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 Next
Last modified: November 3, 2007