Music is widely used for various purposes, it contributes to the treatment of patients since people appreciate it and use it for music therapy. Music therapy is a clinical practice that engages both client and therapist in dynamic musical interaction. In music pitch estimation experiments, the database is adopting western classical music and the original songs. We proposed a multi-resolution feature extraction technique to deal with music therapy signals. We utilize adaptive resonance theory to search their clusters of features. The effectiveness of the clusters is encouraging.