Ανάλυση ηχητικών εγγραφών αστικού περιβάλλοντος και αναγνώριση πλαισίου
KeywordsMachine learning ; Classification ; Audio analysis ; Audio classification ; Urban sounds ; Sound recognition
At this thesis we tried to apply and evaluate some classification techniques through machine learning algorithms aiming at the automatic recognition of an unknown set of urban sounds. So far, the majority of the researchers have investigate the recognition of the human speech or the categorization of different music types. In this study, we tried to examine how some of the most popular classification algorithms correspond to an urban audio library. The first step of the process was the creation of an audio library that would include different activities for each of the eight classes studied. Then followed the manipulation of the audio data, such as splitting into the desired lengths of time and further subdividing after applying the frame functions. After labeling every sound data, we convert the audio signal to small dimension vectors of audio features in order to be able to be classified. The next step was to apply the machine learning algorithms to the processed audio data. The audio attributes were divided into a training and a test set for the estimation of the audio recognition performance. Finally, the machine learning algorithm with the highest accuracy was applied on an unknown set of audio data of varying lengths of time. By this step we tried to investigate its generalization ability.