As the amount of music available via the Web increases dramatically, both music lovers and audiovisual professionals start to feel lost. IRCAM's CUIDADO² project aims to create the audio data equivalent of search engines. The first one, the Sound Palette, is for elementary sounds and the second, the Music Browser, is for musical works.

In the hushed atmosphere of IRCAM's laboratory/studio, Hugues Vinet, the Director of STMS and Coordinator of the CUIDADO project, is listening over and over to recorded sounds, while manipulating and analysing them. “We're trying to perfect procedures for automatic description of audio data based on sound content”, he explains.

From musical color to harmonic structure
CUIDADO researchers are exploring two avenues simultaneously. In a bottom-up approach, the audio signal is the starting point for defining attributes like timbre, musical color, or how a note is attacked. The top-down approach on the other hand focuses on the perceptions of the listener. Experimental studies have shown that differences in timbre can be modelled through three perceptual dimensions. The sound pitch or the perceived energy can also be quantified. These so-called high-level parameters are difficult to interpret throughout an entire piece of music, and other descriptors must be devised.
CUIDADO has experimented with, for example, defining cultural affinity between two artists by statistical analysis of references to both by the same radio programme or the same web pages. Nevertheless, as Hugues Vinet emphasises, “structural analysis of music remains our principal focus.” Variations in acoustical parameters can be used to identify rhythmic motifs, melodies, and harmonic patterns, and from those to decipher the structure of a piece.

A way to thwart pirating
Analysis of music is a tool with a number of possible uses. A summary of a work of music can be produced by placing end to end the first few measures of each movement of the piece. The musical signature of a piece produced with a similar method would represent 1,700 times less data than the piece itself, “thereby making it possible to identify a piece in a matter of seconds”, Vinet points out. “This is also a powerful weapon against music pirating.” Automated publishing of musical scores, assisted musical composition, or organising music storage by style are some of the possible applications.
These description procedures are grouped into two computer programs, the Sound Palette and the Music Browser. The Sound Palette is designed principally with sound-effect technicians in mind, but it also can be useful to composers who will be able to search music for certain timbres or styles of attack. With the Music Browser those who catalogue or program music will be able to search for similar rhythms and harmonies, or cultural affinities, putting together playlists with a few clicks.
The CUIDADO project, which began in 2001, is involved in the development of the international MPEG7 standard for describing audiovisual content. But, as Vinet points out enthusiastically, “beyond the strictly scientific goal of analytic descriptions of sounds, our aim is to bring new value to music on-line.” The legal aspects of music distribution are, of course, another question.