Performance analysis of ASR Model for Santhali language on Kaldi and Matlab Toolkit
Arvind Kumar, Rampravesh Kumar, Kamlesh Kishore · 2020
Speech recognition is the ability of devices to respond to spoken commands. In today's times when we are moving towards an automated world, the area of speech recognition has caught the eye of the researchers. The developments in this area are making waves all around us. Our proposed work presents a method to design a robust digit recognition system in Santhali language using Kaldi toolkit and MATLAB for small vocabulary dataset for varieties of features. Santhali is the most widely spoken local dialect in the state of Jharkhand. Using Kaldi, we trained our system with two training methods; monophone training and triphone training. The triphone method proves to be more efficient than the monophone method because of context mapping. In MATLAB, we obtained 95% accuracy for MFCC+LPC feature extraction applied to a GMM model.