Investigation of Reverse Mode Loudspeaker Performance in Urban Sound Classification

The loudspeaker is a transducer that converts electrical signals to sound. However, it is well-known that in reverse mode, it can convert sound to an electrical signal. In this paper, the reverse mode behavior is investigated through the analysis of its influence on urban sound classification accuracy by comparing the results of deep learning based classifiers. As no audio datasets recorded by loudspeakers are available, a popular traditional dataset was used and transformed into forms as they would have been recorded by reverse mode speakers. These transformations simulated the loudspeakers' electrical responses to acoustical excitation signals based on their reverse mode transfer functions, which were derived from equivalent mechanical circuits. The details of this reverse mode modeling are also included. The transformed datasets were used during the trainings of the classifiers, and the effects of different speaker parameters and noise levels were examined and compared. The results showed that smaller, full-range speakers performed better than bigger woofers. The types of well-classified events revealed that loud, impulsive events could be classified more accurately.