A New Approach for Coding of Speech Signals using Auto Associative Neural Networks

Digital Speech coding is a procedure to represent a digitized speech signal using as few bits as possible, maintaining the speech quality and its intelligibility at the same time. In this paper a new direction in research on speech coding using auto associative neural networks (AANN) is discussed. The AANN acts as a combination of encoder and decoder. The feature extractor extracts the necessary features from the input speech. Instead of coding the speech signal the Linear Predictive coefficients (LPC) and discrete cosine transform (DCT) features of the speech signal which acts as the compressed value of the speech, is passed to the neural network. The signal reconstructor reconstructs the signal based on the decompressed features and the weight matrix. Different features are extracted and the results are compared. The signal to noise ratio (SNR) shows the efficiency of the algorithm. Some of the applications for which this coder is suitable are videoconferencing, streaming audio, archival, and messaging.