Elaine Lu and Julie Chan, University of Toronto & Holland Bloorview Kids Rehabilitation Hospital
ABSTRACT
This study presents the development and evaluation of a vocal cord vibration switch as a communication access pathway. The switch uses vocal cord vibrations as an input to activate a switch, which produced an output in a computer, communication device, or environmental control. The vocal vibration switch was further developed to allow for two outputs based on the duration of the input (short or long). The dual output device was tested on three individuals who had different levels of dysarthria and who did not have another acceptable access solution. As a single output device, it had a specificity of 99.3% and an average sensitivity of 95.3%. As a dual output device, it had an average ‘short’ signal sensitivity of 71.1% and an average ‘short’ signal specificity of 77.5%. The average ‘long’ signal sensitivity was 60.2% and its specificity was 67.3%. The device was found to be easy to set up and non-fatiguing for participants.
BACKGROUND
Individuals with complex communications needs are often limited in their ability to control their environment, communicate with others, and use a computer. Some are able to use augmentative and alternative communication (AAC) systems with access pathways which include mechanical switches, infrared sensors, and computer vision systems.
Using the voice for communication and environmental control is a natural access pathway for those with control of their vocalizations. However, individuals with complex communication needs often are unable to use voice recognition devices and software due to dysarthria. Voice based switch activation has been developed with simple sound activated systems [1], where any sound recorded from a microphone produces a single signal, however there are drawbacks with unintended sounds (i.e. coughs), ambient noise, and user fatigue, in addition these systems are often limited to having one signal from a sound activated device.
PROBLEM STATEMENT
Three children with complex communication needs required an access solution. They were able to vocalize unintelligibly, however they were unable to successfully use current market solutions to access communication devices, computers, and environmental controls.
METHODOLOGY
Single Output Vocal Cord Vibration Device
It was found that vocal cord vibrations, such as hums and speech created highly periodic signals (Figure 1), whereas coughs (Figure 2) and swallows (Figure 3) created low periodic or aperiodic signals. These coughs, swallows and other unintended sounds were filtered, so as to create a signal only when the user intentionally hummed or spoke. Unlike a microphone, ambient noise was not picked up, however low vocalizations were able to be detected.
Figure 1. High periodic accelerometer signal from a hum, voltage of signal versus time.
Figure 2. Low periodic accelerometer signal from a cough, voltage of signal versus time.
Figure 3. Aperiodic accelerometer signal from a swallow, voltage of signal versus time.
The vocal cord vibration device was developed which measured periodic signals from a sensor attached to the neck [2,3]. It consisted of two accelerometers encased in a plastic mold inserted on a neckband. The accelerometers were attached to a signal processing device, which was in turn connected to computers, communication devices, or environmental controls through a USB cable. The neckband, with a safety release in the back was put around the neck of a user, see Figure 4.
Figure 4. Vocal cord vibration switch on a user.
Multiple Output Vocal Cord Vibration Device
It has been shown that even those with severe dysarthria were able to control sound duration and pitch [4], which would enable them to control more than one output signal. Enabling individuals to have multiple responses based on differing activation would give them even greater control of their environment and communication [1].
Figure 5. Schematic drawing of the vocal cord vibration switch used in conjunction with a computer.
This device was altered to separate out signals based on either sound pitch or duration by altering the signal processing. Interviews with the three children found that sound duration was easier for the children to produce than pitch variation. Different words could be associated with different durations, which would produce two different signals. This was more meaningful to the participants than producing two different pitches of the same word. The signal processing was altered to send one signal for a short duration and a different signal for a long duration, which could correspond to two different keystrokes on a computer, or two different functions on an environmental control device. An adjustable dial was then added to adjust the threshold between the short and long duration.
Participant Description and Study Methodology
Three children between the ages of 9.2 and 14.5 with differing levels of dysarthria were identified as being able to have some control over sound duration. They were given a modified 5 point Borg scale [1=not tired at all, 5=very tired] to measure their tiredness before and after the sessions to determine their level of exertion. During the 30 minute sessions they were asked to speak two words, one long (e.g. there you are/Alexander/go right) and one short (e.g. go/Bob/left), to determine the appropriate threshold level. They were then given three different two-switch activities to complete: one with two boys kicking a ball, one with two pistons hitting two balls, and a step scan activity. The sessions were all videotaped and reviewed for true activations for each switch, false activations (when the device would send a signal, but no sound was made by the participant), false negatives (when a sound was made by the participant and the device sent no signal), and unintentionally activating the other switch (when an intended short sound was made, but the long sound switch was activated and vice versa). Sensitivity and specificity was calculated based on the data. The data collection sessions were all conducted at a children’s rehabilitation center.
