Deep Single Shot Musical Instrument Identification using Scalograms
Debdutta Chatterjee, Arindam Dutta, Dibakar Sil, Aniruddha Chandra · 2023
Musical instrument identification has for long had a reputation of being one of the most ill-posed problems in the field of musical information retrieval. Despite several robust attempts made at solving the problem, a timeline spanning over the last five odd decades, the problem remains an open conundrum. In this work, we take on a further complex version of the traditional problem, we attempt to solve the problem with minimal data available - one audio excerpt per class. We propose to use a convolutional Siamese network and a residual variant of the same to identify musical instruments based on the corresponding scalograms of their audio excerpts. Results obtained for two publicly available datasets validate our algorithm, achieving over 80% accuracy with only 5 sets of training data. Moreover, our proposed architectures work for both spectrograms as well as scalograms, and exhibit improvements, albeit marginal (≃ 3%), for the later input class.