Grant: Collaborative Research and Development Grants
University of Toronto
$129,820.00
Towards Articulatory-Based Adaptation in Dysarthric Speech Recognition
FY 2009-10
Summary
Automatic speech recognition (ASR) has produced mature and widespread tools for the general public, but remains effectively unusable by individuals with severe speech disorders. One group of such disorders, called dysarthria, is caused by neuromuscular disorders such as cerebral palsy, multiple sclerosis, Parkinson's disease, stroke. While individuals with dysarthria might be able to form and comprehend full sentences, they would lack control over the muscles used in speech, producing unintelligible speech. The neurological damage that causes dysarthria also affects other physical activity, such as slowing keyboard input 150 to 300 fold in severe cases compared with regular users (Hosom et al., 2003; Hux et al., 2000). Speech recognition has the potential to be a viable alternative to keyboard input if poor speech can be reliably recognized.
The goal of this proposed research is to develop new adaptive ASR software that significantly improves factors such as accuracy and accessibility for dysarthric speech. We will focus on individuals whose speech disabilities limit the use of modern ASR, and for whom keyboard or other alternative interaction methods are insufficient. We intend to accept unrestricted speech entry for novel sentence dictation on computers, telephones, or other devices. Our approach will be unique in that it will be informed by both physiological models of speech production and acoustic data. These models will be constructed by machine learning algorithms upon data to be acquired within the first year of the project. That data will combine 3D and 2D motion parameters of the speech articulators with acoustic data for both dysarthric and non-dysarthric speakers.
In the long-term, we anticipate that a commercialization of resulting technology will provide direct social benefit by enabling access to services currently impeded by poor speech. Since dysarthria is often a consequence of more general neuromuscular disability prohibiting other kinds of physical movement, dramatic improvement of ASR accuracy is essential, as users with disabilities would likely benefit the most from being able to control devices through speech, as it allows a faster and less fatiguing access mode.
Program
Collaborative Research and Development Grants
Recipient
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
364360-2007
Name nom
Hirst, Graeme
Department département
Computer Science
Organization id
31
Institution établissement
University of Toronto
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2009
Competition year année de concours
2007
Award amount
129820
Program id
CRDPJ
Program nane en
Collaborative Research and Development Grants
Program name fr
Subventions de recherche et développement coopérative
Group en
Research partnerships programs
Group fr
Programmes de partenariat
Committee code
96
Committee name en
RPP Internal Decision Cttee
Committee name fr
Cté de décision interne - PPR
Area of application code
801
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Computer software
Area of application fr
Logiciel
Research subject code
2801
Research subject group en
Artificial intelligence
Research subject group fr
Intelligence artificielle
Research subject en
Natural language and speech understanding
Research subject fr
Langage naturel et reconnaissance de la parole
Partie
2
Nb partie
3
Application title
Towards Articulatory-Based Adaptation in Dysarthric Speech Recognition
Keyword
NA
Application summary
Automatic speech recognition (ASR) has produced mature and widespread tools for the general public, but remains effectively unusable by individuals with severe speech disorders. One group of such disorders, called dysarthria, is caused by neuromuscular disorders such as cerebral palsy, multiple sclerosis, Parkinson's disease, stroke. While individuals with dysarthria might be able to form and comprehend full sentences, they would lack control over the muscles used in speech, producing unintelligible speech. The neurological damage that causes dysarthria also affects other physical activity, such as slowing keyboard input 150 to 300 fold in severe cases compared with regular users (Hosom et al., 2003; Hux et al., 2000). Speech recognition has the potential to be a viable alternative to keyboard input if poor speech can be reliably recognized.
The goal of this proposed research is to develop new adaptive ASR software that significantly improves factors such as accuracy and accessibility for dysarthric speech. We will focus on individuals whose speech disabilities limit the use of modern ASR, and for whom keyboard or other alternative interaction methods are insufficient. We intend to accept unrestricted speech entry for novel sentence dictation on computers, telephones, or other devices. Our approach will be unique in that it will be informed by both physiological models of speech production and acoustic data. These models will be constructed by machine learning algorithms upon data to be acquired within the first year of the project. That data will combine 3D and 2D motion parameters of the speech articulators with acoustic data for both dysarthric and non-dysarthric speakers.
In the long-term, we anticipate that a commercialization of resulting technology will provide direct social benefit by enabling access to services currently impeded by poor speech. Since dysarthria is often a consequence of more general neuromuscular disability prohibiting other kinds of physical movement, dramatic improvement of ASR accuracy is essential, as users with disabilities would likely benefit the most from being able to control devices through speech, as it allows a faster and less fatiguing access mode.