Stephen Brewster

Ammar Al-Taie, Thomas Goodge, Shaun Macdonald et al.

Automated vehicles (AVs) must communicate their yielding intentions to pedestrians at crossings. External Human-Machine Interfaces (eHMIs, on-vehicle displays) are promising solutions, but were primarily tested with walking pedestrians. Runners are a significant pedestrian group who move faster and face distinct bodily and perceptual demands, raising questions about how pedestrian activity influences eHMI use. We conducted an outdoor study using an augmented reality simulator. Participants navigated a virtual crossing while walking and running; an approaching AV displayed one of three eHMIs: red/green colour-changing lights, animated cyan lights, or no-eHMI. No-eHMI consistently underperformed. Walkers mostly stopped and validated eHMI signals with vehicle behaviour; they processed both eHMI animations and colour changes effectively. Runners experienced greater time pressure to cross, increasing reliance on the eHMI over vehicle behaviour. They preferred colour changes over animation for rapid decisions. These findings are crucial for promoting eHMI inclusivity and physical wellbeing as AVs join our roads.

Satvik Venkatesh, David Moffat, Alexis Kirke et al.

Segmenting audio into homogeneous sections such as music and speech helps us understand the content of audio. It is useful as a pre-processing step to index, store, and modify audio recordings, radio broadcasts and TV programmes. Deep learning models for segmentation are generally trained on copyrighted material, which cannot be shared. Annotating these datasets is time-consuming and expensive and therefore, it significantly slows down research progress. In this study, we present a novel procedure that artificially synthesises data that resembles radio signals. We replicate the workflow of a radio DJ in mixing audio and investigate parameters like fade curves and audio ducking. We trained a Convolutional Recurrent Neural Network (CRNN) on this synthesised data and outperformed state-of-the-art algorithms for music-speech detection. This paper demonstrates the data synthesis procedure as a highly effective technique to generate large datasets to train deep neural networks for audio segmentation.