Audio source device recognition is one of the most challenging task in digital forensic, but it is important and useful in other areas of digital forensic; for example, tamper detection, copyright protection etc. Recently a Multivariate Audio Database (MAuD, database download link [3] has been published where audio samples were recorded with one speaker’s device. Devices include laptops (HP 14s, Lenovo Ideapad, Lenovo Thinkpad) as well as mobiles (Mi4, Samsung A2, Samsung J2). MAuD dataset has many variations other than device as audio samples were collected in real life scenarios across different conferencing applications, languages, surrounding noises and conversation topics. Objective of this research is to verify if device specific information retains in audio samples while so many other variances are there. Two approaches have been used as no previous device recognition research has been performed on MAuD dataset. Pre-trained CNNs are used for recognition as well as feature extraction purpose. We obtained impressive recognition accuracy (98%) which clearly shows presence of significant device related properties even when so many other variances are present.