The objective is to develop Machine Learning models to detect events using audio files. The list of events is (10 events): dog bark, engine idling, siren, jackhammer, drilling, children playing, gun shit, car horn, air conditioner and street music. The basic idea is to extract features from audio data using Fourier Transform. For more details, see report.pdf. We develop models for two tasks:
The goal is to classify an audio file into one of the possible events. It is a multiclassification problem. The task-1 folder has 3 notebooks corresponding to data preprocessing, training of model and testing of model on unseen data.
The goal is to determine the sequence of audio events in a single .wav file. The sequence of events our model outputs is critical to the accuracy of the model. The evaluation metric is edit distance. The task-2 folder has 3 notebooks corresponding to generating data from single event audio files, training of model and testing of model on unseen data.
Note: The link to the train data ('.wav' files) is: https://drive.google.com/drive/folders/1SLk9xU8bGSb98FGqlb5Q19oqTejqKy-f?usp=sharing