ABSTRACT
In very many contexts, people not only interact with technical systems and the external physical world. People inter act with people. Much of this interaction goes through the auditive channel. We present and discuss the outlines and the first steps in the technical realization of a complex suite of systems that monitor the auditive channel to gather input for contextual workload assessment. We apply the suite to monitor the human-originating acoustic events in side a passenger car. First, we sketch the acoustic environment inside a vehicle. We discuss the microphone sensor equipment and some of the signal processing techniques employed in order to separate audio sources, needed to find out who of the peo ple in the car is talking at any given moment. We then discuss the nature of the auditory events, explaining why we chose to first concentrate on a set of so-called ‘para-linguistic’ events, non-speech sounds generated by the human articulatory apparatus. We outline how speech recognition technology was expanded to handle the wide frequency range needed. We proceed to show how we overcome the inherent uncertainty in ascribing certain categories to a set of events from a certain source over a cer tain time. We show how the technologies mentioned above are integrated in a car and how, using the vehicle’s data bus, a communication is established with a contextual workload assessment module that also receives input from a set of other data sources. We conclude by outlining ongoing and further work on audio monitoring, highlighting a few application scenarios.
