Source-channel coding for CELP speech coders / J.A. Asenstorfer.
John Anthony Asenstorfer · Adelaide Research & Scholarship (AR&S) (University of Adelaide) · 1994
ùn .fù 4.4 Codebook Generation ll Comparison of Method,s in Missing Data Estimation: 128 entry frxed vector codebook, sorting of the Lsp components, repetition of entire LSP vectors and distortion due to undetected errors' Ratio of Detected.Error Vectors to all Vectors with Bit Errors' Comparison of a Fixed Vector Codebook and an Adaptive Codebook' AverageSpectralDistortionVersusLog2Cod.ebookSizeforTreeand Full search vector Quantisers: /s denotes full search and tree tree search' 101 WeightedNegativeSpectralDistortionVe[susGenerationNumber:öesú denotesthebestofthepopulation,auedenot,esthepopulationaverage.ll0 47 53 53 bÐ 56 56 Ðl 57 4.I 4.2 4.3 82 83 84 85 85 4.4 4.5 5.1 5.2 vl1 5.5 5.3 5.4 5.6 6.1 6.2 6.3 6.4 6.5 6.6 6.8 Average spectral Distortion Versus cod,ebook size for 3 Bits of the Index Error Protected: unpernù denotes the unpgrmuted index case and pernù denotes permutation of the indices' AverageSpectralDistortionVersustheNumberofErrorProtectedBits.116Multi-Stage Vector Quantiser Using Two Stages' I2l Average spectral Distortion versus the Total codebook size: The upper curveisforthe2S6vectorcoalsequantiserandthelowercurveforthe 512 vector coarse quantiser, the points show the performance of the better split-vector quantisers' ll4 r22 r27