The LiquidInfo project is researching the development of a secure, low-threshold and efficient provision of AI models for the application of media transformations. Audio-visual information, e.g. from meetings or events, is to be semantically enriched to increase transparency and inclusion and made available in various languages and formats – for barrier-free communication and documentation of information across all modalities in a protected space.
Project description
Initial situation
AI-based services and applications are firmly integrated into our everyday working lives. Most of them are cloud-based and proprietary. SMEs are confronted with several challenges: Services in the cloud are problematic for data protection and data sovereignty, their use makes them dependent on closed formats and platforms, the licence models involve high costs and integration into internal business processes often has to be done manually instead of at data level - or the company's own processes have to be laboriously adapted to the specifications of external APIs. The social costs of massive energy consumption in permanently running cloud data centres are also often underestimated.
Project goal
The project addresses these problems by developing the LiquidInfo-Box; a combination of harmonised hardware and software components for energy-saving mobile use. By simply connecting an HDMI signal, the box enables automatic transcriptions, translations, object recognition, processing, analyses and summaries of meetings, presentations or other events in a variety of formats. Data processing takes place entirely on the device, which guarantees data sovereignty, reduces dependencies on external providers and energy consumption and also greatly simplifies set-up. The open, semantically enriched data structure allows flexible media switching of content for greater accessibility and inclusion.
Implementation
The implementation as an offline on-premise solution enables a data protection-compliant implementation. As the data processing takes place locally on the device without the involvement of external providers, no consent is required for the external processing of data. This makes it possible to use the solution in strictly confidential situations. The information encoded in the AV data is extracted and converted into an abstract data model. This enables further processing across all modalities, flexibilisation and individualisation of the output.
