Deepgram

In healthcare, every word matters. The difference between “????????????????????????” and “??????????????????????” is critical-yet healthcare technology companies have long struggled with general-purpose models that misinterpret specialized terminology, generating errors that compromise patient care. Today, we are thrilled to introduce the most accurate medical transcription model in the world, Nova-3 Medical. Nova-3 Medical is built to capture clinical interactions with the highest transcription accuracy, reducing word error rates by 63.7% compared to next-best ASR. With Nova-3 Medical, developers can easily integrate reliable transcription into AI healthcare applications for patient consultations, appointment scheduling, and more. TL;DR:? ?? Capture medical terminology, acronyms, and jargon (i.e. symptoms, prescriptions, etc.) with precision ?? Fine-tune for specific medical specialties using Keyterm Prompting to boost accuracy ?? Maintain HIPAA compliance with flexible deployment modes including on-premise and VPC configurations ?? Transcripts integrate seamlessly with EHR systems and existing workflows, without the latency and cost barriers of general-purpose models From ambient scribes to voice agents, our customers like Lyrebird Health and Chartnote are using Deepgram to build the future of healthcare! Dive deeper into the benchmarks from this release: https://lnkd.in/gacF4n39 If you're at #HIMSS this week, stop by the Deepgram booth (#136) to learn more and see Nova-3 Medical in action. Get started with Nova-3 Medical in our API Playground: https://lnkd.in/gAXJeQUh

OpenAI released gpt-4o-transcribe and gpt-4o-mini-transcribe yesterday to a lot of fanfare. We immediately tested them with extremely disappointing results, especially compared to Deepgram Nova-2, which was released almost 2 years ago. Voiceflow wants to deliver the best conversational experience on our platform and speech-to-text is a very critical piece of the stack. Often it's easy to get caught up in hype cycles and great looking benchmarks. But sometimes you can just easily tell if a conversation feels natural or not, and the best way is just to start speaking. Feel free to reproduce my results yourself, or tweak the settings where needed. https://lnkd.in/eVMpubfh

It’s our last day at #GTC25 but there’s still plenty of time to catch a demo of our AI voice agent and snag a branded Owala while you’re at it (while supplies last). ?? #NVIDIAGTC2025 #VoiceAI NVIDIA GTC

Need a hand with #VoiceAI? ?? Swing by our booth #1709 to learn more! Shout out to our AI hand model, Yorick from Cloudflare. ? #GTC25 #NVIDIAGTC25

It's always nice running into our customers and DG Startup Program members. Great time catching up with the Pactto team at #GTC25! ?? Swing by booth #1709 to chat, grab some swag, and see how devs are building with #VoiceAI. #NVIDIAGTC

Voice AI, GPUs, Deepgram... oh my! ?? If you're at #GTC25 follow the yellow brick road to our booth #1709. ??? Test our voice agent that makes AI conversations as seamless as clicking ruby slippers ?? Win a Founders Edition NVIDIA GPU (more powerful than a flying monkey) – Live drawing at our booth today at 6PM PT! There's no place like Deepgram's booth! Stop by and say hi ?? #NVIDIAGTC

Final day at the Enterprise Connect expo hall! There’s still a chance to enter our partner passport to prize! Make sure to drop off your passport before our 3PM drawing!

The Deepgram team is live at Enterprise Connect! Stop by our booth #1219 to see how Deepgram's Voice Agents and Nova-3 speech model are helping companies boost customer experience with natural, responsive voice interactions that transform CX outcomes. Participate in the Partner Passport to Prize program and enter for your chance to win one of five amazing prizes. ? Don't miss our lightning talk at the Twilio booth #811 today at 1:30PM: "Talk Like a Human: Mastering Conversational Pacing in Virtual Agents" - discover how to create more natural voice interactions. ?? Later, join us for our collaborative case study session with Twilio at 4:30PM at the EC Case Study Theatre, where we'll share real-world success stories you can apply to your business. See you on the show floor!

"Once you have these models, you need to orchestrate them so they all kind of sound and feel human." Some great insights from Jordan Dearsley, Founder & CEO at Vapi on the AI Minds Podcast discussing orchestrating Voice AI models for human-like interactions. From diving into workflows and handling real-time infrastructure for tens of thousands of calls, Vapi embeds Deepgram's Voice AI to achieve less than 200ms latency with impressive conversation quality.

Heading to Enterprise Connect in Orlando next week? Visit the Deepgram booth #1219 to see how our Voice Agent API transforms customer service, automates order processing, and handles appointment scheduling without human agents. Don't miss a thing onsite! Grab our exclusive session guide at the booth to catch all our game-changing voice AI demos and talks! ?? BONUS: Grab our Partner Passport to Prize and visit participating booths for your chance to win premium prizes! Winners will be announced live at our booth on March 19th at 3PM! See you next week!