Improving Keyword Search in Sign Language using Similarity Models

Although it is a well-established research subject for domains like speech processing and text analysis, keyword search studies for sign language has not attracted enough attention from the community. In this paper, we explain the components of the previously proposed keyword search system. We demonstrate that the existing model can also be trained with similarity based methods using cosine and triplet losses. We show that fusion of existing models with similarity based models improves the retrieval performance. All experiments are conducted on RWTH-PHOENIX-Weather 2014T dataset and mean Average Precision scores are reported.