Babaz

Babaz is a audio database management system with an audio-based search function.
It is licenced under the terms of the GNU General Public License v3.0.

Brief functional overview

1. Audio descriptors are computed from the audio files using the filter banks from SPro.
2. The database descriptors are index using KM-LSH (by default) or PQ-codes (for licensed users).
3. The approximate nearest neighbors of the query descriptors are computed.
4. Optionally, the reciprocal distance is used to improve the score (as described here).
5. Hough matching is performed on the matches to find the matching audio tracks and time offset.
6. If the groud truth is known, various scoring measures are computed.

A complete system description is given in the ICASSP 2012 paper (see below)

Latest release:

• babaz v1.0.114

Dependencies:

• Yael
• SPro

Related publications:

• Hervé Jégou, Jonathan Delhumeau, Jiangbo Yuan, Guillaume Gravier, Patrick Gros. Babaz: a large scale audio search system for video copy detection. In IEEE International Conference on Audio, Speech, and Signal Processing, ICASSP'12, Kyoto, Japon, January 2012.
• Mohamed Ayari, Jonathan Delhumeau, Matthijs Douze, Hervé Jégou, Danila Potapov, Jérôme Revaud, Cordelia Schmid, Jiangbo Yuan. INRIA@TRECVID'2011: Copy Detection & Multimedia Event Detection. In TRECVID 2011 Workshop, USA, 2011.

This software was developped in the TEXMEX team by Hervé Jegou, Jonathan Delhumeau and Guillaume Gravier.
Software development was partially funded by the EIT ICT Labs OpenSEM project and by the OSEO Quaero project.