The organization and identification of sound files in sample libraries is a topic of concern in the fields of Music Information Retrieval and Timbre Classification. Of interest is the notion of labelling and searching for files using sonic and timbral criteria.

The author proposes classifying percussive sounds using the ‘Percussive Audio Database’ (PAD). PAD is a multi-dimensional taxonomy for drum and percussion libraries, employing an Object-Relational model. It describes a given sound by a set of 34 ‘classes’; each defines a specific physical, spectral, temporal or perceptual/timbral attribute. Examples include Stroke Type, Spectral Density, Harmonic Content, Attack Time, Decay Rate, Brightness and Attack ‘Hardness’. Classes are labelled with an alpha-graphical notation and displayed as a three dimensional matrix of colour-coded objects, using hierarchical layers and distance measures to connect them*.

A hybrid analyser/database is currently being designed that automatically analyses samples for spectro-temporal data, and permits user-input of various physical and perceptual attributes via menus and listening surveys. Samples are then automatically classified across all 34 attributes. End users may then identify, search for, filter and compare collections of samples in a library using any or all attributes. Artificial Intelligence models are also explored, with a view to allowing PAD to ‘learn’ patterns of similarity and difference between different samples in a collection; for example, PAD learns to assume that different beater types can create different levels of ‘brightness’ in a given instrument’s sound, and automatically classifies relevant samples accordingly.

PAD is being developed as part of the author’s PhD candidature at Swinburne University.

*Please see attached jpg file if an example image is desired by the panelists

Authors: Robert Bell, Dr. Anthony Bartel, Assoc. Prof. Stephen Barrass

Event: SF08: Search and Information Extraction from Audio Data Workshop

← View all submissions for this event.