Using Fingerprinting to Aid Audio Segmentation
Lopes, C.
; Veiga, A.
;
Perdigão, F.
Using Fingerprinting to Aid Audio Segmentation, Proc GTM, RTTH and SIG-IL VI Jornadas en Tecnología del Habla and II Iberian SLTech Workshop - FALA, Vigo, Spain, Vol. -, pp. - - -, November, 2010.
Digital Object Identifier:
Abstract
Audio segmentation is an important preliminary task in audio description systems. In this paper a three-phase audio segmentation scheme is proposed, where the first phase
detects silence events, the second phase detects audio repetitions trough fingerprinting and the last phase uses a hybrid HMM/ANN system to classify the remaining segments. Fingerprinting is an important aspect of audio segmentation in broadcast audio, due to the omnipresence of advertisements, jingles and even repeated programs. Detecting repetitions can be extremely fast and accurate and also permits to enhance the consistence of the annotations. Results using the Catalan broadcast 3/24 TV channel are reported.