This item is non-discoverable
Comparison of encoding schemes for spoken words in spiking neural networks
Many recent studies have focused on speech recognition in the human brain and an increasing part of this research uses spiking neural networks (SNNs) for these investigations. SNNs, similar to the human brain, consist of neurons that can communicate with each other by sending out spikes. The LKD model is an example of such an SNN and an adapted version of this model is used as a starting point for this research, which focuses specifically on the way speech inputs are encoded in a spiking neural network and the human brain. This encoding can be done in several ways and here three encoding schemes are compared: the biologically plausible auditory encoding (BAE), cochlear encoding and rate encoding. An accuracy-based comparison shows that the BAE is the most optimal encoding for this SNN. These results are unexpected because the Cochlear encoding is more biologically plausible than the BAE and a sensible expectation would be that the most biologically plausible encoding results in the highest accuracy.
Faculteit der Sociale Wetenschappen