Deepgram Nova 2 is Deepgram's second-generation automatic speech recognition model, built on their end-to-end deep learning ASR architecture. It significantly improves on the original Nova in word error rate, speaker diarization accuracy, and real-time transcription throughput, making it one of the fastest commercially deployable ASR models available.
Nova 2 is optimized for production workloads where both latency and accuracy matter — call centers, live captioning, and voice-enabled applications. Deepgram's infrastructure-native approach means it runs efficiently at scale, and Nova 2 supports a broad range of English accents and speaking styles reliably.
Key Features
Low word error rate on diverse English speech samples
Real-time and batch transcription with low latency
Speaker diarization to distinguish multiple speakers
Streaming transcription for live voice applications
Robust performance on phone-quality and call-center audio
Custom vocabulary and keyword boosting support
Ideal Use Cases
Call center conversation transcription and QA
Live captioning for broadcasts and meetings
Voice command processing in real-time applications
Podcast and media transcription at scale
Customer service voice analytics pipelines
Example Prompts for Nova 2
Technical Specifications
| Provider | Deepgram |
| Category | Audio |
| Modality | Audio -> Text |
Frequently Asked Questions
Try Nova 2 now
Start using Nova 2 instantly — 100 free credits, no credit card required. Access 750+ AI models through one platform.