Using a Harmony Search Algorithm/Spiking Neural Network Hybrid for Speech Recognition in a Noisy Environment

The Harmony Search Algorithm (HSA) is a recently constructed algorithm that is based on the improvisation process that musicians often demonstrate in performances [1],[2]. In the HSA each musician plays a note in an effort to find the best harmony with his/her fellow musicians. In the context of Computer Science, this improvisation process corresponds to a decision variable (musician) generating a value in order to find a best global optimum solution (a pleasing harmony). The HSA involves three possible choices; repeating a tune or pattern exactly from memory, repeating a tune or pattern with slight variation of one or two variables (adjusting the pitch slightly), or using new random values or notes. Geem [1],[2] formalized these processes as harmony memory, pitch adjusting, and randomization. Recently the HSA has been applied to neural networks. This has been done in two ways. Firstly, by modification of the weights of the neural network by using the HSA as the network is trained, and secondly by using the HSA as a postprocessing tool as a way to reduce the chances of selecting a sub-optimal solution. We propose using the HSA in a new way and on a new generation of neural networks. Spiked Neural Networks (SNN) [3],[4],[5] encode information according to the temporal differences and correlations between spike trains. This new type of neural network is not only biologically more realistic than other neural networks; we propose that SNN have a natural symbiosis with the HSA.