Deep Metric Learning for Music Information Retrieval

Abstract

This master thesis explores the application of Deep Metric Learning (DML) for creating effective audio representations in tasks like audio classification, music retrieval, and speech recognition. DML uses deep neural networks to learn hierarchical representations from raw audio waveforms, capturing intricate relationships between audio samples. The thesis evaluates different deep neural network architectures and loss functions, including triplet loss and contrastive loss. The models are tested using various distance metrics and normalization techniques. The research aims to enhance our understanding of DML for audio representations and its potential applications. The findings contribute valuable insights to guide the design of powerful audio representations for diverse audio-related tasks.