How to Make Artificial Intelligence Recognise the Sounds that Contribute to Soundscape

Artificial intelligence (AI) is increasingly adept at recognising sounds and acoustic scenes, and generative AI can create meaningful narratives about them. However, in everyday environments, humans do not notice or pay attention to all sounds. This selective attention reflects nature's strategy to avoid overloading the energy-expensive human cognitive system. The biological mechanism of focusing on relevant sounds while filtering out irrelevant or repetitive stimuli is not well-matched in most current AI systems. Sounds stand out in the sonic environment due to their salience. A sound can be salient because of its (1) acoustic characteristics such as loudness, sharpness, and tonality; (2) semantic incongruity or unexpectedness; or (3) special meaning to an individual or society. Prior work focused on the first form of salience and integrated it into artificial soundscape analysis. Here, we report primarily on the latter forms. We identify sounds that would typically be recognised as foreground, salient sounds by a person. A processing layer is then added to the sound recognition system to identify these salient sounds and further process them into soundscape perception, thereby creating context awareness in the AI.