Nova-3
Deepgram's flagship speech recognition model, supporting multilingual, high-accuracy, low-latency real-time transcription.
Timeline (last 90 days)
Deepgram worked with the AWS Forward Deployed Engineering team to port its Nova-3 speech-to-text model to AWS Trainium2 chips for offline batch inference.
Nova-3 is Deepgram's flagship speech recognition model, supporting multilingual, high-accuracy, low-latency real-time transcription.
During the porting process, the Deepgram team verified the numerical correctness and performance of Nova-3's custom operators on Neuron one by one, and implemented fallback for some operators through Neuron's custom operator interface.
According to official Deepgram and AWS blog posts, after migrating to Trainium2, Nova-3's batch transcription throughput increased significantly and the cost per unit of audio processed dropped noticeably, but no precise benchmark numbers were published.
All Facts (4)
Nova-3 is Deepgram's flagship speech recognition model, supporting multilingual, high-accuracy, low-latency real-time transcription.
50%UnverifiedAccording to official Deepgram and AWS blog posts, after migrating to Trainium2, Nova-3's batch transcription throughput increased significantly and the cost per unit of audio processed dropped noticeably, but no precise benchmark numbers were published.
50%UnverifiedDeepgram worked with the AWS Forward Deployed Engineering team to port its Nova-3 speech-to-text model to AWS Trainium2 chips for offline batch inference.
50%UnverifiedDuring the porting process, the Deepgram team verified the numerical correctness and performance of Nova-3's custom operators on Neuron one by one, and implemented fallback for some operators through Neuron's custom operator interface.
50%