Abstract: “Keywords and descriptors are important metadata for multimedia content. Such metadata are associated with the programme, and are generated to facilitate indexing, search, clustering, archiving, semantic analysis and many other potential applications. Information about performing or recording venues provides a useful cue for content search and authentication, but is difficult to obtain form the media content. This paper proposes to determine the recording venues from extracted room acoustic features contained in the soundtracks. Room acoustic decay curves are obtained via maximum likelihood estimation from recording. The decay curve is a statistical description of the impulse response of a room, and provides a good discriminator for recording venues. Machine learning is then performed on the estimated decay curves to make the decision. This paper presents the rationale of the method, describes the algorithms and validates the method by simulations.”

Keywords: acoustic feature, recording venue, maximum liklihood estimation, machine learning, media content