Deep Learning applied to recovery music information.

This seminar covered the main concepts and advances in the area of ​​Music Information Retrieval (MIR), starting with the conversion of audio into discrete digital representations, from the sound wave to more interesting techniques such as the Short Time Fourier Transform (STFT), revealing frequency information over time, and the Chromagram and Melspectrogram, which capture tonal and timbre characteristics of the audio. We discuss fundamental tasks in MIR, such as identifying musical genre, retrieving similar songs and estimating chord progressions, essential for the harmonic analysis of a composition. From this, we explore the applications of Deep Learning models, such as Recurrent Neural Networks (RNNs) and Transformers, in MIR.