Year: 2024
Type: article
Abstract: An in-depth study of audio separation, delving into avant-garde and conventional methodologies for isolating musical tones. The exploration aims to investigate various techniques for isolating musical tones and extracting individual components of sound. By comparing traditional and advanced approaches, the study seeks to offer insights beneficial to researchers, educators, musicians, and composers. This paper briefly investigates conventional approaches like Non-Negative Matrix Factorization (NMF), Independent Deeply Learned Matrix Analysis (IDLMA), Independent Low Rank Matrix Analysis (ILRMA), Independent Component Analysis (ICA), and Principal Component Analysis (PCA), detailing their working principles and advantages. Also, Analysis of the numerous forms of machine learning techniques. Afterwards, it explores how the models are used to dissect the instrumental acoustics when deep learning techniques like Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) are applied. Additionally, coupled deep learning frameworks that include High-Resolution Long Short-Term Memory (HR-LSTM), Dense-U-Net, Wave-U-Net, Conv-tasnet, Res-U-Net and Long-term Recurrent Convolutional Network (LRCN) are analyzed. DenseLSTM and Audio Spectrogram Transformer are evaluated because the combined architecture is more efficient than the individual architecture. This Paper bridges avant-garde and conventional audio separation methodologies, offering valuable insights for various stakeholders and indicating a path towards enhanced practical applications in the field of audio separation.
Authors: Sakthidevi S P, Divya C, Kowsalya V
Institution: Manonmaniam Sundaranar University
Language: English
Topic: Music and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4404006217