Topological learning for acoustic signal identification

We consider the problem of classifying highly concentrated acoustic signals. We implement a topological learning technique to examine the shape and topology of signals' higher dimensional embedding. Next, rather than considering classical features for acoustic signal classification such as Cepstral features, instead a new feature extraction is presented which reveals information about the underlying structure of the signals. The performance of this new methodology is then compared with the Cepstral features in real data provided by the Army Research Lab.