LSP parameter interpretation for speech classification

This paper describes several methods of interpreting line spectral pair (LSP) parameters such as those generated by CELP and LPC speech coders. Analysis methods are proposed to assist in automatic classification of the phonetic content of an encoded speech signal. In particular, the advantages of analysing the lower bitrate parameters available at a speech coder output rather than the higher bitrate input parameters, are discussed. Comparisons are made with existing parametric measures in terms of both performance and implementation complexity, with a view to maximising classification performance whilst minimising complexity.