Appeal No. 1996-2198 Application 08/077,506 "dimension" corresponding to each of the twelve parameters 64 (col. 13, lines 20-27). Each dimension is assumed to have the general form shown in Figure 3A, which is a Laplacian probability distribution definable by only two variables, mu and sigma (col. 13, lines 27-35). Baker explains that this simplifies the computation and storage required to represent each dimension (col. 13, lines 35-38). As appellants correctly note (Brief at 32), Baker's Laplacian phonetic frame models are not developed from the analysis parameters used in Markov models, as suggested by the examiner. In fact, the Laplacian phonetic frame models shown in Figures 2, 3A, and 3B are developed without using any HMM analysis. Instead, each Laplacian phonetic frame model simply represents the features of an incoming segment of speech and thus is more akin to the feature information extracted by Kuroda's feature extraction block 4 than to the HMM parameters stored in Kuroda's parameter table 11. We note in passing that although Baker discloses using the Laplacian phonetic frame models in a "smooth frame labeling" technique that employs HMMs (Figs. 6- -19-Page: Previous 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 NextLast modified: November 3, 2007