Similar Singer Search by Property of Vocal Sound through Weighted Euclidean Distance

Although using the singer's name to search for songs is the common function of music retrieval systems, people often only remember famous singers. For new singers, people often only remember that the singer’s voice is similar to a famous singer, but cannot recall his/her name. Among current studies, there is little discussion on the music search method by voice. Therefore, this study proposed a method of music search by similar vocal sound. Users can enter the name of a singer, and then the system can find the names of other singers with similar features in vocal sound. In the proposed method, the sound signal is processed by Discrete Fourier Transform, and the statistical value of frequency spectrum and formant is calculated as the characteristic value of the singer's vocal sound. Then, the similarity of the singer's vocal sound is calculated by the weighted squared Euclidean distance function to obtain similar singers' names, so as to search for music by using similar vocal sound. The experiment verified that the feature extraction method can realize the vocal sound-based search effectively.