Multi-source neural networks based on fixed and multiple resolution analysis for speech recognition

This paper reports the results obtained by an automatic speech recognition system when MFCCs, J-RASTA perceptual linear prediction coefficients (J-Rasta PLP) and energies from a multi-resolution analysis (MRA) tree of filters are used as input features to a hybrid system consisting of a neural network (NN) which provides observation probabilities for a network of hidden Markov models. Furthermore, the paper compares the performance of the system when various combinations of these features are used showing a WER reduction of 20% with respect to the use of J-Rasta PLP coefficients, when J-Rasta PLP coefficients are combined with the energies computed at the output of the leaves of an MRA filter tree. Such a combination is practically feasible due to the use of a NN architecture designed to integrate multiple features, exploiting the NN capability of mixing several input parameters without any assumption about their stochastic independence. Recognition is performed on a very large test set including many speakers uttering proper names from different locations of the Italian public telephone network