Deepgram

Voice is becoming the primary interface for AI. So what does it take to make those conversations actually feel natural? In this interview from theCUBE + NYSE Wired’s Mixture of Experts series, Deepgram CEO and co-founder Scott Stephenson joins John Furrier to discuss the rise of real-time voice agents and the launch of Flux TTS, Deepgram’s context-aware text-to-speech model. Scott breaks down how Flux TTS uses conversational context, including turn-taking cues, tone shifts, and prior dialogue, to generate more natural responses with latency as low as 80 milliseconds. He also explores the architecture behind real-time voice AI, the role of fast response models alongside deeper reasoning systems, and how Deepgram supports deployment across public APIs, AWS, and customer-controlled environments. The conversation also covers Deepgram’s growth to more than 200,000 developers and $100M in ARR, the 2,000+ products already built on its platform, and what’s next as Deepgram expands its voice AI capabilities globally. Topics include: Flux TTS, conversational AI, real-time voice agents, context-aware speech generation, AI infrastructure, latency, enterprise deployment, and the future of voice as an AI interface.

Building real-time voice apps like call centers, drive-thrus, or customer service bots? We put two text-to-speech (TTS) platforms head-to-head — Deepgram vs ElevenLabs — to see which one delivers the accuracy, latency, cost, and scalability enterprise voice AI actually needs. ElevenLabs shines in media and entertainment (dubbing, narration, cinematic voiceovers), but when it comes to live AI conversations with customers, we break down where Deepgram was built to win. ---------------------------------- ▶️ Try Deepgram free: https://deepgram.com 📚 Docs: https://developers.deepgram.com 🚀 Get started: https://console.deepgram.com/signup ---------------------------------- In this comparison we cover: ✅ Accuracy on technical terms (Kubernetes, PostgreSQL, quantum entanglement, HIPAA) ⚡ Ultra-low latency for natural, real-time conversations (sub-1-second response) 💰 ~40% cost savings vs ElevenLabs Flash for enterprise voice workloads 📈 Enterprise-grade scalability for thousands of concurrent conversations 🏆 Blinded evaluation results — why users consistently preferred Deepgram's voices If you're a developer or tech leader choosing a TTS API for voice agents, this breakdown will help you pick the right one for production. Working on a voice project? Drop it in the comments — we'd love to see what you're building. 👍 Like this video and subscribe for more voice AI tutorials and comparisons. #Deepgram #ElevenLabs #TextToSpeech #TTS #VoiceAI #SpeechSynthesis #VoiceAgents #AIVoice #DeveloperTools #ConversationalAI