Abstract of Thesis presented at COPPE/UFRJ as a partial fulfillment of the requirements for the degree of Master of Science (M.Sc.)

Musical Notes Identification in Solo Recordings of Acoustic Guitar and Piano

Alexandre Leizor Szczupak

June/2008

Advisors:

Luiz Wagner Pereira Biscainho Luiz Pereira Calôba

Department:

Eletrical Engineering

      This dissertation presents methods developed for the identification of musical notes in acoustic guitar recordings. These methods are based on multilayer feed-forward neural networks, trained with frequency domain representations obtained via a constant-Q transform. Versions of these methods, developed for the identification of musical notes in piano recordings, are also presented.
      The proposed methods can be divided in two categories: methods based on a single neural network, used to identify the notes in a signal excerpt; and methods with two neural networks used in sequence, the first one to identify the bottom note of a signal excerpt and the second to determine the intervals between the bottom note and the remaining ones.
      Encouraging results were obtained on the identification of musical notes in acoustic guitar recordings, but not on the identification of musical notes in piano recordings. For both instruments, the best results were obtained using methods of the second category, especially regarding the isolated performance of the neural network used to determine intervals between the bottom note and the remaining ones.