I stumble upon this video today about features extraction in audio files using Adaboost algorithm. Douglas Eck, from University of Montreal, is playing an important role in the BRAMS and his group aims to develop some kind of algorithm to automatically label songs by genre. It’s interesting to see how such technology could be scaled to a massive amount of information for a search engine like Google. He also shows up differences between his kind of approach compared to those used by Amazon, last.fm or Pandora. Eck and his students claim achieving 83% success rate using Boosting algorithm.
[Via IM2 | OQP]