Data Sonification for Anomaly Detection - Prototype
Prototype v. 3 - Overview
The current version of our Data Sonification prototype in modelled on the needs of real-time monitoring of anomalies in a medium-sized industrial Internet network. The anomaly detection algorithm distinguish several types of data, among which fields (aggregation of the network’s components and variables); field impacts (the weight on the system’s behaviour on each aggregation); anomaly index (the overall level of anomaly present by the network in each moment); anomaly true/false (when the algorithm identified an anomaly as “True” vs non malicious or non identified anomalous behaviour). The sonification tool represents these types of data through different sounds and their behaviour. In particular, as seen in the video, the prototype represents types of data as follows:
Field 1 = Sound of birds
Field 2 = Sound of wind (leaves moved by the wind)
Field 3 = Sound of insects
Anomaly Index = Sound of rain
Anomaly True = Sound of thunder
The totality of the sounds compose a natural soundscape which is thought to merge nicely with an office environment thus remaining at the periphery of attention for several hours (unless an emergency occurs) without burdening the listener. At the same time, using natural, real sounds leverages our biological ability in detecting alarming events or events that create a disturbance in the environment therefore need our attention. In this case a real emergency (a cyber-attack or a major failure of the network) is represented sonically by a thunderstorm approaching. In normal conditions, the intensity of each field is determined by the weight that each field has on the behaviour of the network at any given time, as represented by the incoming data.
The prototype is programmed in Max/MSP and uses sounds designed by us. It plugs directly into the artificial intelligence algorithm’s data sets which are streamed in real-time and converted into sound parameters.
Disclaimer: in the video, you will see the Thunder (Anomaly True) being activated manually. This is due to a last minute issue in the data set’s threshold which we are solving right now, thanks for your patience!