I am done with simple speech recognition now i am trying to do Keyword spotting. I read some papers about it in which they proposed separate modeling for keywords and fillers(remaining speech). Can somebody help me what does it mean by separate models. Is it simply use isolated keywords utterances and exteraneous speech files and then model them or is it means to make two separate hmms for keyword and fillers and then by some why (I am not sure about it) concatenate two models?

Best Regard

--- (Edited on 1/19/2014 9:43 pm [GMT-0600] by khawar163) ---