Speech recognition using ANN and predator-influenced civilized swarm optimization algorithm

Abstract

This paper proposes a hybrid optimization technique, predator-influenced civilized swarm optimization, by integrating civilized swarm optimization (CSO) and predator-prey optimization (PPO) techniques. CSO is the integration of the attributes of particle swarm optimization and a society civilization algorithm (SCA). In the SCA, the swarm is divided into a few societies, and each society has its own society leader (SL); other individuals of the society are termed society members. The combination of all such societies forms a civilization, and the best-performing SL becomes the civilization leader (CL). In CSO, SLs and members update their positions through the guidance of their own CL and SLs, respectively, along with their best positions. In the proposed technique, the PPO technique is integrated with CSO, in which a predator particle is included in the swarm. The predator always tries to chase the CL in a controlled manner, which maintains diversity in the population and avoids local optimum solutions. The proposed optimization technique is applied to optimize the weights and biases of an artificial neural network (ANN) trained for speech recognition. Two databases have been used; one is a TI-46 isolated word database in clean and noisy conditions, and the other is a self-recorded Hindi numeral database. To evaluate the performance of the proposed technique, 2 performance criteria, correlation coefficient and mean square error, are applied. The results obtained by an ANN with the proposed technique outperform the results obtained by an ANN trained with particle swarm optimization, PPO, CSO, and backpropagation techniques in terms of correlation coefficient and mean square error.

A Normalization Methods for Backpropagation: A Comparative Study

“…The most successful applications of NN are classification such as in (B. Dębska, 2011;Cal, 1995;Diane M. Miller, 1995;Markku Siermala, 2008;Sajad JASHFAR, 2013;Tamer Ölmez, 2003). In addition, NN is widely used in pattern recognition such as in (Birendra Biswal, 2013;Bishop, 1995;Jiří Grim, 2008;Seref SAGIR OGLU, 2000;Teena MITTAL, 2016). Normalizing data such as scaling data between (0, 1) may enhance the accuracy and the performance of the mining algorithms including NNs.…”