhance transcription quality and accuracy for all existing users without requiring modifications to API requests.
Deepgram released the July 2026 version of its self-hosted suite (260714). Key features include improved number and percentage formatting, such as maintaining trailing zeros in decimals, and enhanced Malay language numeral conversion. Additionally, the update changes how streaming sessions report no-audio timeouts, providing a specific 'no_audio_timeout' reason instead of generic closing signals to facilitate better error handling for client applications.
Deepgram has updated its Voice Agent features, introducing an 'interrupt' behavior for injected messages that allows messages to immediately override ongoing speech. The company also moved its LatencyReport feature from experimental to fully supported, providing automated, granular reporting on LLM, TTS, and end-to-end latency metrics for each turn in a conversation.
