Deepgram Nova-2

Real-time speech recognition API with sub-300ms latency and high accuracy

Audio & SpeechFreemium

Deepgram Nova-2 is a speech-to-text API delivering state-of-the-art accuracy combined with extremely low latency for real-time transcription applications. Nova-2 supports 30+ languages, handles noisy audio environments, provides speaker diarization, and returns structured transcripts with timestamps. The API is used for voice agents, meeting transcription, captioning, and voice command systems. Nova-2 significantly outperforms Whisper-based solutions on both accuracy and speed benchmarks for real-time workloads.

Key Features

• ✓Sub-300ms latency
• ✓30+ languages
• ✓Speaker diarization
• ✓Noisy audio handling
• ✓Timestamped output
• ✓Real-time streaming