Deepgram Nova 3 is the latest iteration in Deepgram's Nova ASR lineage, advancing accuracy and robustness beyond Nova 2. It extends multilingual capabilities, tightens word error rates on accented and spontaneous speech, and improves timestamp precision — all while maintaining the low-latency characteristics that Deepgram's architecture is known for.
Nova 3 targets enterprise applications that have already adopted Nova 2 and need incremental quality gains, as well as new deployments where transcription accuracy directly impacts downstream workflows. Its improved handling of overlapping speech and diverse acoustic conditions makes it relevant for complex real-world audio environments.
Key Features
Improved word error rate over Nova 2 on diverse speech
Enhanced multilingual transcription support
More precise word-level timestamps for alignment tasks
Better handling of overlapping and spontaneous speech patterns
Low-latency streaming transcription maintained from prior generation
Improved robustness in challenging acoustic environments
Ideal Use Cases
Enterprise call analytics requiring high accuracy
Multilingual customer support transcription
Subtitle and caption generation with precise timing
Voice-to-text for accessibility and assistive technology
Audio content indexing and search at scale
Example Prompts for Nova 3
Technical Specifications
| Provider | Deepgram |
| Category | Audio |
| Modality | Audio -> Text |
Frequently Asked Questions
Try Nova 3 now
Start using Nova 3 instantly — 100 free credits, no credit card required. Access 750+ AI models through one platform.