2026 年热门更新
Deepgram upgrades Nova-3 Medical batch model with expanded medical vocabulary
Deepgram released an upgraded Nova-3 Medical batch model with expanded medical vocabulary and improved medical term recognition (97.20% KRR). The update maintains word error rate parity and is availab…
Speech-to-Speech vs Cascade: Voice Agent Architecture
Deepgram’s guide compares Cascade and Speech-to-Speech (S2S) voice agent architectures, emphasizing tradeoffs in cost, debuggability, and compliance. Cascade pipelines expose text at each stage, aidin…
Dynamic Range Compression for Voice AI
Deepgram’s product marketing manager argues that dynamic range compression (DRC) is often unnecessary for voice AI pipelines and can degrade transcription accuracy. The article provides a decision fra…
AI Voice Agents in Healthcare: 7 Production Use Cases (and What Makes Them Work)
Deepgram’s article highlights how health systems deploy AI voice agents for scheduling, refills, and triage, emphasizing the critical role of the speech-to-text (STT) layer in production success. It d…
Everybody's building Voice AI for restaurants right now. Let's draw the map.
The restaurant Voice AI market is rapidly saturating with developers, tech platforms, and enterprise brands building solutions. Deepgram positions itself as the foundational speech recognition layer e…
Deepgram Self-Hosted May 2026 release adds profanity filtering and Korean spacing fixes
Deepgram’s May 26, 2026 self-hosted release (260528) introduces profanity filtering for multilingual Nova-3 models and improves Korean word spacing in transcripts. The update also preps deployments fo…
Pricing change detected for Deepgram
Pricing updated for Deepgram: - Pay-As-You-Go: promotion changed — Free $200 Credit - New tier: Custom (Custom pricing) - Removed: Growth tier
Voice Agents That Prioritize Data Security and Run Where Your Data Lives
Deepgram launched a Voice Agent API that integrates NVIDIA Nemotron models (Nemotron 3 Nano and Super) to enable sub-700ms end-to-end latency for voice agents deployed in customer VPCs, on-prem, or hy…
Deepgram launches self-hosted voice AI deployment with enterprise prerequisites
Deepgram introduced self-hosted deployment options for its voice AI services, targeting use cases with strict performance or security requirements. The offering requires an Enterprise Plan, direct lic…
Pricing change detected for Deepgram
Pricing updated for Deepgram: - Pay-As-You-Go: promotion changed — $200 free credit then pay-as-you-go - New tier: Growth ($4000/yr ($333.33/mo annually, ~12.5% off)) - Removed: Enterprise tier
AI Voice Agents Improve Patient Engagement: What the Evidence Actually Shows
Deepgram’s analysis finds no peer-reviewed studies validating vendor claims of 30–50% appointment booking lifts from AI voice agents in healthcare. Most evidence comes from vendor case studies with sh…
Hinglish: The Language 600M+ Indians Speak and Why Your Voice AI Keeps Failing
Deepgram introduced multilingual code-switching capabilities in its speech-to-text API to handle Hinglish, a blend of Hindi and English spoken by 600M+ Indians. The feature detects language shifts wit…
Why Word Error Rate Is Broken for Indian Languages: The BRIDGE 7-Metric Stack Explained
Deepgram argues that Word Error Rate (WER) systematically overstates errors for Indian languages due to morphological agglutination, script diversity, and code-switching. They propose the BRIDGE 7-met…
Evaluating Voice AI Agents for Healthcare: The Compliance and Accuracy Checklist You're Missing
Deepgram released a detailed checklist for evaluating voice AI agents in healthcare, emphasizing the intersection of HIPAA compliance and transcription accuracy. The guide highlights medical-specific …
What is Code-Switching? A Complete Guide for ASR Builders
Code-switching in speech can cause ASR error rates to spike up to 11x higher, with monolingual systems failing at language boundaries. Unified multilingual models and specialized metrics like PIER are…
Deepgram adds Gemini 3.5 Flash to Voice Agent API and deprecates older models
Deepgram introduced the managed Google Gemini 3.5 Flash model in its Voice Agent API, replacing the older Gemini 2.5 Flash family. The new model improves performance and efficiency, while the 2.5 Flas…
显示前 20 项。点击上方月份查看 Deepgram 的完整存档。