By using machine learning technology for analyzing multimedia such as images and sounds, this research intends to be able to search songs more efficiently and accurately. In contrast with previous keyword-based methods, this method adopts images and sounds as a mediation. This method intended to reduce error due to verbalizing our feelings.