Analysis of LSTM, BiLSTM, and CNN Methods for Environmental Sound Identification in Smart Cities

Identification and analysis of environmental sounds in smart cities is important for city safety and comfort. In this study, sounds taken from eight different environments were used and a new sound dataset was collected. Long Short Term Memory (LSTM), Bidirectional Long Short Term Memory (BiLSTM), and Convolutional Neural Network (CNN) models were used for the automatic identification of environmental sounds. The collected sound data were segmented and their spectrograms were obtained. The obtained features were classified into LSTM, BiLSTM, and CNN. In the training results, $99.4 \%, 99.23 \%$, and $93.35 \%$ accuracy were calculated with LSTM, BiLSTM and CNN, respectively. In the test results, $97 \%$ accuracy was calculated for these three models.