MEMORIES - Design of an audio semantic indexation system allowing information retrieval for the access to archive content
MEMORIES will design an OAIS-based application combining computer assistance for developers of audio databases and a smart search engine for audiovisual information retrieval.
- Project type: STREP (Specific Targeted Research Project)
- Start date: 1 June 2006
- Duration: 36 months
- EU funding: € 1 225 000
- Number of partners: 7
- Project coordinator: MEMNON Audio Archiving Services, Belgium
- Contact: Mr. Michel Merten
Acquiring media content, structuring and attaching metadata and controls, archiving and exploiting it in various modes are complex processes. This is particularly true for sound signals, which are often composite and can basically be described with one or all of the descriptors: noise, speech, music.
Whilst available indexation systems are satisfactory for professional use as regards mono-component sound signals, their performance drastically decreases as soon as several components are present. High level information is difficult to obtain, for instance if the operator's task is to transcribe speech content of a recording in presence of noise or music. The MEMORIES project team aims at overcoming these problems through a generic software library that facilitates extraction of high level information from audio signals.
From the archivist point-of-view, the proposed application will present a flexible and effective tool for information indexation in audio resources and support the development of audio databases. It will provide operators with a maximum of computer assistance for attaching ontology and semantics to the content and with an innovative facility of source separation, combined with classical 'Speech to Text' and 'Wave to Midi' transcription functions.
From the user point-of-view, this application will appear as an accurate and robust tool for audiovisual information retrieval from large databases. Core facility of the search engine will be advanced search based on semantic associations. Its development will apply strategies generated for biology and genetics, using the textual annotation associated with and within the media.
Finally, the project will design an implementation, named AXIS, of the reference model presented in the ISO standard 'Open Archival Information Systems' (OAIS). This will guarantee that the system is open to the acquisition and to the exploitation of data from both old archives and new productions. It will also assure persistence (integrity and long term availability of the data), interoperability (capacity to exchange subsets of the databases), scalability (capacity to operate from small to large systems), and adaptability (capacity to be adapted to specific context-dependent needs).
The indexation technology developed by MEMORIES will help to turn huge amounts of archival audio data into valuable knowledge that eventually can lead to the creation of products that may finance digitisation of analogue audio material held by the institutions. Outlets for the smart search engine could emerge in fields such as education, broadcasting or telecommunications.