Deepgram

The energy at #DeveloperWeek is 🔥 We've been connecting with builders, partners, and teams who are shaping what's next in voice AI — real use cases, bold ideas, and a lot of momentum. There are only a few hours left to come say hi. Let's talk about what you're building! #VoiceAI #AIBuilders #PoweredByDeepgram

In 2026, Voice AI Agents are powering real ROI — from faster service and 24/7 support to measurable cost savings and smoother workflows — across healthcare, quick-service restaurants, contact centers, and more. Curious how Voice AI Agents work, where they deliver the biggest impact, and what you should consider before implementing one? 📈 👉 Read the complete guide here: https://lnkd.in/g_PC8S-n #VoiceAI #ConversationalAI #CustomerExperience #Automation #AI #SpeechRecognition #NLP #TechTrends #BusinessTransformation

‼️ Demo video below: Arabic is now available on Nova-3 — across 17 dialects. Nova-3 is built to capture real-world Arabic as it’s actually spoken around the world. That means higher accuracy, better understanding of regional nuance, and more reliable Voice AI experiences. Whether you're building call center automation, AI voice agents, or transcription workflows, you can now deploy Arabic with confidence. 🎥 Watch the demo to hear it in action. 🚀 Try it yourself in the Playground: https://lnkd.in/gwYmvWEb #VoiceAI #SpeechToText #Arabic #AI #Developers

In financial services, "car" vs. "card" isn't a typo. It's a liability. Gradient Labs built AI voice agents for banks and financial institutions and needed transcription accurate enough to trust without question. Deepgram Nova-3 delivered: fewer "can you repeat that?" moments, SOC 2 compliance, and EU data residency out of the box. Read the case study → https://lnkd.in/eyGSwzqQ

🚀 Text-to-Speech APIs are no longer a niche; they’re essential for voice-enabled apps, accessibility tools, and real-time AI experiences. But with so many options out there, how do you choose the right one in 2026? Our latest guide breaks down the top Text-to-Speech APIs, comparing latency, voice quality, use cases, pricing, and integration tradeoffs to help you pick what fits your project. 📊 👉 Dive into the full list here: https://lnkd.in/giufzuYn #TextToSpeech #TTS #VoiceAI #APIs #ConversationalAI #SpeechTechnology #DeveloperTools #AI

⚡ Thanks for having us! Deep learning in noisy environments and the ethics of AI voice cloning are two extremely important topics to discuss. We hope everyone who watches this pivotal interview can take home important ideas from it. Thanks again to Stack Overflow and AWS re:Invent!

We made voice a first-class citizen in OpenClaw. Last week, we made it easy to call your existing Moltbot, but that still required manual setup. This week, we’re making it as simple as calling your friend on the phone. We also made some optimizations to OpenClaw itself to improve the voice-first experience. Call 910-MOLTBOT. In under a minute, you'll have your own OpenClaw instance running in the cloud — with a personal phone number, persistent memory, and full agent capabilities. Call it. Text it. It remembers everything. No git clone. No API keys. No Docker compose. No signup form. You just call a phone number. Under the hood: Deepgram's Voice Agent API handles speech with semantic turn detection (Flux) and 90ms TTS (Aura-2). DeepClaw bridges it all to OpenClaw — tools, memory, web search, code execution. This is an experiment from Deepgram Labs. It's free, provided as-is, and we're building in public. We want to see what happens when an AI agent is as easy to reach as a friend on speed dial. Try it: 910-MOLTBOT (910-665-8268) Open source: https://lnkd.in/gt6QhJvA

The future is real-time, voice-first experiences. We’re proud to partner with Cloudflare to make building Voice AI more accessible. Deepgram's voice models run natively on Cloudflare's AI Cloud, deployed across 330+ locations worldwide. Build faster, more securely, and at global scale without infrastructure complexity.

Nova-3 Multilingual Speech-to-Text just got a major upgrade. The improved production model delivers: ✅ Lower Word Error Rate (WER) across batch and streaming ✅ Significant improvements in code-switching scenarios ✅ No API or configuration changes required 📉  That’s a ~34% relative reduction in batch mean WER and a ~21% relative reduction in streaming mean WER The updated model is now the default Nova-3 Multilingual production model. Blog post in the comments 👇

🚀 Hebrew, Persian, and Urdu speech-to-text are now available on Nova-3. These new monolingual language models are live in production and extend coverage across the Middle East and South Asia – delivering Nova-3 level accuracy, Keyterm Prompting, and support for both streaming and batch workflows. Designed for call centers, voice agents, media transcription, and analytics, this release further establishes Nova-3 as the foundation for global voice applications. Full launch blog in the comments →