A2IR: Audio-to-Image Representation
Published: 28 September 2021| Version 2 | DOI: 10.17632/wdng2cjhmy.2
Contributors:
Steven Camacho, , , Description
A2IR is a dataset for synthetic audio detection using deep learning. It includes five audio-to-image representations for natural and synthetic audio: spectrograms, histograms, scatter plots, bispectrum phase plots and bispectrum magnitude plots. Each category is divided into 3 subsets: training 56.72% (11,400 images), validation 33.83% (6,800 images), and test 9.47% (1,900 images). In each subset, the images are separated into two folders, natural and synthetic, with a balanced classification (i.e. each class has the same number of images as the other or very similar).
Files
Institutions
Universidad Militar Nueva Granada
Categories
Computer Vision, Deep Learning