A control-theoretic model of speech production and recognition for use within prosthetic communication devices

点击翻译按钮获取中文摘要

Millions of people in North America have some sort of speech disorder. Many of these have normal mental abilities and are able to process language, but they lack normal control over the muscles used in speech (such as the lips and tongue), resulting in speech that can be very difficult or even impossible to understand. Other individuals have cognitive disorders (for example, caused by Alzheimer's disease) where it becomes difficult to understand what is being said, to remember the right words, or to produce sentences. I am starting a new lab called SPOClab whose purpose is to produce software to help people with disabilities communicate. Our goal is to improve the core technology in speech recognition and to use that technology within applications that help people in their daily lives. Improving speech recognition for individuals with speech and language disorders will involve fundamentally changing the way speech recognition is done currently, because modern speech recognition simply doesn't work properly for very different kinds of speech. Our new way of doing speech recognition will involve teaching the computer to understand how speech is physically produced. For example, if the computer knows that the speaker tried to close their lips, it becomes easier to tell the 'm' sound in 'map' apart from the 'n' sound in 'nap'. The advanced speech recognition developed in SPOClab will be used to help people communicate. For example, we will transform hard-to-understand speech signals to be more intelligible to human listeners and we will use video as a mode of input to computer applications. This technology can have a tremendous positive impact for many Canadians, including those Canadians for whom a loss of linguistic abilities accompanies ageing.