Frequency-domain Features for Environmental Accident Warning Recognition

In real life, the analysis and classification of sound signals is a significant part of environmental recognition systems. This work aims to present the performance of different combinations of frequency-domain features in the process of building a recognition system of environmental accident warnings. In this paper, a context-aware accidental signal recognition system is proposed for detecting sound events in three different scenarios. This approach is based on Multi-layer Perceptron Neural Network classifier using different combinations of frequency-domain features. The experimental performance evaluation on the different combination of frequency-domain features is significant and effective. Typically, we obtained recognition as high as 84.83% using the combined Mel Frequency Cepstral Coefficient, Mel Spectrogram and Spectral Contrast technique.