Isolated words recognition using neural networks
Hadi Harb, A.H. Husseiny · 2002
Because anyone would like to speak with his machine, we decided to undertake a speech recognition project. Our objective is to recognize a word to use it as an industrial machine's command, with good accuracy and performance. Until now methods used in speech recognition are analytical or statistical methods. Analytical methods like DTW or Euclidian distance have been used for isolated words recognition, but the performance was not good enough (noise causes problems with these methods). Statistical methods, especially Multi-Layer Perceptron with Hidden Markov Model (MLP+HMM) are commonly used these days, for both continuous and isolated speech, because of their good performance (better than analytical methods). Our method consists of using just neural networks for the recognition of a number of words (commands).