lead to both (involuntary) asynchrony between articulatory ges- tures and (voluntary) alternative articulatory strategies, which in turn yield insertion of sounds not found in the target words, splits of target sounds (one to many), repetitions of target sounds, substitutions of entire sound sequences (many to many) and in- sertions of pauses inside words. For example, too early opening of the nasal passage to airflow, which reduces air pressure above the vocal folds and facilitates their vibration, often results in the presence of an unexpected nasal consonant (like n or m) before another consonant, or of nasal coloring of an entire syllable. Recognizing the complexity and the unique needs and char- acteristics of the users it serves, the core technology that forms Voiceitt’s multi-layered solution is built upon two main ASR archi- tectures adapting state-of-the-art ASR techniques to recognize dysarthric speech. Voiceitt's is an accessible app, designed to serve individuals with dysarthria caused by medical conditions, such as Cerebral Palsy, Stroke, ALS, Parkinson’s disease, Down's syndrome, Trau- matic Brain Injury and any other type of non-standard speech to communicate and control smart assistants with their OWN VOICE for the first time. Voiceitt’s app identifies and adapts to the individ- ual’s unique impaired speech patterns like breathing pauses and
non-verbal sounds so that individuals with severe speech impair- ment can use it to communicate. Voiceitt does not need an Inter- net connectivity for usage and can be used everywhere. Technologically, Voiceitt lives in a world where all state-of-the- art ASR (Automatic Speech Recognition) systems are designed to recognize standard speech in a variety of native languages from around the world. Being a speech recognition technology that aims to serve millions of individuals around the world with im- paired speech, Voiceitt cannot use existing ASR systems as they were initially designed - we innovatively combine systems, tools, methodologies, sometimes even pure theoretical research in the field of speech recognition, and apply everything in an app that intelligently recognizes the user’s unique speech patterns and translates his speech into standard pronunciation which other people (and smart assistants) can understand. Voiceitt builds a custom ASR speech model, to each user, us- ing machine learning, utilizing deep neural networks, we extract unique speech features used to train our system with the individ- ual’s speech. Once ready, the personal custom ASR model is load- ed into the app where it is used to identify the user’s speech and translate it. It is very unique in a sense that it naturally uses speech/voice
4
www.closingthegap.com/membership | December, 2020 / January, 2021 Closing The Gap © 2020 Closing The Gap, Inc. All rights reserved.
BACK TO CONTENTS
Made with FlippingBook Ebook Creator