AUTOMATIC EXTRACTION OF MUSICAL PORTIONS OF AN AUDIO STREAM

摘要

Music and non-music portions in an audio stream are identified. The audio stream is digitized and segmented into frames. Selected frames are passed through a filter bank which includes filters having bandwidths approximately proportional to their center frequencies. The spectral flux for each selected frame is calculated and smoothed. Frames having a smoothed spectral flux below a threshold value are associated with music, and frames having a smoothed spectral flux above a threshold value are associated with non-music.