A novel approach for segment level audio retrieval using singular value decomposition

Audio files with different file names may contain same data in it. This duplicity of data occupies the system storage without providing any benefit to the user. It is obligatory to eliminate this duplicity, to optimize the system storage efficiently. To solve this problem, a segment level audio retrieval system is presented in this paper. The system is used to retrieve the duplicate audio files from the database using audio retrieval technique. The segments of the audio files are given as a query to the system. These segments are the audio pieces range between 1 to 2 seconds. Use of segments is preferred over original file because of their smile sizes, to reduce the system complexity. Using the system, duplicity of two files is decided by the similarity between their feature matrices. The system is purely application driven and does not depend on the particular audio characteristics. Experiments are done on the large audio databases and system is evaluated accordingly. The system is provided high retrieval accuracy for all the experiments. © 2014 IEEE.