In this work, a query-by-singing (QBS) content-based music retrieval (CBMR) system is proposed. The proposed QBS-CBMR system shows high efficiency and portability. The proposed QBS-CBMR system uses a music clip as a search key. First, a 13 dimensional Mel-frequency cepstral coefficients (MFCCs) is extracted from an input music clip. Second, each dimension of MFCCs is transformed into a symbolic sequence using the adapted symbolic aggregate approximation (adapted SAX). Each symbolic sequence corresponding to each dimension of MFCCs is then converted into a tree structure called advanced fast pattern index (AFPI) tree. In order to evaluate the similarity between the query music clip and the songs in the database, a partial score is calculated for each AFPI tree first. The final score is obtained by the weighted summation of all partial scores, where the weighting of each partial score is determined by its entropy. The experimental results show that the proposed music retrieval system outperforms other approaches.