Generation of a vocal-tract MRI movie based on sparse sampling