Appeal No. 1996-2198
Application 08/077,506
circuit (15, 15A) which, using code vectors (R) stored in a
code vector table (14, 14A) converts the time sequence of
feature vectors into a time sequence of feature codes (C, Ck),
a process referred to by appellants as "quantization of the
feature codes into feature codes" (Brief at 13). As noted by
appellants, the advantages and disadvantages of this
quantization technique are described as follows in Juang et
al. (Juang) Patent No. 4,783,804:
The recognition scheme disclosed in U.S. patent
application Ser. No. 434,516, filed Sept. 2, 1982
[now Patent No. 4,587,670], discloses an arrangement
that utilizes vector quantization to generate Markov
model output symbol probability signals b(O ). While
t
vector quantization techniques permit the use of
permanently stored probability tables, the resulting
model probabilities are only an approximation to the
actual likelihood. The graph of FIG. 2 illustrates
the effect of the vector quantized approximation.
In FIG. 2, curve 201 represents the actual
likelihood b(O ) as a function of acoustic features,
t
and the vertical lines correspond to the discrete
probability values of the discrete vector quantized
features. An input acoustic feature derived from a
speech pattern is first quantized to the nearest
prototype feature and the probability of the
prototype feature is selected as b(O ). It is
t
readily apparent that an input feature at x on the
1
feature axis corresponds to a probability y from
1
curve 201 but that the probability for the nearest
prototype vector 205 is y . The difference between
2
y and y is the error due to quantization and the1 2
-6-
Page: Previous 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Next
Last modified: November 3, 2007