AudioDB

What is AudioDB ?

AudioDB is a feature vector database management system that is optimized for passage-level retrieval from very large collections (1M+ tracks). It supports sequence-of-vectors searching on large databases. Developed by Michael Casey and Christophe Rhodes as part of Goldsmiths' OMRAS2 development.

Features:

• Exhaustively search for matching feature sequences across tracks
• Scales to searching over millions of tracks in milliseconds
• Locality-sensitive hashing (LSH) indexing and retrieval engine
• Automatic LSH parameter estimation for optimal performance
• Search high-dimensional vector sequences
• Efficient nearest-neighbour reporting of LSH results
• Command-line and Web services interfaces (over SOAP)
• Library and C API for developers
• Result format supports arbitrary primary key formats (e.g. URI, RDF)

Download latest development release from: http://rvw.doc.gold.ac.uk/sullivan/cgi-bin/gitweb.cgi?p=audioDB.git;a=summary.

Download latest development release of feature extraction utility fftExtract from: http://rvw.doc.gold.ac.uk/sullivan/cgi-bin/gitweb.cgi?p=fftExtract.git;a=summary.

Developers

AudioDBDeveloperDiscussionPage

• Michael Casey (Goldsmiths / Dartmouth)
• Christophe Rhodes (Goldsmiths)
• Ian Knopke (Goldsmiths)
• Malcolm Slaney (Yahoo)

• AudioDBBugReporting

Tutorials

• Tutorials 1, 2 and 2a

• Tutorial 3
• UsingKeysTutorial
• LargeADBTutorial
• IndexingTutorial

Formal Description of audioDB

• Document prepared by Mark d'Inverno and Christophe Rhodes: AudioDBSpec3.pdf

AudioDB in the PRESS

• Atelier report on mHashup demo at SIGGRAPH in LA, Aug 11-15 2008

-- MichaelCasey - 08 Oct 2007

• Set ALLOWTOPICCHANGE = AudioDBGroup