Multilayered network for LPC based speech recognition

This paper presents an efficient artificial neural network based linear predictive coding. Speech recognition is fundamentally a pattern classification task. The objective is to input the speech pattern, and then to classify it as a sequence of patterns. The linear predictive coefficients of the slowly varying speech signals are stored. A feedforward network is determined by the linear predictive coefficients. A three layered feedforward network was used with backpropagation as the training algorithm. The network and learning techniques are proven for their correctness and applied to the problem of speech recognition. The suggested novel scheme yields good results in finite accuracy and recognition performance.