"On a time-frequency blurring operator with applications in data augmentation"Halvdansson, SimonInspired by the success of recent data augmentation methods for signals which act on a phase space representation of the signal, we introduce an operator which convolves the short-time Fourier transform of a signal with a specified kernel. Analytical properties including boundedness, compactness and positivity are investigated from the perspective of time-frequency analysis. The operator is also implemented numerically in several different versions and examples are presented. To explore the effectiveness of the operator for data augmentation, a simple convolutional neural network is trained to classify audio signals using spectrograms with different augmentations setups with results indicating that the operator significantly improves test performance, especially in the data-starved regime. |
| https://ps-mathematik.univie.ac.at/e/talks/strobl24_Halvdansson_2024-02_On_a_time_frequency_blurring_operator (19).pdf |
« back