A Connectionist Framework For Feature Based Speech Recognition System Using Artificial Neural Networks

In this paper we study the various methods employed to recognize discrete speech.We design a recognition system which is capable of recognizing spoken language. The software takes spoken language and translates it into written text, or follow the spoken instructions to perform other functions. Here, we propose an unexampled method to recognize speech.We provide a basic connectionist framework to analyze speech wave.The spoken words are digitized(turned into sequence of numbers) and matched against pretrained samples in order to identify the words. The system is trained, requiring samples of actual words that will be spoken by the user of the system. The sample words are digitized, stored in the computer to match against future words. We propose a novel combination of extracting the characterestics of the audio signal using linear predictive coding and a computational approach of using artificial neural networks in indentifying the correct sample. The analog audio is converted into digital signals. This requires analog-to-digital conversion. Linear Predictive Coding is a correlation measure, a measure of similarity between two signals, and is used in the analysis of speech in our implementation. As speech recognition involves the ability to match a voice pattern against a provided or acquired vocabulary, a neural net is constructed to achieve maximum accuracy We show that this method gives salutary results using experimental observations. Then we provide conditions under which the system gives optimum results.