Similarity of Musical Timbres Using FFT-Acoustic Descriptor Analysis and Machine Learning

Musical timbre is a phenomenon of auditory perception that allows the recognition musical sounds. The challenging task because instrument or sound source complex and multifaceted influenced by variety factors, including physical properties source, way it played produced, recording processing techniques used. In this paper, we explore an abstract space with 7 dimensions formed fundamental frequency FFT-Acoustic Descriptors in 240 monophonic sounds from Tinysol Good-Sounds databases, corresponding to fourth octave transverse flute clarinet. This approach us unequivocally define collection points and, therefore, timbral (Category Theory) different any type its respective dynamics be represented as single characteristic vector. geometric distance would allow studying similarity between audios instruments datasets. Additionally, Machine-Learning algorithm evaluates similarities through Euclidean distances was proposed. We conclude study allowed distinguish audio categories instruments, same relative dynamics,