Robust Automatic Speech Recognition System: Hmm Versus Sparse

Speech Recognition has been an ever growing and challenging area for the researchers as well as the industry. It is defined in the computer domain as the ability to ability of computer systems to accept spoken words in audio format - such as. wav or raw and perform tasks accordingly. Despite the wide diffusion of commercial applications most of the research works are done in either English, Arabic or Mandarin and the technology underlying is known to only a few laboratories[1]. Thus the development of such a system is still on the primitive stage towards the local Indian languages. This paper discusses regarding an attempt to develop a digit recognition system for Malayalam language using the HMM Toolkit (HTK). Application of Sparse Imputation to recognition algorithm is discussed so as to increase the robustness rather than the conventional Hidden Markov Model technique.