Switched- basis LSP quantization

Line Spectral Pair information is routinely

generated by speech coders, and transmitted to

convey vocal tract information for

reconstruction of speech. Quantization of these

parameters is necessary to reduce bitrate, but

causes a degradation in quality. Many

quantization schemes quantize a differential

value - either using a temporal or a frequency

basis. Although neither scheme is shown to

have significant performance advantages for

LSP quantization through results presented

here, both relate differently to speech content.

Therefore this paper proposes an adaptive

quantizer switching method using both bases,

and assesses the advantages of such a switching

scheme based on a number of common distance

measures.