In the vocal communication, we are able to detect the position of a source sound in 3D space, extract particular sound from mixed sounds, and recognize what the sound is. An ability that extracts and recognizes particular sound in noisy environment is called cocktail party effect. By realizing this mechanism with a computer, new applications will be presented to be utilized in the communication with humans. In this study, musical sounds are dealt for the tracking of sounds generated by musical instruments. Musical sounds have different features from human voice. For example, instrumental sounds have variety of spectra distributed in wide frequency range. In this study, a system that can deal with various kinds of sounds will be developed. For the estimation of musical pitch, this study focuses on musical sounds with different pitches generated from musical instruments, and a filtering technique using comb filters is being developed. A recorder, harmonica and keyboard-harmonica were selected as target musical instruments in the experiment and the result showed this technique was effective for the estimation of musical pitches.