A Comparative Study Of Auditory-Based Front-Ends For Robust Speech Recognition Using The Aurora 2 Database

In this paper a comparative study of robust speech recognition using three different auditory-based front-end processors is presented. The front-ends investigated are perceptual linear prediction (PLP), the PEMO algorithm proposed by Tchorz and Kollmeier, and the front-end processor proposed by Li et al. Results obtained using the Aurora 2 connected digit database are presented for each of the three algorithms. For comparison purposes the recognition results obtained using the ETSI basic front-end for distributed speech recognition, defined in standard ES 201 108, are used as a baseline. The HTK-based recogniser provided with the Aurora database is used as a classifier for the front-end evaluation. A structural comparison of the main features and commonalities of the different front-end processors is also included