Automated Speech Recognition for a Personal Emergency Response System (PERS)
Keywords: Speech recognition, speech-based PERS, automated emergency response.
Background
The use of traditional push-button personal emergency response systems (PERS) in the home of the older adult has been shown to ease both caregiver and user anxiety, support aging-in-place, and decrease overall healthcare costs. However, only a small proportion of the total older adult population uses these devices. Reasons for non-use include stigmatization or burden from always having to wear a button, fear of losing independence, inability to access/use their system (e.g., difficult to push button), system cost, and other personal reasons [1,2]. As well, the majority of calls to the emergency call centre tend to be false alarms or accidental, further stressing already limited emergency resources.
With an increasing senior population and the subsequent rise in need for healthcare and social services it is important to ensure that potentially life-saving assistive technologies that support aging-in-place, such as the PERS, be made usable, effective and efficient for future generations.
Research indicates that older adults are open to using various assistive technologies, however, the technology must work well and fill a need before the technology is adopted [3,4]. Older adults have also been found to be very receptive to technologies which are controlled using speech alone [5,6].
It is hypothesized that if the user is not required to wear any actuators, then the burden and stigmatization of wearing “the button” will be eliminated. Furthermore, if the system was intelligent, allowing the user to cancel accidental calls for assistance, change their mind over the course of time and call for help when desired, or adapt the dialogue to the user’s previous calls for assistance, then these combined changes should improve overall PERS efficiency, effectiveness and usability.
Overview of Research
To improve PERS effectiveness, efficiency and usability, IATSL is working on developing an automated, speech-based PERS as an alternative to the basic push button activators, shown in Figure 1.
To make the system more intuitive and simple to use, a communication and response module (CRM) was developed for the PERS incorporating speech recognition and artificial intelligence [7,8]. This prototype was successfully tested in a controlled laboratory environment with younger adults using a custom-built microphone array. The next step in this project is to further optimize the system’s artificial intelligence and speech recognition, to make it more robust in a less controlled, more home like laboratory environment and to perform testing with lower cost electronics and with older adult subjects in different emergency scenarios.
Figure 1. Flow diagram of a conventional PERS (solid lines) with the addition of the HELPER system (dotted lines).
Project Methodology
This research project has been divided into three phases: (I) System Design, (II) System Development and (III) System Evaluation; each phase has specific research objectives.
Phase I: System Design
- To examine how older adults interact and use existing personal emergency response services.
- To identify limitations of the existing communication and response module (CRM) personal emergency response system (PERS) prototype.
- To identify a new communication and response procedure for the PERS CRM.
- To develop a speech corpus of older adult and adult emergency speech.
Phase II: System Development
- To further develop the speech recognition system.
- To further develop the artificial intelligence system.
- To conduct preliminary testing throughout the development stage and make necessary modifications
Phase III: System Evaluation
- To evaluate the PERS CRM ASR trained with older adult specific voices and older adult and younger adult voices with older adult speech input.
- To evaluate the PERS CRM technology using various emergency scenarios, older adult speech input, and the best ASR system from the previous experiment.
Progress to Date
Currently, this project is in Phase I: System Design. Emergency call recordings are being analyzed to examine various call features such as older adult and call taker dialogue, type of calls, and common word usage. Next, with this information we will start development of an older adult speech corpus consisting of emergency type speech, spoken and read, with which we can use to train the speech recognizer.
Future Work
To complete phase I, the older adult speech corpus will be used to expand the speech recognition vocabulary and will also be used to target the speech towards recognizing older adult voices. The system dialogue will also be improved to recall previous requests and to handle situations such as a silence, calls for help, and different emergency situations.
References
- Mann, W.C., Belchior, P, Tomita, M. and Kemp, B.J. (2005). Use of personal emergency response systems by older individuals with disabilities. Assistive Technology, 17, 82-88.
- Porter, E.J. (2005). Wearing and using personal emergency response systems. Journal of Gerontological Nursing, Oct, 26-33.
- Mann, W.C., Marchant T., Tomita M., Fraas L. and Stanton K. (2001-2002 Winter). Elder acceptance of health monitoring devices in the home. Care Management Journals, 3(2), 91-8.
- Demiris, G. et al. (2004). Older adults’ attitudes towards and perceptions of ‘smart home’ technologies: a pilot study. Medical Informatrics & The internet in Medicine, 29(2), 87-94.
- Johnson, J.L., Davenport, R. and Mann, W.C. (2007). Consumer Feedback on Smart Home Applications. Topics in Geriatric Rehabilitation, 23(1), 60-72.
- Anderson, S. et al. (1999). Recognition of elderly speech and voice-driven document retrieval. IEEE, proceedings of the ICASSP, p.145-148.
- Hamil, M., Young, V., Boger, J. and Mihailidis, A. (2009). Development of an automated speech recognition interface for personal emergency response systems. Journal of NeuroEngineering and Rehabilitation, 6(26).
- Mihailidis, A. et al. (2006). An Intelligent Health Monitoring and Emergency Response System. ICOST paper. Gerontechnology, 4(4),209-222.
Funding Sources
Natural Sciences and Engineering Research Council (NSERC)
CIHR strategic fellowship in Healthcare Technology and Place
Engineers Canada-TD Insurance Meloche Monnex
Toronto Rehabilitation Institute
Graduate Department of Rehabilitation Science, University of Toronto
Research Team
Alex Mihailidis, University of Toronto
Elizabeth Rochon, University of Toronto
Vicky Young, Ph.D. student, University of Toronto
Babak Taati , University of Toronto
Jennifer Boger, University of Toronto
Rozanne Wilson, University of Toronto
