Discovering Key Phrase in Acoustic Audio Data

We present a method to automatically discover key phrase from audio recordings in this paper. Chroma features are firstly selected to represent melodic information and a modified similarity matrix based on this feature is then established for repeated pattern searching. We also run a pruning process to discard invalid phrase pairs and a novel boundary correction algorithm is put to make the phrases more precisely. Final key phrase is determined by both its occurrence frequency and energy information. Key phrases of nearly 200 pop songs are extracted using our proposed methodand 5 person´s evaluations to the experiment results show the effectiveness of our method.