The authors have been studying sign language recognition using color gloves and optical camera. We extract features using AlexNet as a method of transfer learning from sign language motion data which is obtained after extracting the colored region from raw data. Long Short Time Memory (LSTM) is applied to learning and recognition for time series feature data obtained by this method. In this paper, we examine the relationship between frame intervals and recognition performance by changing the frame interval of sign language motion data, and describe the result of clarifying the frame interval required for sign language recognition.