Audio phylogenetic analysis using geometric transforms

Abstract

Whenever a multimedia content is shared on the Internet, a mutation process is being operated by multiple users that download, alter and repost a modified version of the original data leading to the diffusion of multiple near-duplicate copies. This effect is also experienced by audio data (e.g., in audio sharing platforms) and requires the design of accurate phylogenetic analysis strategies that permit uncovering the processing history of each copy and identify the original one. This paper proposes a new phylogenetic reconstruction strategy that converts the analyzed audio tracks into spectrogram images and compare them using alignment strategies borrowed from computer vision. With respect to strategies currently-available in literature, the proposed solution proves to be more accurate, does not require any a-priori knowledge about the operated transformations, and requires a significantly-lower amount of computational time. This work has been partially supported by the University of Padova project Phylo4n6 prot. BIRD165882/16. This material is based on research sponsored by DARPA and Air Force Research Laboratory (AFRL) under agreement number FA8750-16-2-0173. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright notation thereon. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of DARPA and Air Force Research Laboratory (AFRL) or the U.S. Government.

Phylogenetic Minimum Spanning Tree Reconstruction Using Autoencoders

“…), finding the matching points, computing the homography matrix H k,t and warping I k onto I t after an equalization of color histogram values. This approach proves to be more flexible and less computational demanding, and therefore, it has also been adopted for spectrograms of audio tracks in [11,12]. As for video contents, three-dimensional descriptors permit to match and align chunks of video sequences whenever simple cuts and interpolations were applied [13,3].…”

Stay True to the Sound of History: Philology, Phylogenetics and Information Engineering in Musicology

“…A more flexible and computationally-efficient approach was proposed in [17]; in this latter approach, audio tracks are time-and frequency-aligned by representing each audio file with a spectrogram and using image registration techniques investigated in the field of computer vision. Moreover, the set of possible transformations includes time-frequency operations, as well.…”

“…As outlined in the Introduction, this paper presents an innovative approach to the musicological analysis using phylogenetic techniques, typically adopted in forensics, in order to reconstruct the stemma codicum, which can be considered as an audio phylogenetic tree (APT) [17].…”

“…This introduces a variable to the phylogenetic problem described in [17]: the digitization. An analogue tape can be read and digitized several times, creating similar digital versions of a unique document.…”

“…Although, in the last few years, several works have been targeting the analysis of images and video contents, little effort has been put toward audio phylogenetic approaches. To the best of our knowledge, the only algorithms proposed in the literature are represented by [12,17]. Moreover, the phylogenetic analysis problems have been extensively investigated for digital multimedia contents since online material offers the largest datasets of near-duplicate contents.…”