Broadcast News Segmentation with Factor Analysis System

This paper studies a novel audio segmentation-by-classification approach based on Factor Analysis (FA) with a channel compensation matrix for each class and scoring the fixed-length segments as the log-likelihood ratio between class/no-class. The system described here is designed to segment and classify the audio files coming from broadcast programs into five different classes: speech (SP), speech with noise (SN), speech with music (SM), music (MU) or others (OT). This task was proposed in the Albayzin 2010 evaluation campaign. The article presents a final system with no special features and no hierarchical structure. Finally, the system is compared with the winning system of the evaluation (the system use specific features with hierarchical structure) achieving a significant error reduction in SP and SN. These classes represent 3/4 of the total amount of the data. Therefore, the FA segmentation system gets a reduction in the average segmentation error rate that is able to be used in a generic task.