Implementation of Segmentation and Classification at the Same Time
In the introductory talk, a background on audio segmentation will be given to the students: description of the task, applications, main issues to deal with, and the evaluation framework as defined in Albayzin evaluations. After this, the most habitual audio features (MFCCs and its derivatives, LPCs, etc) and the main state-of-the-art approaches for audio segmentation and classification will be introduced (BIC algorithm, HMM segmentation, SVMs). Right after, fusion techniques for combining different audio segmentation systems will be presented. Lastly, the Albayzin Audio Segmentation Evaluation carried out in 2010 and 2012 will be described: description of the task, datasets, evaluation metrics and performance achieved by the proposed systems. A brief summary of the main issues found by the participants will be given.