Machine learning paradigms for speech recognition of an Indian dialect

Present era is full of speech recognition based services and products. The machine learning paradigms is at the centre stage of speech recognition methodology. Automatic speech recognition (ASR) technology has vastly evolved in recent years including emerging applications in mobile computing, natural user interface, and man-machine assistive technology. In this paper, it's the first time we are presenting ASR designs based on two important machine learning paradigms Artificial Neural Network (ANN) and Support Vector Machine (SVM) for an rare and geographically important Indian dialect `Chhattisgarhi'. The conventional feed-forward ANN and SVM have been applied on the dataset of maximum 50 isolated words of 15 speakers. The performance of these machine learning paradigms is compared with state of art Hidden Markov Model (HMM). The tendency of ASR to be speaker dependent and independent has been extensively investigated with speaker variation experiments. Furthermore the reliability and stability of ASR has been confirmed with numerical validation. The exhaustive review of ASR techniques from the literature along with the ASR systems designed on Indian languages is presented.