Peruvian Sign Recognition (LSP) to the Native Quechua Language Using LSTM
Seline M. Maquera, Jesus E. Rocca, Honorio Apaza Alanoca, Victor Yana, Carlos Antônio da Silva · 2024
Sign language is not universal; there are over 300 recognized sign languages worldwide, according to the World Federation of the Deaf. This number adds to the nearly 7000 spoken languages, such as English or Spanish. More than 430 million people worldwide, representing over 5% of the global population, experience disabling hearing loss that requires rehabilitation. Of these, an estimated 34 million are children. By 2050, this number could exceed 700 million, affecting 1 in 10 people. In Peru, approximately 532,000 people face daily challenges due to permanent hearing loss, relying on various communication methods, including voice, gestures, and sign language. However, sign language, although effective, is adopted by a minority due to the time and dedication required to learn a new language. This project focuses on developing a model to recognize Peruvian Sign Language (LSP) and translate it into Quechua, one of the native languages of Peru, aiming to bridge the communication gap. The manuscript details the implementation of a Long Short-Term Memory (LSTM) model, incorporating MediaPipe for sign language recognition. After data preprocessing, the LSTM model accurately recognizes and translates signs into Quechua. The research aims to facilitate a greater understanding of sign language, empowering individuals to participate fully in society. Moreover, advancements in artificial intelligence, particularly in LSTM models, demonstrate the potential for sign language recognition with 91.16% categorical accuracy, offering effective solutions to the communication barriers people with hearing loss face.