| Summary: | 碩士 === 國立暨南國際大學 === 資訊工程學系 === 95 === The word “Cerebral Palsy” is used to describe a medical condition that affects control of the muscles. If someone has cerebral palsy, it means that he is not able to use some of the muscles in their body in the normal way. People, who have cerebral palsy, may not be able to walk, talk, eat or play in the same ways as most other people.
One student who has cerebral palsy is currently studying in our school, and he has problems communicating with others. He cannot use keyboard to type normally, cannot write with the pen, and cannot speak fluently.
Some of the papers analyze the speech recognition with Dysarthric speakers [1][2][8]. They found that the patients, including serious cerebral palsy sufferers, can have acceptable speech recognition results by collecting more voice samples. It gives us an idea: we can develop the speech recognition system for the student as input interface.
In this thesis, we explored the vocal ability of people who have cerebral palsy, looked for the robust vocabularies, and constructed a suitable speech recognition system.
We tried some methods of speech recognition, like Dynamic Time Warping and discrete hidden Markov model (dHMM), taking Mel-Frequency Cepstral Coefficients (MFCC) as features, to find robust vowels and robust consonants that people who have cerebral palsy can pronounce.
Finally, we combine theorems and experimental results to design the speech recognition system that is convenient for the cerebral palsy sufferer. The system encodes frequent Symbols to Morse code. Tester can pronounce based on Morse code table, and combines with Morse code to match the Symbol which he think. We can associate with other assistant to increase the convenience and performance of the system.
|
|---|
