Context Recognition

The environmental sounds consist of a variety of everyday audio noises and events. Most of the work done in the area of sound is in voice and music recognition. And non- human sounds are still unexplored. The need to solve the problem of environmental sound classification, arises from the potential applications of using environmental sounds as a tool and from the fact that until now most of the research has been done on human sounds only and non-human sounds are still unexplored. The representations are often transformed further into more compact forms of audio features.The performance for a robust task, which is recognising sounds in a noisy environment, it was observed that the best performance was given by input features that are in fact images. The approach used in this project currently proposes the idea to use convolutional neural networks (CNN) with the spectrogram image features.

Understanding or recognizing context of sounds in environmental surroundings is very important in terms of making the next move based on a sound that occurred. Examples include evacuating a building on hearing a fire alarm or attending to a baby on hearing the cries. Such activities depend on human hearing which intelligently filters out sounds and quickly recognizes it thereby signaling the brain to take the next step. According to the World Health Organization, over 360 million individuals worldwide suffer from some form of disabling hearing loss. Deaf individuals can neither hear sounds nor distinguish between many sounds. Such situations could be life threatening at times and also inconvenient. Providing access to sound in some other form of communication are in demand for its practical applications towards assisting deaf individuals in their daily activities. This model has been made by keeping such utilities of context recognition in mind.