Home

This grant will involve the development of a diarization system, using state-of-the-art methods that can be adapted to the constraints imposed by privacy-preserving techniques. This system will consist of different modules: - a pre-processing/feature extraction stage; - a Voice Activity Detection module; - a Speaker Modelling module, using x-vector embeddings obtained from a neural network trained for Speaker Verification; - a final clustering stage. This grant will also include procuring and preparing data for the training and development of the system’s modules, as well as for their evaluation.