Nova 2 EARLY ACCESS

[lukeocodes]

Introducing Nova-2 Early Access - our next generation speech-to-text model

At Deepgram, we are constantly pushing the boundaries of what’s possible in Language AI. I’m excited to share our latest model for superhuman speech-to-text, Nova-2.

Nova-2 achieves:

• Next-level accuracy for pre-recorded and real-time audio: 18% more accurate than Nova-1 and 36% more accurate than OpenAI Whisper (large)
• Unmatched speed: Up to 40x faster than alternatives
• Same low costs as Nova: 3-5x lower costs

Link to blog: https://deepgram.com/learn/nova-2-speech-to-text-api

[bmaluijb]

Really cool! Will the Json result stay the same (utterances, paragraphs, etc..), as I'm now relying on that in my application.

[lukeocodes]

Yes our transcription response never differs between ASR models!

[JackBarker21]

Hi,

I would like to test the new nova-2.

What do I need to use in my Python script?

options = { "punctuate": True, "diarize": True, "model": '', "tier": '' }

[lukeocodes]

You only need to change the model from nova to nova-2-ea, but in the future nova will just default to the latest version!