Thursday, November 20, 2008

Extracting meaning from audio signals

Associate professor Jan Larsen, from the Intelligent Signal Processing Group at IMM, Technical University of Denmark, gave a presentation on extracting meaning from audio signals. He gave several examples on how to process audio signals to extract information of, for instance, used instruments and genre. Other examples showed how several information sources can be combined to create search engines for music.

He highlighted the increasing importance of cognitive systems and machine learning in future research. As current emerging interests he stated sparse unsupervised models, generation of high-level contextual information, semisupervised learning and user modeling. Games with a purpose, such as the ESP game and reCAPTCHAs, were mentioned. There users typically entertain themselves, and at the same time happen to tag images, recognize digitized text or otherwise augment raw data with high-level meta-data. The results can be used as additional training data for machine learning methods for creating better applications.

The next day, he acted as the opponent in Jarkko Salojärvi's dissertation on using wrong models.

No comments: