Sources
Loading...
Additional media
Loading...
DeepNewz, mobile.
People-sourced. AI-powered. Unbiased News.
Deepgram has announced the launch of Nova-3, its latest speech-to-text technology, which enhances accuracy for enterprise applications. The new model boasts a 54.2% improvement in streaming accuracy and a 47.4% increase for pre-recorded audio compared to its closest competitor. Nova-3 introduces real-time multilingual capabilities and allows for self-serve customization, enabling users to adapt vocabulary without the need for retraining. The release is part of Deepgram's ongoing efforts to lead in the voice AI sector, following the recent release of Ultravox v0.5, which has also shown competitive performance against proprietary models like GPT-4o Realtime and Gemini 1.5 Flash in speech understanding benchmarks.
Deepgram Nova-3 is now available with Vapi. What changes for Vapi builders? 1) Superior accuracy: 54.3% reduction in word error rate (WER) for streaming 2) Real-time multilingual transcription 3) Instant vocabulary adaptation without model retraining https://t.co/dCoYcQKO6F https://t.co/F6fkUYQG4N
Nova-3, our most advanced speech-to-text, is here! ⚡️ ✅ Leads in accuracy by 54.2% (streaming) and 47.4% (pre-recorded) vs. next-best ASR ✅ New real-time multilingual capabilities ✅ Self-serve customization (vocabulary adaptation without retraining) See the benchmarks (🧵)… https://t.co/PFXz6GOfNt
Deepgram Launches Nova-3, Enhancing Speech-to-Text Accuracy for Enterprise AI https://t.co/GqLchD1GlY #AIwire https://t.co/grIHna0vQF
