A comparison of different spectral analysis models for speech recognition using neural networks

This work presents an application of Artificial Neural Networks (ANN) in Speech Recognition. The aim of this article is to compare the performance of a neural network-based recognition system when using different spectral analysis models. Different sets of coefficients, such as AutoCorrelation and MelCepstron, are extracted from the speech utterances. We have made experiments using separately, different sets of coefficients as the Neural Network inputs. The development of a hybrid system, combining two different sets of coefficients, has also been performed. The results indicate that the hybrid approach outperforms the other models.