I have been trying to run the Aurora4 corpus (subset of WSJ0) in the HTK as MFCC_Z_0_D_A features, but having problem generating the transcription .mlf files. Since only the sentence/word utterances are given as .ptx files, I am generating the word-level transcriptions first, and then generate the phone-level transcriptions using the 'timitdict' dictionary, but getting many errors that words are not found in the dictionary. Is there a simple way to get around this - or can somebody help with references to UNIX scripts for generating the Aurora4 transcriptions? Thank you, Serajul

--- (Edited on 5/2/2013 3:21 am [GMT-0500] by serajul) ---