Continuous Translation of Brazilian Sign Language in the Context of Health with Deep Learning
Accessibility, Libras, Deep Learning, Sequence-to-sequence, Machine Translation.
In Brazil, the deaf represent about 5% of the population, approximately 9.7
millions of Brazilians. Although Brazilian Sign Language (Libras) is one of the
official waters of Brazil, knowledge and mastery of Libras among non-deaf people is
an obstacle, creating language barriers in accessing basic rights, especially
in access to health services. This motivated the development of government policies.
that oblige service providers to provide Libras interpreters to enable
access to these services by the deaf community. However, this type of solution
presents a very high cost of implantation and maintenance. From this perspective, it
necessary to develop research and automated methodologies for translation
Automatic Libras. Thus, in this work, we proposed a methodology to
continuous production of Libras. The proposed solution does not require any additional hardware,
relying entirely on images or image sequences (videos). Furthermore,
a new dataset for continuous recognition of Libras was introduced, con-
having 10500 videos of 105 distinct sentences in the context of clinical screening. The experiences
computational data obtained a WER of 21.62 and a maximum accuracy of 92.68%,
for a set of tests with samples and interpreters never seen by the model during
the training.