Our brain can recognize a wide variety of noises that are present in our environment with ease. In addition, our brain continually interprets the sound signals it receives and gives us pertinent environmental information. Deep learning categorizes audio signals like speech, music, and background noise using large datasets for high accuracy. However, appropriate audio signal representation is crucial for effective classification. Techniques for representing signals, such as spectrograms, Mel-frequency Cepstral coefficients, linear predictive coding, and wavelet decomposition, can be used aimed at this. The audio signal can be used in a DL model after proper encoding, with various algorithms compared for their ability to handle complex datasets and make accurate predictions. The carefully chosen dataset utilized in this study is the UrbanSoud8k Dataset, which consists of 8732 labelled sound samples (<= 4s) of urban sounds from 10 classes: air conditioner, car horn, kids playing, dog barking, drilling, engine running, gunshot, jackhammer, siren, and street music. Several models were applied, including the CNN, CNN-LSTM, and CNN-Transfer Model. Results show that the CNN-YAMnet model attained an inspiring accurateness of 98%.