Sound indexing in large audio archives

Download the project report (in danish)
This project will develop a system for general segmentation and recognition of sound. When large amounts of audio and video material becomes available, e.g. through the digitalisation of the Danish Radio archives, new services such as TDC YouSee "TV archive etc. automatic segmentation and recognition of audio becomes an important part of being able to access this data.  The high degree of broadband penetration in Denmark allows for access to the media files and interactive use for ordinary users.

The system is being developed in order to identify shifts in the sound environment that users consciously or unconsciously register and thus seperates the sound recordings into cognitively coherent parts. These segments will be parts that can be described adequately with a few simple labels. Natural labels are e.g. audio type (voice or music), identities of peaple speaking, the sound environment (on the street, at a train station, in nature), and other specific sound sources.

Danmarks Radio og Institut for Kunst og Kultur, KU, Biblioteksskolen og DTU Informatik er blandt deltagerne i projektet.The Danish Broadcasting Corporation and the Department of Art and Culture at Copenhagen University, The Royal School of Library and Information Science and DTU Informatics is among the participants in the project.

Projektet skal munde ud i et værktøj, der kan præsentere lydfiler i en segmenteret form, således at filen er let tilgængelig for brugeren for videre analyse. Projektet knytter an til det større LARM projekt, hvis hovedformål er at skabe en humanistisk infrastruktur med digitale værktøjer, der gør det muligt for forskerne at søge i og beskrive den radiofoniske kulturarvs mange optagelser. The culmination of the project is a tool that can present audio files in a segmented form so that the files are easily accessible to the user for further analysis. The project ties in with the larger LARM project whose main objective is to create a humanistic infrastructure with digital tools that allows researchers to search and describe in the many recordings in the radiophonic heritage .


For more information contact 

Project Manager:

Jan Larsen
Associate Professor

Technical University of Denmark
Department of Informatics and Mathematical Modelling
Section for Cognitive Systems
Richard Petersens Plads, Building 321
Technical University of Denmark
DK-2800 Kongens Lyngby, Denmark


Direct: (+45) 45 25 39 23
Mobile: (+45) 22 43 00 25
Danish Sound Technology Secretariat: (+45) 45 25 34 11
Fax: (+45) 45 88 26 73
Skype: janflynut