Título: Discriminação Neural de Instrumentos Musicais Baseada no Espectro
Autores: Santos, Cristiano Nogueira dos; Calôba, Luiz Pereira; Biscainho, Luiz Wagner Pereira
Resumo: Some works have recently approached the problem of musical instrument recognition. This paper presents a first step towards a pitch independent recognition system of which the main target is to employ a reduced set of input parameters. The problem was constrained to instrument identification from emitted individual notes in steady state. A neural network receives as inputs the relative spectral magnitudes of the harmonic components of each sound sample, extracted by DFT. Training and testing data were obtained by recording synthesized (midi) notes for eight instruments from the reed family. Validation tests used audio samples from real recordings with success.
Código DOI: 10.21528/CBRN2001-074
Artigo em pdf: 5cbrn_074.pdf
Arquivo BibTex: 5cbrn_074.bib