Deaf and Hard of Hearing

This work releases new resources from the Public DGS Corpus, a long term project that aims to build a reference corpus of German Sign Language and research portals. Extension of the corpus involves the addition of new data formats, including OpenPose pose information, and corrections to the subtitles and annotations.

A dataset of Bangla sign language digits and characters is collected to enable research in automatic sign language recognition and develop educational tools. A total of 2800 images are captured from unspecified native Bangla sign language signers. 10 digits and 36 characters were performed, with 100 and 50 images each, respectively.

The study aims for modeling automated recognition of grammatical facial expressions from Brazillian Sign Language (Libras). Videos of a fluent signer in a Libras performing Grammatical Facial Expressions were captured and represented by spatial and temporal information from image frames using Microsoft Kinect Sensor.

This work aims to build a corpus of multilingual sign language using stero cameras and annotate with iLex. Stero cameras recorded the gestures and iLex was used to annotate the corpus.

Naturally occurring sign language videos are collected from online sources like YouTube and Deaf social media for American Sign Language (ASL) fingerspelling recognition. ChicagoFSWild+ contains 55,232 fingerspelling sequences signed by 260 signers, along with crowdsource annotations.

This multimodal and multiview continuous American Sign Language (ASL) dataset consists of a parallel corpus to facilitate progress in the areas of sign language recognition, translation, and production. A total of 80 hours of sign Language videos were recorded from multiple angles, and gloss annotations and a coarse video categorization were performed.

The Swedish Sign Language Corpus aims to document sign language discourses and is freely available with the web-based corpus tool STS-corpus, for use in teaching, research and sign language lexicography. The recordings consist of free conversations and storytelling, as well as retellings of, for example, “The Snow Man”. The recorded materials are annotated and transcribed.

This work collects video clips showing Deaf people using Polish Sign Language to create a corpus-based dictionary. Each recording involves two signers reacting to visual stimuli, e.g. retelling the content of picture-stories and video clips, naming an object, talking about themselves, and discussing topics of interest to them.

The dataset of the Bangladeshi Sign Language letters is released for deep learning classification and detection, which comes with raw images, background removed and augmented images, and labelled bounding box. 1200 images were captured by volunteers at their convenience and natural environment using their phone cameras or webcam.

Language recording sessions were conducted with deaf native or early learner/near-native users of the Australian Sign Language to identify the principles that need to be adhered to in the creation of signed language corpora. Unedited video footage data was collected where each participant took part in language-based activities that involved an interview, the production of narratives, responses to survey questions, free conversation, and other elicited linguistic responses to various stimuli such as a picture-book story, a filmed cartoon, and a filmed story told in Auslan.