Paper Title
Use of Neural Networks for Speech Recognition

Abstract
Feature based Speech reorganizations is very important factor in the field of Artificial intelligence. In the last decayed HMM-GMM is very popular in the field of speech recognition, but this system having some limitations like in GMM, for calculating the probabilities it needs assumptions close to the data distribution, difficult to handle high dimensional acoustic feature & data fragmentation. To overcome this limitation researches move towards DNN. A neural network made-up of simple computational elements which operates in parallel, the network function is depending on the connection between the elements. We train the neural network accordingly so we get the specific target for the particular input. This paper work shows the comparative results like world level, letter level, sentence level and number of layers between speech recognition system using acoustic model of HMM and Neural Networks. In proposed system we are using Marathi language data base in male and female voice also the features used are the MFCC, Fundamental frequency & Voiced & Unvoiced part of data. Keywords - Deep neural Network (DNN), GMM (Gaussian Mixture Model), MFCC (Mel Frequency Cepstral Coefficients), HMM (Hidden Markov Model), ANN (Artificial Neural Network), WER (Word Error Rate).