Albayzin Evaluation: The PRHLT-UPV Audio Segmentation System
Joan Albert Silvestre-Cerdà, Adrián Giménez Pastor, Jesús Andrés Ferrer, Jorge Civera Saiz, Alfonso Juan Císcar · RiuNet (Universitat Politècnica de València) · 2012
This paper describes the audio segmentation system developed by the PRHLT research group at the UPV for the Albayzin Audio Segmentation Evaluation 2012. The PRHLT-UPV audio segmentation system is based on a conventional GMM-HMM speech recognition approach in which the vocabulary set is defined by the power set of segment classes. MFCC features were extracted to represent the acoustic signal and the AK toolkit was used for both, training acoustic models and performing audio segmentation. Experimental results reveals that our system provides an excellent performance on speech detection, so it could be successfully employed to provide speech segments to a diarization or speech recognition system.