RWC Multi-modal Database of Gestures with Speech

Next we have performed experiments to evaluate to proposed recognition method which uses the HMM based recognizer. In this experiment, we used image sequences from RWC multi-modal database of gestures with speech[13], which was collected under the Real World Computing Project. The contents of the database are images of gestures of an upper body and speech. The database is available for research purpose as CD-ROMs. The current version includes 12 subjects and each person has 4 repetitions of 25 patterns of gestures. Each gesture was sampled with video rate (33 frames per second). The size of each image is $320 \times 240$ pixels. The length of gestures is about 3 or 4 seconds.