Deepgram
64 tracked updates · May 2026
13 feature updates·8 new integrations·3 product launches·3 content pieces·2 case studies·1 FEATURE UPDATE·1 legal & regulatory updates·1 PRICING UPDATE
Browse other months from Deepgram
Deepgram introduced the managed Google LLM model `gemini-3.1-flash-lite` in its Voice Agent API, replacing the preview version. The deprecated `gemini-3.1-flash-lite-preview` will be removed on May 26…
Call center compliance in 2026 will require managing four regulatory layers: federal (TCPA, HIPAA, PCI DSS 4.0), state AI disclosure laws (Utah, California, Texas), international rules (GDPR, EU AI Ac…
Deepgram highlights a critical gap where playground TTS demos mask pronunciation failures that emerge in production, particularly with raw user inputs like numbers, proper nouns, and domain terms. The…
Deepgram released a comprehensive guide explaining how medical voice recognition works in clinical settings, including HIPAA compliance requirements, EHR integration challenges, and accuracy benchmark…
Google’s official deprecation page outlines the end-of-life timelines for numerous stable and preview models in the Gemini API, including Gemini 3, 2.5 Pro/Flash, 2.0, and others. Shutdown dates are t…
Google’s official deprecation page outlines the end-of-life timelines for numerous stable and preview models in the Gemini API, including Gemini 3, 2.5 Pro/Flash, 2.0, and others. Shutdown dates are t…
Deepgram introduced Flux Multilingual, the first conversational speech recognition model supporting multiple languages without routing. The model expands speech-to-text capabilities to Thai, Cantonese…
Deepgram’s Aura-2 TTS voices now support runtime speed (0.7x–1.5x) and pronunciation overrides via inline IPA notation in English and Spanish. These controls are available on batch and streaming endpo…
Deepgram expanded its Numerals feature to three new languages—Russian, Romanian, and Hebrew—using monolingual models. The update allows spoken numbers to be converted to digits in transcripts via the …
Deepgram’s Nova-3 speech-to-text model now supports Thai, Cantonese, Mandarin (Simplified and Traditional), and improved accuracy for Bengali, Marathi, Tamil, Telugu, and Gujarati. These additions tar…
Jobcase adopted Deepgram’s Aura-2 text-to-speech to enhance its AI voice agents, reducing latency and improving naturalness in calls for job-seeking members. The integration supports both inbound and …
Klubi, a Brazilian digital consórcio platform, scaled voice-led growth using Deepgram’s Nova-3 speech-to-text to automate pre-sales, qualification, and post-sales workflows. The integration enabled re…
Creditas, an Indian digital debt collections platform, adopted Deepgram’s speech-to-text API to automate and enhance collections while ensuring compliance and trust. The solution provided 100% call au…
Abby Connect, a 24/7 human virtual receptionist service, integrated Deepgram’s speech-to-text API to power its new AI receptionist, automating repetitive tasks like scheduling and FAQs. The integratio…
Vida, an AI Agent OS for enterprises, selected Deepgram’s Aura-2 TTS and multilingual STT to power high-volume healthcare voice agents, citing superior naturalness, low latency, cost predictability, a…
A global cloud communications platform serving 19,000+ businesses replaced its in-house Whisper-based transcription with Deepgram’s AI speech-to-text platform on AWS. The move addressed scalability, a…
SigmaMind AI integrated Deepgram’s Nova-3 and Flux speech-to-text models to power its no-code voice AI platform, reducing end-to-end agent response latency by 300ms and enabling mid-utterance API call…
A Fortune 50 U.S. retail pharmacy chain replaced its legacy Nuance IVR with Deepgram’s Nova Medical speech-to-text and Aura text-to-speech to handle over 1 million pharmacy calls daily across 7,000+ l…
GetVocal AI integrated Deepgram’s real-time streaming speech-to-text into its voice automation platform to support production-grade telephony interactions. The integration improved structured entity c…
Deepgram launched profanity filtering for over 50 languages, enabling automatic detection and redaction of offensive language in transcripts via a simple API parameter. The feature targets cleaner, sa…
Deepgram introduced the Browser Agent SDK, four composable npm packages enabling rapid integration of voice agents into any web app. The SDK abstracts complex audio handling, reconnection logic, and s…
Deepgram introduced Flux Multilingual, a Voice AI model enabling real-time code-switching across 10 languages for restaurant ordering. The feature maintains monolingual-grade accuracy and latency whil…
Maki partnered with Deepgram to embed real-time streaming speech-to-text at the core of its voice pipeline, enabling AI hiring agents to conduct natural, responsive candidate conversations with high a…
Deepgram launched Flux, a conversational speech recognition model, and partnered with Lindy to power Gaia, a no-code AI voice agent for handling business calls. Flux enables ultra-low latency and natu…
Deepgram updated its Nova-3 Portuguese model to improve transcription accuracy for both Brazilian and European Portuguese variants. Users can now leverage the enhanced model by specifying `model="nova…
Deepgram published a detailed 2026 comparison of its speech-to-text service against Amazon Transcribe, highlighting differences in accuracy benchmarks, streaming latency, custom vocabulary features, p…
Deepgram explains when to use its direct API versus Twilio’s managed paths (Gather, ConversationRelay) for real-time transcription. Direct API access unlocks full STT control, Keyterm Prompting, and m…
A comparison of self-hosted speech-to-text (STT) options from Deepgram, Speechmatics, AssemblyAI, AWS, and Google Cloud highlights varying levels of data control, air-gap support, and compliance certi…
Deepgram introduced AI drive-thru ordering using its Nova-3 speech-to-text model, trained on real drive-thru audio, achieving a 5.26% word error rate. The system integrates with POS in real time and u…
Deepgram shipped SDK updates across JavaScript, Rust, Python, and Java, adding Flux multilingual support in Rust, restoring the Agent interface in JavaScript, fixing WebSocket query parameters in Pyth…
Deepgram introduced a new Browser Agent SDK with four composable packages—Widget, React UI Components, React Hooks, and JavaScript SDK—enabling quick integration of voice agents into web apps. Each pa…
Deepgram introduced the Browser Agent SDK, offering four composable packages to connect web apps to the Voice Agent API, including a drop-in widget and React components. The SDK simplifies integration…
Pricing updated for Deepgram: - New tier: Pay-As-You-Go (Custom pricing) - New tier: Enterprise (Custom pricing) - Removed: Free tier - Removed: Custom tier
Deepgram introduced Deepgram for Restaurants, a Voice AI system designed to address the QSR industry's labor-driven margin crisis. Rising wages and 130% annual turnover are crippling QSR operators, le…
Deepgram introduced three agentic engineering tools—the dg CLI, MCP server, and deepgram/skills repo—to streamline voice AI development in AI coding tools like Claude Code and Cursor. These tools auto…
Deepgram’s TTS team will host a live webinar on May 5, 2026, to teach engineers and product teams how to build scalable TTS evaluation pipelines for voice agents. The session covers defining scoring c…
The article provides a framework for evaluating voice AI platforms in banking, emphasizing accuracy under real-world noise, compliance architecture, and cost predictability at scale. It highlights Dee…
Deepgram published a comprehensive buyer's guide comparing AI voice agent services based on latency, noise tolerance, pricing, and compliance. The guide emphasizes testing under real-world conditions,…
A third-party guide ranks Vapi alternatives for voice applications based on STT accuracy, pricing transparency, and deployment flexibility. Deepgram is highlighted for production-grade STT and bundled…
Track Deepgram on autopilot