İlker Türker, Serkan Aksu
Applied Acoustics
Volume 191, 30 March 2022, 108660
We propose a novel time-graph representation method for sounds and time series.
Quantized amplitude levels serve as nodes, while neighborhood of these nodes in time domain define connections.
Applying various undersampling rates, a multilayer graph representation is achieved, further composed as RGB images.
Flattened graph representations from each time frame are tiled to compose a time-graph representation.
The proposed representation improves classification accuracy when used in combination with mel-spectrograms.
Read the full article at: www.sciencedirect.com
Complexity Digest 