NOTE RECOGNITION FOR THAI XYLOPHONES
Chanwit Janraksukhum, Montri Karnjanadecha · 2007
This paper describes the design and implementation of note recognition for Thai xylophones. Thai xylophone is a percussion musical instrument which can be played monophonically and polyphonically. A set of 21 digital bandpass filters are used to determines energy. An energy compensation technique is applied to normalize the loundness of the energy of each note. This paper uses an artificial neuron network to determine if the sound is monophonic (one note) or polyphonic (two notes). The output from the neuron networks specifies the number of highest energy notes (1 or 2) that must be selected for actual notes from the sound. The recognition accuracy obtained is 98.24%. The duration of each note is determined from the sound waveform. The output notes and durations are presented in a MIDI file. The algorithms are implemented in Matlab - as an application program. The user can record sounds, play sounds, and convert the recorded sounds to notes. The program also allows the user to set the threshold value and design a new set of filters that suitable for other xylophones.