Nowadays a big research effort is being made in the development of CSR systems, both ANN-based and HMM-based. Up to now, the HMM-based systems seem to have the best performance, although the ANN-based ones are being developed quicker than the HMM-based due to the new topologies that are being tested with increasingly good results.

This paper presents a new framework developed to apply Alphanets to CSR. For this purpose, a modular system is proposed. This system is made up by three different modules: LVQ module, SLHMM module and DP module. The SLHMM module is an expansion of an Alphanet, and therefore, can be interpreted as a HMM. The system can be trained globally applying backpropagation techniques. The used pruning procedure is based upon recognized units instead of observations, which reduces the number of nodes needed to recognize a sentence, compared to HMM-based systems using the same parameters for the models in both systems. Besides, the training procedure re-adapts the weights according to the new architecture in a few iterations since the initial parameters can be estimated from a classical HMM CSR system.<>

The main goal is automatic speech recognition by using artificial neural networks. The authors define a generalized type of neuron that, grouped in a recurrent neural network (an Alphanet), implements a semicontinuous hidden Markov model (SCHMM). The neurons are grouped in a single layer that generates the Alphanet in such a way that some of its inputs come from the outputs. The network allows an interpretation according to SCHMM models, evaluating symbol sequences that constitute the second type of inputs. The network is trained using the backpropagation algorithm and has been applied to an isolated word recognition task. The experimental results show recognition rates reaching multi-speaker recognition rates of 97.81%.<>

A method is presented for obtaining an algebraic representation of boolean functions. The proposed method consists in considering the boolean function as a training set for a pattern recognition system in which the variables arc of the binary type. The orthogonal Rademacher-Walsh polynomials are used as a basis for the probability density functions in a Bayes classifier.

In this paper, a very simple electronic system for the recursive computation of the Walsh Transform for delta-modulated functions is presented. The proposed method avoids the need of demodulating the function.