In this paper we present a new method for robust audio identification. Based on our existing audio indexing technology, we developed new methods to query large audio data bases with highly distorted versions of an audio signal or parts of them. For instance the data base could be queried by transmitting a piece of music using a cellular phone. In contrast to recent approaches, arbitrary segments of a piece of music are allowed as a query. We demonstrate that our method for any short audio fragment with length exceeding approximately five seconds, is able to identify the corresponding piece of audio along with the exact position of the fragment within the original signal. Our approach only relies on features extracted from the audio signals hence making the embedding of, e.g. watermarks obsolete. In our work we furthermore give an overview on our extensive tests using a database of several 1000 items of audio (approximately one month of audio) demonstrating the capability of our new method.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=11325
(1044KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
