Musical instrument recognition with wavelet envelopes

Hacıhabiboğlu, Hüseyin
Canagarajah, Nishan
Automatic recognition of instrument type from raw audio data containing monophonic music is a fundamental problem for audio content analysis. There are many methods for the solution of this problem, which use common spectro-temporal properties like cepstral coefficients or spectral envelopes. A new method for instrument recognition utilising short-time amplitude envelopes of wavelet coefficients as feature vectors is presented. The classification engine is a distinctively small multilayer perceptron (MLP) network. A correct classification rate which is comparable to previously reported correct classification rates is attained for a set of three instruments containing flute, clarinet and trumpet.
Citation Formats
H. Hacıhabiboğlu and N. Canagarajah, “Musical instrument recognition with wavelet envelopes,” presented at the Forum Acusticum (16 - 20 Eylül 2002), Sevilla, İspanya, 2002, Accessed: 00, 2021. [Online]. Available: