Steve’s ALS Journey: Team Gleason Overview. Until There is a Cure for ALS, Technology is the Cure. A Caretakers Perspective on Independence Providing Technology. Accessibility Across all Operating Systems.
The spectrum of communication disorders treated by speech-language pathologists and audiologists will be described in relation to current service delivery challenges and some of the more prevalent unmet needs of people with communication disorders. Opportunities for optimizing detection and diagnosis, service delivery and clinical outcomes, evidence generation, access and communication participation will be discussed.
Language disorders affect the core linguistic functions that help us to formulate words and sentences. This complex system is supported by networks of brain regions that prepare these utterances for articulation. This presentation will describe how a brain injury can affect the language system, and will set the stage for discussions of how AI can assist people who suffer from such disabilities.
Our voice is our acoustical fingerprint, yet to date, authentic representation of self when using augmentative communication technology, remains elusive. This discussion will provide a brief overview of current methods of message banking and voice banking and examples of multilingual voices created and currently used by people with ALS will be demonstrated. The process for creating the demonstrated voices will be reviewed. Current challenges, both in terms of the current process and the need for efficiency through machine learning and the need to incorporate suprasegmentals of spe.
The tools that have resulted in our current AI boom have largely become commodity, and as a result increasingly what matters is data, design, and resources. I believe there will be a great "harvest" of technologies leveraging these tools in the coming decade driven largely by people at the intersection of AI and human-centered fields like HCI. I will share a few ideas on how we might increase and quicken accessibility's harvest in this AI Autumn.
What’s a LivingLab and how it can bring the best for patient dailylife
Application failure to implement accessibility standards is a failure of that application, but such failures are also shaped by many factors. We have proposed large-scale analyses of Android accessibility to identify factors that impact the accessibility of many applications, and we have also demonstrated how such large-scale analyses might inform large-scale repair of those failures. I will present highlights from both threads of research, the opportunity for leveraging AI in scaling such approaches, and how this can contribute to a vision of community-driven accessibility.
ASL-Search is a way for ASL learners to find the English meaning of American Sign Language (ASL) signs they see in the wild. A query to ASL-Search is a list of features (number of hands, hand shape, hand orientation, location and motion) they observe. Straight-forward feature matching is not sufficient for accurate search because of observation errors and sign variation. Latent semantic analysis helps accuracy. Importantly, successful queries will add additional data to improve ASL-Search for future users.
We describe Parrotron, an end-to-end-trained speech-to-speech conversion model that maps an input spectrogram directly to another spectrogram, without utilizing any intermediate discrete representation. The network is composed of an encoder, spectrogram and phoneme decoders, followed by a vocoder to synthesize a time-domain waveform. We demonstrate that this model can be trained to normalize speech from any speaker regardless of accent, prosody, and background noise, into the voice of a single canonical target speaker with a fixed accent and consistent articulation and prosody. We further show that this normalization model can be adapted to normalize highly atypical speech from a deaf speaker, muscular dystrophy, and speakers with ALS, resulting in significant improvements in intelligibility and naturalness, measured via a speech recognizer and listening tests. Finally, demonstrating the utility of this model on other speech tasks, we show that the same model architecture can be trained to perform other speech tasks, such as speech separation.