Glottal Gap tracking by a continuous background modeling using inpainting - Database

The corpus used for evaluation consists of 54 high-speed sequences: 36 females (67%) and 18 males (33%). Each sequence has 400 frames, thus the number of images analyzed is 21.600. The recording took place at the ENT service of the Gregorio Marañon Hospital in Madrid. The videos were recorded during a sustained vowel phonation, including in some cases the vocal onset. The high-speed sequences were acquired using the camera system WOLF HRES ENDOCAM 5562 and a rigid endoscope with an angle of view of 70◦. The light source was the AUTO LP 5132 and all the videos were recorded in color. The sampling rate was 4000 fps and the spatial resolution of 256 × 256 pixels. The distance between the head of the camera in the oropharynx and the vocal folds is variable. The database includes usual phonatory modes as Mode I (the most common phonatory mode), lesions in the vocal folds (polyps and nodules), paralysis, paresis, postoperative papillary thyroid cancer, patients with multinodular goiter, and diplophonia postoperative. The recordings present different illumination levels, contrast, partial occlusion of the glottis, and lateral displacements