MEMORIES : Design of an audio semantic indexation system allowing information retrieval for the access to archive content

“Sound is a very complex signal. It is often composite and is basically described with one or all of those descriptors: noise, speech, music. Performance of indexation systems is satisfactory for professional use with mono-component sound signals. They drastically decrease as soon as several components are present. High level information is in this case very difficult to obtain. It is for instance very difficult to transcribe speech content of a recording in presence of noise or music. Sound indexation with noise/speech/music is an automated processing that deals only with representing perceived content, it does not take into account high level information that match user needs.  

The proposed project aims at elaborate a generic software library to facilitate extraction of high level information from audio signal. This library will act as a front-end processing for all kind of information retrieval from audio files. It will also propose an information retrieval system that matches archivist needs.

The main expected innovations of the MEMORIES project are:
  • a usable, user-friendly system that matches archivist needs for information retrieval in audio databases
  • the definition of a formalism for database structuration an information contents descriptors
  • a generic front-end system for all kind of information retrieval tasks in audio data,
  • an efficient tool for database structuration that take advantage of the indexation tool derived from single sensor source separation,
  • an efficient tool for audio restoration.
The partnership is highly complimentary since it gathers SME, well-known research centers, end-users and a famus cultural organisation from 5 different countries.