Continuous Speech Recognition System for Malayalam Language Using Kaldi

An automatic speech recognition system aims to enable communication between human and computer. Here a continuous Malayalam speech recognition system is developed and realized using Kaldi toolkit. For building the proposed system, MFCC features and its transformations such as LDA and MLLT are extracted from Malayalam speech sentences. Then, the system has been trained to estimate the HMM parameters for speech sound units. Performance of both monophone and triphone models using bi-gram language model is compared in terms of word error rate(WER). A substantial drop in WER is obtained for triphone models.