Transcrição automática de acordes: análise espectral e classificação

Abstract

Chord transcription is a technique with a high degree of difficulty, even for some experienced musicians. That said, this work aims to present the method for automatic transcription of chords, evaluating different spectral analysis techniques for identifying the frequency components of the signal and through the use of dynamic programming algorithms to classify chords. As a result, it was found that the constant Q transform and the Viterbi algorithm proved to be efficient for identifying less complex compositions, composed of major and minor chords, bringing significant results for more complex songs that feature dissonant chords. Finally, the model used, even without presenting a 100% hit rate for songs composed of dissonant chords, proved to be a formidable tool to assist in the transcription of chords, both for simple and more complex songs.