| Autor: |
Ngoc N.D, Vinh D.P, Nam L.H, Hai T.N, Minh N.L, Thanh V.N |
| Rok vydání: |
2018 |
| Předmět: |
|
| Zdroj: |
2018 4th International Conference on Green Technology and Sustainable Development (GTSD). |
| DOI: |
10.1109/gtsd.2018.8595695 |
| Popis: |
Timing is an important characteristic of human speech, which influences its rhythm, stress, duration, and intonation. In this study, we build and analyses a simplified model of a cerebellum-like neural network and investigate its learning mechanism in speech timing. The model is built based on the neural anatomic structure of mammal cerebellum using Matlab Simulink. The model is feed with encoded sound signal as the learning input. The learning mechanism occurs in the neural network for several cycles until one of the connection weight is dropped by more than 80%. The output is the prediction of the timing in the input sound. We found that the cerebellum is capable of learning short timing in human speech of around 400ms to 1500ms, but cannot learn the longer signal of above 2000 ms. For signal between 1500ms to 2000ms sometime the bio-realistic neural network can detect but most of the time it cannot recognized. This is due to the size of the neural network is much smaller than the size of the real cerebellar neural network in mammal |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
