MARVEL D4.4: Optimal audio-visual capturing, analysis and voice anonymisation

This report focuses on the final version of the optimal audio-visual capturing, analysis and voice anonymisation. With regard to optimal audio capturing, the first section of the deliverable offers a detailed description of the development progress of the hardware used within the scope of this project, since the initial version of the document. Moreover, the applied methodologies to provide pre-processing and data analysis through the usage of Edge AI techniques are presented. The last part describes the deployment process of the final version of the hardware within the different project use cases as well as the final experimental results. The second section presents the progress made with regard to intelligent audio analysis with the Voice Activity Detection module within the scope of devAIce SDK, the modular technology encapsulating multiple AI modules designed for cross-platforms deployment, as well as the upgrades applied to the data collection and user annotations toolkits, SensMiner and the AVER application. A third section in this document describes the toolkit used to anonymise the sensitive speaker information, AudioAnony, as well as the approach adopted to design and build the voice anonymisation pipeline on the edge along with a detailed description of the deployment process along the different use cases. Furthermore, the progress of the KPIs set for the tasks and the relevant components is discussed in a separate section.