Effects of adaptive postfilters on the LSF quantisation for low bit rate speech coders in tandem connections

This paper examines the effects of postfiltering techniques on the performance of the line spectrum frequency (LSF) parameters for low bit rate speech coders in tandem connections. The speech coder platform consists of a mixed multiband excitation (MMBE) linear predictive coding (LPC) algorithm, that encodes voiced frames at 1.75 kb/s and unvoiced frames at 0.4 kb/s. The analysis is performed for the well known adaptive spectral enhancement (ASE) technique and a recently reported scheme, called spectral envelope restoration combined with noise reduction (SERNR) postfilter, using the same MMBE platform. Spectral distortions and percentages of outliers of a switched predictive LSF vector quantiser, when the speech coder is operating in tandem connections, are presented. These results, along with subjective listening tests, show that the SERNR technique is clearly superior to the ASE postfilter in tandem connection situations.