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]