Information overload, inability to quickly browse through audio, poor added-value to music via Internet distribution, keyword dictatorship, inability to search for similarities among sounds: these are music consumer complaints addressed by CUIDADO. It aims at developing content-based technologies using MPEG 7 output. Building reusable modules for audio feature extraction, statistical indexing, database management, networking and constraint based navigation, CUIDADO target two pilot applications:
1) The Music Browser features musical paths and automatic compilations according to user's tastes, search for music similarities, learning systems based on user's personal choices. It also serve as a monitoring system for Web music files.
2) The Sound Palette involves music creators for developing an authoring tool in an existing professional audio environment taking full advantage of the extracted audio features for innovative retrieval, editing and processing.
Develop content-based audio modules and applications using the MPEG 7 Media representation standard. The project covers the analysis process (extraction of descriptors), the navigation process (retrieval methods and interfaces implemented in a leading database system with Web interoperability), up to the creative process (consuming and authoring tools) involving content creators and consumers in each stage. The project addresses both the audio (samples) and the music domain (titles) since high-level descriptors for music (style, rhythm) should rely on robust lower level audio descriptors (pitch, energy or spectral features) in order to cover a wider range of applications. This approach answers the needs of record labels and copyright societies for Information management methods for both marketing and protecting their contents. CUIDADO is also a first attempt to go beyond content retrieval with an Authoring system using content features for professional musicians and studios.
The project involve music creators and consumers, online quality control and regular specification update according to MPEG7 inputs (WP1) throughout:- Two Pilot applications (WP3). The Music Browser include sound similarity search, constraint based music selection, personal keyword use and user's profiling. One version is tied to Web music monitoring and another to Web music sales and customized radios. The Sound Palette has an online version with retrieval and processing features scalable to the network performances and an offline version includes content-based editing/processing in an existing professional authoring environment. Reusable technical modules (WP2). Audio features extraction will provide content descriptors to the other modules, first derived from current Instrument Timbre Description proposed in MPEG 7 by the CUIDAD Group (Esprit 28793) and then from signal analysis and perceptual studies following MPEG 7 Multimedia Description Scheme. Statistical music indexing will derive high-level descriptors (such as music style or genre) from previous task audio descriptors and MIDI files. Database management system will be optimised for large audio & descriptors databases using Java beans and Mpeg7-DDL (or XML). Distant interoperability will use Corba Network Computer Architecture. Sound processing and networking modules for raw synthesis and fast listening over the network will be packaged with existing secured music transaction systems (IST project RAA and SDMI). Music sequence generation applies constraints on music titles/audio descriptors. - User validation, standardisation and dissemination (WP4). User groups for requirements, creativity and validation will gather music creators and consumers but also record labels, sound archives and copyright societies using a feedback loop methodology. Standardisation in MPEG 7 and dissemination to content providers is a key issue.
M1: Working methods and specifications (Month4) providing agreed online quality control and involving user groups
M2: Preliminary descriptor modules for application prototypes (Month 13)
M3: First Application Prototypes (Month18) for first user validation and specs revision with first version of the technical modules
M4: Final middleware technical modules (Month 24), documented and reusable in several applications
M5: Final feature extraction modules (Month31), documented and reusable in several applications
M6: Final Pilot Applications (Month33)
M7: Final validation and dissemination (Month35)
M5: Validation/Dissemination of Results (Month29) with strong contributions to MPEG 7
The project's actual outcome conform to the initial objectives in terms of - design of low- and high level audio/music descriptors adapted to targeted application features, together with low-level descriptor extraction and similarity computation software modules, with a systematically MPEG7 compliant approach, - design of a generic software architecture including database management system, access to audio/music material, user and user right management, client-server middleware, etc. - development of targeted applications : Music Browser, Online Sound Palette, Offline Sound Authoring Tool, Web Music Monitoring System, - scientific, professional and public dissemination of the project results through various media : publications, conferences, workshops, shows and exhibitions, etc.
Resulting innnovations related to audio/music content management features include: - automatic audio sample classification, - content-based sample editing - intelligent title play-list generation, - automatic audio summary generation, - identification of unknown audio excerpts from reference copyrighted databases (fingerprinting), - new navigation interfaces in sample and music title databases, - etc. These innovative features are integrated into the two resulting applications: 1) The Music Browser features musical paths and automatic compilations according to user's tastes, search for music similarities, learning systems based on user's personal choices. It also serves as a monitoring system for Web music files. This approach answers the needs of music listeners on one side, record labels and copyright societies for Information management methods for both marketing and protecting their contents on the other side. 2) The Sound Palette involves music creators for developing an authoring tool in an existing professional audio environment taking full advantage of the extracted audio features for innovative retrieval, editing and processing. CUIDADO is also a first attempt to go beyond content retrieval with an Authoring system using content features for professional musicians and studios.
Funding SchemeCSC - Cost-sharing contracts
84105 Beer Sheva
28230 Las Rozas
92110 Clichy La Garenne