A silence/noise/music/speech splitting algorithm

In this paper, we present techniques to warp audio data of a video movie on its movie script. In order to improve this script warping, a new algorithm has been developed to split audio data into silence, noise, music and speech segments without training step. This segments splitting uses multiple techniques such as voiced/unvoiced segmentation, pitch detection, pitch tracking, speaker and speech recognition techniques. The 102.47 minutes of the film movie « Contes de Printemps » produced by E. Rohmer have been indexed with these techniques with an average shifting lower than one second between the time-code script and audio data.