Juny Pina, Ton (2021) Keyword Spotting with the Time Difference Encoder. Master's Thesis / Essay, Physics.
|
Text
Ton_Master_Thesis.pdf Download (9MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (138kB) |
Abstract
In this thesis, the contribution of introducing the Time Difference Encoder to neuromorphic keyword spotting models is researched. Previous studies show that implementing a layer of this neural structure into these models, can improve the performance in the classification task. The reasons for this improvement are discussed by analyzing how the speech is encoded and processed in the biological nervous systems. Information Theory is used in order to research the stimulus encoding in the different parts of the models. In the first set of experiments, a realistic model based on a Python implementation of a biological cochlea is bench-marked, showing successful results by the TDEs in the encoding of temporal patterns from the cochlea representation of the human speech. The second set of experiments tests a model that uses the extracted formants from human speech to classify the spoken words. As in the first experiment, the TDEs show successful results in the encoding of temporal patterns in the formant shape. Finally, a simple binary classifier is designed that performs the keyword spotting from the formants model outputs, showing a solid performance.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Chicca, E. and Noheda, B. |
| Degree programme: | Physics |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 17 Dec 2021 15:25 |
| Last Modified: | 17 Dec 2021 15:25 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/26391 |
Actions (login required)
 |
View Item |