An optimized machine translation technique for multi-lingual speech to sign language notation

Due to the lack of assistive resources, hard-of-hearing people cannot live independently. Sign language or gesture language is the natural language and it is the primary mode of communication for hard-of-hearing people. Researchers and IT companies are continuously trying to find the best solutions to minimize the communication barriers for hearing-impaired people. Existing translation techniques for speech to sign language on the web platform are consuming higher resources. This study presents an optimized technique for direct machine translation of multi-lingual speech to Indian sign language using the HamNoSys notation system, whereas existing techniques were translating speech-text-HamNoSys. Performance comparison of both existing and the proposed techniques is analyzed in this study. The proposed technique optimizes the resources for the following parameters: CPU, heap memory, primary memory, and classes load. The result shows that the existing technique takes 220 MB heap memory, 10 threads, 2236 classes, and CPU for 12 s. The proposed technique consumes only 210.4 MB, 9 threads, 2113 classes, and CPU for 9 s.