In a Nutshell
Company:
Fondazione Bruno Kessler
Open Source License:
CC BY 4.0
About the Repository
Functionality
The component includes a suite of energy- resource- and data-efficient neural networks for audio/speech processing. The networks are designed to be cloud-edge native. Being flexible in their computational compute they can be deployed in a variety of nodes and can adjust to the operational conditions and available resources.
The component is under development as the project ends in 2029.
IPCEI-CIS Referene Architecture
AI Layer, Sustainability Domain
Keywords
Dynamic neural models (layer dropping, early exits), Efficient neural models (custom conformer; custom splitformer), Speech recognition
