Mexican Spanish viseme phoneme /u/ dataset

Multimodal data was captured using a MS Kinect V2 sensor connected to a laptop. RGB images (24 bits per pixel) and depth images (16 bit per pixel) sequences were captured at 30 frames per second.

The synchronized audio of the phoneme was recorded in a WAV file at 44 Khz, 16 bits per sample. For each phoneme performance the following data are available for the mouth region: 90, 256x256 RGB images (24 bits per pixel) in PNG format 90, 256x256 depth images (16 bits per pixel) in PNG format One WAV file at 44 Khz, 16 bits per pixel To facilitate analysis, processing, machine learning and inference methods to be developed the following additional files are provided which are derived from the above raw data: 191 facial landmarks (x,y) extracted with MediaPipe software for each RGB image.