This paper describes a very efficient Parallel Phone Recognizersfollowed by Language Modeling (PPRLM) system in terms of bothperformance and processing speed. The system uses context-independentphone recognizers trained on MLP features concatenated with theconventional PLP and pitch features. MLP features have severalinteresting properties that make them suitable for speech processing,in particular the temporal context provided to the MLP inputs and thediscriminative criterion used to learn the MLP parameters. Results ofpreliminary experiments conducted on the NIST LRE 2005 for theclosed-set task show significant improvements obtained by the proposedsystem compared with a PPRLM system using context-independent phonemodels trained on PLP features. Moreover, the proposed systemperforms as well as a PPRLM system using context-dependent phonemodels, while running 6 times faster.