Abstract. This paper deals with the CLEAR 2007 evaluation on the detection of acoustic events which happen during seminars or meetings The implemented system consists in a front-end that converts an audio sequence in a stream of MFCC features and in a detecting/classifying block whose aim is to identify the acoustic events with time stamps and assign to them an event label among all possible event labels. Identification and classification are based on statistical models and in particular on Hidden Markov Models (HMM). The results, measured in terms of two metrics (accuracy and error rate) are obtained applying the implemented system on the interactive seminars collected under the CHIL project. Final not very good results highlight the task difficulty and complexity.