Indonesian Sign Language (SIBI) Recognition and Extraction Using Convolutional Neural Networks - Symmetric Deletion Spelling Correction

Communication through sign language is a profound and essential form of human interaction, particularly for individuals who are deaf or hard of hearing. Sign language, with its roots dating back to the 17th century, has evolved into a sophisticated visual language and communication method used globally by the deaf community. Technological advancements, notably the integration of sign language into video communication platforms, have significantly improved accessibility and inclusivity. However, converting sign language gestures captured in video to text presents technical challenges, including duplicated letters and inconsistent formatting due to articulation irregularities and video frame rates. To address these challenges, innovative solutions leveraging machine learning, computer vision, and natural language processing are necessary. This paper explores a novel approach, utilizing the Indonesian Sign Language system (SIBI), to improve the accuracy and coherence of transcribing sign language gestures into text, enhancing communication for the deaf community and individuals with hearing impairments. This research shows that the classification of SIBI using keypoint features achieves an accuracy of 97%. Additionally, spelling correction yields favorable outcomes through preprocessing steps, such as eliminating duplicates and distortions, thereby enhancing the correction process. Using Symspell for spelling correction also yields good results with a fast computational process, taking only 0.883 seconds.