Fig. 1
From: Classification of audio signals using spectrogram surfaces and extrinsic distortion measures
High-level schematic overview of the proposed model. a Highlights the ‘embedding’ process. Subsequent to the preprocessing, the spectrogram of the 1D signal is computed and presented as a surface. This allows the conversion of the original problem of 1D signal analysis to geometric analysis of manifolds. The second part of the proposed method begins with this ‘geometric object,’ which is the output of the first part. b Summarizes how this geometric object is used for the extraction of features. First, the spectrogram is converted to a mesh, by means of the Delaunay algorithm. Then, the meshes are projected into a target domain by using the ABCD algorithm. The meshes are clustered in this target domain, with reference to a well-defined metric. Finally, distortions measures are computed from the obtained meshes