Recognition of Musical Instruments from Audio Signals Using Deep Learning

Abstract

Currently, deep learning has been involved in different areas of science, being useful in tasks related to computer vision. However, there are other areas of technology that have benefited, among them we can mention digital signal processing. The processing of audio signals generated by musical instruments is of great relevance for the music market, but also for the preservation of the musical heritage. The proposal aims to identify the advantages of implementing deep learning in the analysis of audio signals, through the classification of musical instruments. Two neural networks were implemented that are trained using audio signals and audio feature images. The audio signals used are part of a public domain repository of the London Philharmonic Orchestra. The results obtained indicate that the proposed deep network neural result that differ by 0.01% with respect to those obtained with a convolutional network. It has been demonstrated Manuscript received on 18/05/2023, accepted for publication on 02/09/2023. E.M. Ramírez Rodríguez, M. Saldana Perez, J.L. Oropeza Rodríguez are with the Instituto Politécnico Nacional (IPN), Centro de Investigación en Computación (CIC), CDMX, México ({rramirezr2023, amagdasaldana, joropeza}@cic.ipn.mx). that the effectiveness of the proposal presented is close to that existing in similar works present in the state of the art. In which feature extraction is used as part of the results obtained, as opposed to only taking the input signal as in this work. I