Abstract

In this work, we present a system for detecting a specific musical instrument note in polyphonic mixtures based on machine learning method. By using extreme learning machine as a classifier, we generate input features from magnitude data of FFT spectrum, which are processed by an adjustable dynamic level processor. We use the particle swarm optimization to tune the parameter values of this level processor to make the system have the optimal detection performance for the specific instrument signal. Three musical instruments were used in the experiments. These were trumpet, flute and clarinet. We also compare the detection results between using ELM with linear and non-linear output functions for selected features. Furthermore, we show that our system can be used as a musical instrument source tracking system by creating a trajectory of its fundamental frequency using the information from the outputs of the system and illustrating over the spectrogram.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call