Abstract: Content-based cover song detection is a hot research program in the field of music information retrieval in recent years．With the consideration that the different versions of music may change in many aspects,such as tempo,speed,pitch,timbre and structure,the key factor in this research is to extract robust audio feature,which can represent songs’ main melody progress,and compare their similarities between cover songs．Chroma features express the distribution of audio energy among different pitch class,consider presence of harmonics ,timbre independence and robustness to noise,so are used by a lot of cover song detection systems．Based on the theory,we designed and conducted the experiment to explore the effect of different forms of audio interference,and verified Chroma’s robustness in the aspect of pitch invariance.

Key words: Chroma feature,Robustness,Pitch invariance,Cover song detection