Linguistic mapping in LSF space for low-bit rate coding

Abstract

In this paper we investigate the spectral density of line spectral frequency (LSF) content in languages. The results show that the phonetic variation of languages is reflected in the LSF space. This leads to an alternative approach to the design of LSF quantisers. A trained LSF codebook, like the phonetic inventory of a language, is a static description of spectral behaviour of speech. As clear relationships exist between phonetic segments and LSFs, the structure of an LSF codebook can be analysed in terms of the phonetic segments. The new approach incorporates phonetic information into the structure of LSF codebooks through combining individual phonetic codebooks. The investigation leads to the conclusion that phonetic information can be usefully employed in codebook training in terms of perceptual performance and bit-rate reductions.