We address the problem of situated acoustic road event detection. We present a general probabilistic framework to detect and annotate acoustic events. The framework relies on two generative probabilistic models – a “good” model that will learn the characteristic on-the-spot (“situated”) acoustic dynamics, and a baseline model – on which we run signal processing tasks. We define the involved signal processing stages in the approach (parameter estimation, state estimation, and model comparison) as probabilistic inference tasks. We introduce generative models focusing on the characteristic acoustics on the road, allowing us to validate the framework for acoustic road event detection specifically. Our experimental results show that we are able to detect acoustic road events such as a crash and tire slipping using our presented probabilistic framework.