Environmental alarm sound (EAS) is a rich source of information that can be used to infer awareness of the surrounding, especially for events out of sight and attention. The events may include fire alarm, car horn, dog barking and others. Most of human beings can recognize these alarm sounds and can take a prompt action. However, hearing-impaired (HI) individuals get difficulty to detect these sound events. This thesis addresses the design and development of a model for detection and recognition of environmental alarm sound (EAS) on mobile phones for the awareness of hearing impaired. We compare performance of different Artificial Neural Network (ANN) sound classifiers (Convolutional Neural Network (CNN) and Multilayer Perceptron (MLP)) and sound feature extraction techniques (Mel Frequency Cepstral Coefficient (MFCC), Mel Spectrogram (Mel), Spectral contrast (SC) and Chroma gram (Chroma)) to select the best ones in the case of MLP classifier. The performance of classifiers evaluated on a set of five EAS classes (dog barking, car horn, gun shot, engine idling and fire alarm). The offline training and testing result show that CNN classifier with log-mel-spectrogram feature outperforms than MLP classifier, in terms of recognition accuracy. The audio feature vector is selected based on its suitability for CNN classifier for offline training and testing and verified to be optimal features. Android-based EAS recognition model is then implemented using CNN classifier on the selected feature. The recognition performance of the EAS recognition technique is evaluated under 20% of the EAS dataset for 10 times trial. In this case, EAS recognition technique achieve 89% recognition accuracy, which outperforms the recognition performance of previously investigated efforts on the area of environmental sound recognition (ESR).