When Rommel Fernandes undertook a capstone project as part of his graduate school work in electrical engineering at Loyola Marymount University, he turned to his parents for inspiration. Both of the Seattle native’s parents had suffered strokes, and his father was left with the inability to talk. Fernandes wondered if he could use his electrical engineering skills to help his father communicate and spent a semester devising a non-audible speech classification system.
“Rommel came to me a long time ago about this idea,” said Lei Huang, his faculty advisor and a professor of electrical engineering and computer science. “He said it was heartbreaking when he saw his father trying to say something and he couldn’t get the words out.”
The semester-long project resulted in a program that uses electromyography technology – in this case, sensors placed around the mouth and throat to detect the movements of muscles used during speech. That data is fed into a deep-learning program and classified to look for speech patterns. Fernandes had to submit his proposal to an Institutional Review Board, recruit 10 study participants, gather the data and assess the technology.
The small data set showed the technology could predict speech patterns, said Fernandes, who earned his master’s degree in electrical engineering in May 2019. “With more data, the neural network we used could hopefully identify speech patterns with higher accuracy.”
The project validated the concept of non-audible speech classification, Huang said.
“We wanted to see if this would work or not,” she said. “This is really where the learning is happening for students. Deep learning only happens when students can think through a problem and apply things they learn and make connections among different aspects of knowledge and the other realistic factors.”
Fernandes currently works as an engineering data scientist at Boeing in Seal Beach, working on reliability engineering problems. In December, he presented a paper on his project at the International Conference on Computational Science and Computational Intelligence in Las Vegas.
Fernandes said the project reinforced the reasons he chose LMU for his master’s.
“I was trying to find a smaller school that had a good reputation, small class sizes and professors always willing to help. The location near Silicon Beach was very important in connecting with industry,” he said. “I learned valuable research skills during this project.”