A COMPARISON OF PLP~RPS AND PLP~IMELDA ACOUSTIC REPRESENTATIONS IN AUTOMATIC SPEECH RECOGNITION
SM RICHARDSON, MJ HUNT · 2024
This paper explores two acoustic representation.lMELDA and PLP-RPS.both of which had given good results in speech recognition tests.llvELDA is examined in the context of some related representations developed at NTI'.Lincoln Laboratory and IBM.Experimental results suggest that the effectiveness of PLP-RPS stems not from its modelling of perceptual properties but from its approximation to a desirable statistical propeny attained exactly by IMELDA.A combined PLP-IMELDA representation is found to be more effective than PLP-RPS.but notclearly better than an IMELDA representation derived directly from a lter-bank.Some preliminary experiments with PLP incorporating dynamic inforrnalion are described.