DICTIONARY LEARNING BASED TARGET SOURCE SEPARATION USING INDUCTION ALGORITHM

摘要

The present invention relates to a learning technique applied method for separating a sound source based on a dictionary learning algorithm. More specifically, the method includes: (1) a step of learning the dictionary learning algorithm with regard to the sound source; (2) a step of creating a mixed sound source by using the sound source and learning the relation between the mixed sound source and target sound sources; and (3) a step of separating an input sound source into the target sound sources by using the relation learned in step (2). According to the learning technique applied method for separating the sound source based on the dictionary learning algorithm the present invention suggests, it is possible to effectively separate the sound source even when the target sound sources have a similar characteristic and have excellent sound source separating performance by sufficiently incorporating various noise characteristics, by using the dictionary learning algorithm and separating the input sound source into the target sound sources by learning the relation between the mixed sound source and the target sound source and using the learned relation.