MAMI: Musical Audio-Mining ("Query by Humming")

Project Description:

MAMI is a data-mining project for audio recognition that investigates ways of searching an audio archive as easily as you can search a text archive. 
The project starts from the observation that given the current state-of-the-art in telematica, the technological orientation of the music culture and the interest of the music industry to sell musical commodities and services via the Internet, there is a high need to develop advanced tools that support new ways to deal with content concerning musical audio and associated processing. Current technology makes it possible to retrieve music from a database using new content-based methods. Performing feature extraction on a wide range of sound characteristics opens the possibility for multiple ways of querying on data not only by text queries but also by music-based query techniques such as query-by-humming or query by specification of a list of musical variables.
A main characteristic of the MAMI-project is its focus on music as audio signal. This includes all kinds of music, including electro-acoustical music as well as ethnic and world music.

Project aims

• Develop a background epistomology for audio mining that is based on auditory modelling and perception theory.
• Work out methodologies, techniques and software tools for content-based musical audio mining taking into account all kinds of music.
• Development of an integrated system for audio description using different levels of representation.
• Work towards a practical application which demonstrates its usefulness by means of the so-called "query-by-humming" paradigm.
• Allow users to retrieve a musical piece by describing sound characteristics, either by humming or playing or describing the piece on the basis of its sound characteristics.
• Set up representational structures in compliance with the MPEG-7 standard, an interface for Multimedia Content Description.

Description levels

The MAMI-research project uses different description levels to describe music, such as:

• Waveform representation.
• Frame-based representations.
• Parameter-based representations.
• Event-based representations.
• Gesture-based representations.
• Concept-based representations.