Bewdy, Maaate: Audio analysis in the MPEG compressed

Thomas Vincent
CSIRO Mathematical and Information Sciences

Tuesday 29 January at 11am
*** Please note the special date of this seminar***

Abstract

With increasing amounts of digital audio being available on CD, the Web, digital radio broadcasts, and as part of digital Television, the user asks for more capabilities to handle sound information aside from just the file names and possibly title and artist. We have to consider the specificity of sound itself as a communication channel. For example, an audio web search  engine should not only be based on the file names but also on the audio content. 

Most digital audio files are available in compressed format because of reduced storage requirements. Audio analysis directly in the compressed domain thus has the advantage to be costless as it avoids part of the decompression steps. 

Maaate is an audio analyis toolkit developed by CSIRO, which allows extraction of relevant features from compressed sound files (MPEG-1 audio files). Bewdy is a visualization interface for the feature calculation results of Maaate, also developed by CSIRO. 

The talk will start with a brief introduction of MPEG-audio encoding and the information available directly from the compressed domain. The Maaate architecture is then explained and the kinds of features that can actually be extracted with Maaate. The features range from signal analysis features such as bandwith or energy density to perceptual features such as pitch, loudness or brightness. Features with more semantic content are silence-based segmentation, music or speech likeliness. A live demonstration of feature results will be given using Bewdy and it will be explained how the features can support the applications mentioned above.

Short resume

Thomas VINCENT is a final year student at the Departement of Telecommunication at INP Grenoble, France. His study contains a general background in math and physics and various fields in computer sciences and electronic. During his current internship at CSIRO he worked on audio analysis in the compressed domain implementing a large set of new features for Maaate and improving the visualisation tool Bewdy.

Back to HAIL Home Page