RESULTS
It was found that as a single‐output solution, that is it would create a signal when there was any vocalization, the device achieved on average 95.3% sensitivity and 99.3% specificity. As a dual‐output device, it had an average ‘short’ signal sensitivity of 71.1% and an average ‘short’ signal specificity of 77.5%. The average ‘long’ signal sensitivity was 60.2% and its specificity was 67.3%.
When participants were asked to rate their exertion levels on a five‐point scale before and after each 30‐minute experiment, no increases in exertion levels were observed.
Verbal feedback on the device was positive from all participants, in addition all continued using the device after the study was completed.
DISCUSSION
This device was developed in conjunction with the help of an interprofessional team of therapists, engineers, and end-users. Using user centred approach has made the device more user friendly and practical.
The Hummer as a single output device had good sensitivity and specificity. It was very good at not picking up environmental sounds. The device was not overly tiring to use for those in the study, as they did not need to vocalize too loudly. It was found to be easy to set up, as there was no mounting required. In addition, unlike microphone based voice activating devices, it did not interfere with hearing aids.
Some of the challenges of the Hummer included not picking up unvoiced letters or overly breathy sounds, as these did not vibrate the vocal cords. Another drawback was that the Hummer would pick up laughter and crying, which did vibrate the vocal cords in a periodic way. In addition, the neckband would sometimes move the accelerometers away from the vocal cord folds, causing the device not to pick up sound. This could be remedied through better neckband design.
There were many advantages to a dual output device. It opened up more opportunity when using computer programs. For example, while using scan typing programs, one signal was used to start scanning and one to select. While using environmental controls, one signal could be used to turn on a device, and another to turn off that device. Another advantage was that the long duration signal may be used as a filter, effectively filtering out times when the user would like to talk to their communication partner, laugh or cry. This would allow the user to not have to take off the device before talking to their partner. Varying the duration may also open up more possibilities for multi-output devices.
As a dual output device, the sensitivity and specificity would need to improve for the device to be practical. As the dual output was only tested for a short period of time with each participant, the lower accuracy of the device may be due to lack of practice. In addition during the sessions, some words which were used to produce the long duration signal were not as good as others. For example the word ‘Alexander’ would often produce a pause in between the word, causing the device to send a short duration signal. Some short words, such as ‘go’ would produce a long signal when participants would become excited and extend the word. The sensitivity and specificity were often dependant on the words chosen to elicit the signals.
COST
The vocal cord vibration switch cost approximately $306 in material costs and $127 in engineering and labour costs to produce. This cost would be reduced if the product was mass produced.
FUTURE DIRECTION
This study shows that the proposed switch access solution provides a promising new alternative for individuals with severe and multiple disabilities who are able to hum or produce vocalizations. Further development of the device may include designing a better neckband, and making it wireless. The vocal cord vibrations could also be analyzed, just as sound is analyzed for voice recognition systems; this could potentially create multiple outputs that could be used as a multiple output switch.
ACKNOWLEDGEMENTS
Special thanks to those at Holland-Bloorview Kids Rehabilitation Hospital – Gail Teachman, Tom Chau, Tiago Falk, Ka Lun Tam, Pierre Duez, the children who participated as well as their caregivers and therapists.
REFERENCES
[1] G. Lancioni, “Extending microswitch-based programs for people with multiple disabilities: use of words and choice opportunities,” Research in Developmental Disabilities, vol. 24, Apr. 2003, pp. 139-148.
[2] T.H. Falk, J. Chan, P. Duez, G. Teachman, and T. Chau, “Augmentative communication based on realtime vocal cord vibration detection.,” IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society, vol. 18, Apr. 2010, pp. 159-63.
[3] J. Chan, T.H. Falk, G. Teachman, J. Morin-McKee, and T. Chau, “Evaluation of a non-invasive vocal cord vibration switch as an alternative access pathway for an individual with hypotonic cerebral palsy – a case study.,” Disability and rehabilitation. Assistive technology, vol. 5, Jan. 2010, pp. 69-78.
[4] R. Patel, “Phonatory control in adults with cerebral palsy and severe dysarthria,” Augmentative and Alternative Communication, vol. 18, Mar. 2002, pp. 2-10
Author correspondence: Elaine Lu, 54 Longford Crescent, Toronto, ON M1W1P4,