IDENTIFICATION OF AMERICAN SIGN LANGUAGE TO SPEECH CONVERSION USING TRANSFER LEARNING (SHUFFELNET)

IDENTIFICATION OF AMERICAN SIGN LANGUAGE TO SPEECH CONVERSION USING TRANSFER LEARNING (SHUFFELNET)

Vijay J¹, Vijayakumar J², Mervin Paulraj M³

¹Student, Department of Electronics and Instrumentation, Bharathiar University, Coimbatore

2Associate Professor & Head, Department of Electronics and Instrumentation, Bharathiar University, Coimbatore

³Research Scholar, Department of Electronics and Instrumentation, Bharathiar University, Coimbatore

---------------------------------------------------------------------***---------------------------------------------------------------------

Abstract - Sign language is used to facilitate communication between normal people and deaf and hard of hearing people. Many people are unaware of sign language, which creates a barrier to knowledge sharing and comprehension of thoughts and feelings. The convolution neural network is used in this study to train sign languages using the ShuffleNet transfer learning model. During training, the network accuracy is 99.89 percent and the testing accuracy is 84.76 percent. Matlab's converter is used to translate the output (Text to Speech).

Key Words:  Sign Language, Deep Learning, Transfer Learning, ShuffleNet, Neural Network.