NVIDIA Updates Riva Speech, Recommender System, and Hyperscale Inference

NVIDIA has updated its NVIDIA AI platform, including its speech, recommender system, hyperscale inference, and more.

The company also announced the NVIDIA AI Accelerated program, which helps ensure performance and reliability of AI applications developed by NVIDIA solution partners.

"NVIDIA AI is the software toolbox of the world's AI community, from AI researchers and data scientists to data and machine learning operations teams," said Jensen Huang, founder and CEO of NVIDIA, in a statement. "Our GTC 2022 release is massive. Whether it's creating more engaging chatbots and virtual assistants, building smarter recommendations to help consumers make better purchasing decisions, or orchestrating AI services at the largest scales, your superpowered gem is in NVIDIA AI."

NVIDIA AI includes NVIDIA Riva for speech AI and NVIDIA Merlin for smart recommendations. Updates have also been made across the software suite, including tools such as the NVIDIA Triton, NeMo, Maxine, and TAO Toolkit.

NVIDIA AI is comprised of the following:

• NVIDIA Riva 2.0 - A speech AI SDK that includes models pretrained with leading recognition rates, enabling developers to customize real-time speech AI applications for their industries. Riva 2.0 includes speech recognition in seven languages, deep learning-based text-to-speech with both male and female voices, as well as custom tuning with NVIDIA TAO Toolkit. NVIDIA also announced NVIDIA Riva Enterprise, a paid offering.
• NVIDIA NeMo Megatron 0.9 - A framework for training large language models, it enables users to train any model to convergence and scale to trillions of parameters for applications such as conversational AI, recommenders, and genomics. The latest version includes new optimizations and recipes that shorten end-to-end development and training time, and adds support for training in the cloud.
• NVIDIA Merlin 1.0 - An accelerated, end-to-end recommender AI framework to build recommenders at scale, which includes two new libraries: Merlin Models and Merlin Systems. These allow data scientists and machine learning engineers to determine which features and models are bestfor their use cases and deploy recommender pipelines as microservices.
• NVIDIA Maxine - An audio and video quality enhancement SDK with AI and acoustic echo cancellation and audio super resolution.
• NVIDIA Triton - An open-source hyperscale model inference solution. The latest release contains a Model Navigator for accelerated deployment of optimized models, Management Service for efficient scaling in Kubernetes, and Forest Inference Library enabling inference on tree-based models.

NVIDIA AI Enterprise 2.0 also introduces more NVIDIA AI software containers to support training and inference. Support for the NVIDIA TAO Toolkit allows enterprise developers to fine-tune and optimize NVIDIA pretrained AI models, simplifying the creation of custom, production-ready models.