Department of Computational Perception
Department of
Computational Perception
 Johannes Kepler Universität Linz

Home  –  Mission  –  Teaching  –  People  –  Research  –  Media  –  Awards  –  Impressum

TV & RADIO ANALYSIS

<< back to overview


Audio Fingerprinting

Audio Fingerprinting aims at identifying an audio snippet given a large database of reference audio files (e.g. SoundHound, Shazam). Our institute developed an audio fingerprinter that is robust to all kinds of distortions, including severe time stretching and/or pitch shifting.

We offer the following topics for theses and practical works:

  • Exploring Attack Vectors on (Landmark-Based) Audio Fingerprinting Systems
    Find and evaluate methods that change audio in a way that fools audio fingerprinters but is inperceptible to humans.
  • Developing a Client/Server Architecture and/or a Mobile Application
    Make our fingerprinter usable from from the web.
  • Implement (and/or Improve) Modules of Our Fingerprinter In C for Speedup
    For example: A Robust, Scalable Method to Extract Peaks from Spectrograms
Contact: Reinhard Sonnleitner, Gerhard Widmer

Recognising Sound Objects in Audio Streams

Given the gigantic and constantly growing amount of audiovisual material that is archived in digital form, there is a tremendous need in the media industry for methods that automatically extract semantic information from audiovisual data streams. In order to index and effectively search in huge multimedia databases, we need computers that can detect and name all sorts of recognisable sound objects in audio streams. The image to the right pertains to a real-world application, where TV channels are automatically monitored for the presence of music (in the foreground or background), using a music detection algorithm developed by us (in cooperation with the Austrian Research Institute for Artificial Intelligence, Vienna).
  • Detection of Rap/HipHop Music
    Detecting Rap/HipHop challenges music detection systems because the vocals are difficult to distinguish from spoken language (for the computer). The goal is to develop new features and/or methods that overcome this problem.
  • Automatic Detection and/or Recognition of Commercials and Jingles in TV and Radio
  • Detection of Specific Musical Instruments
    Develop methods that detect the presence of an instrument in the audio.
Contact: Reinhard Sonnleitner, Gerhard Widmer

Speech/Singing Detection and Classification

Contact: Reinhard Sonnleitner, Gerhard Widmer


last edited by kf on Sep 28, 2015