[Adapt] [Seminar] Auditory Scene Recognition

[huaxinyu at sjtu.edu.cn]

Hi ADAPTers,

Xun and I will give a talk in this week's seminar.The one-hour seminar will feature two separate parts of our ongoing project"Auditory Scene Recognition". As the name suggests, our goal in this project is to infer the most likely scene given a sound clip.The main difference between our approach and conventional approaches is,we assume those scenes are composed of multiple primitive events which can presumably tell us some characteristics about the scene. We use textual corpus to build such scene-event map(Text part), and use GMM model to detect primitive events(Audio part), then combine the scene-event map with the detected events to infer the most likely scene where the sound clip was recorded. Hope you enjoy the talk.

See you tomorrow at 5pm. 

All the bests,
Xinyu Hua