Music plagiarism has been a legally grey area for many decades. The complexity of music making contributes to the difficulty of defining plagiarism among songs. According to U.S. copyright law, in the absence of a confession, musicians who accuse others of stealing their work must prove "access"—the alleged plagiarizer must have heard the song—and "similarity"—the songs must share unique musical components. The focus of this project will be on the similarity between two songs, i.e. we will train a network as a similarity metric between two audio files. In the past, researchers have attempted to define music similarity using string-matching type of algorithm. But we hope to use recent developments in music information retrieve to tackle this problem again. We hope this work could serve as an insight into how machines discern the similarity of two songs.