How To Write Your Own
Application Programs
Programming
Hints
Multiple
Training
Samples
When the
vocabulary
size is less
than 32 words,
significant
improve-
ments in
recognition accuracy are possible
through
the use of
multiple training
samples. By associating
2 or more
speech
table
entries with the same word
the
system can better
handle
the normal
variations of
speech. Coding with multiple
training
samples is
simplified by use
of
a
string array.
For example:
DIM
WD$
(32)
WD$ (1)="ENTER"
WD$ (2)="
ENTER"
would set both indexes 1
and
2
of WD$ equal
to "ENTER."
If
you
trained the
speech recognition system so that
words 1
and 2 of its
tables were both "ENTER" you
would
now have
2 training
samples
of the word
"ENTER." In Recognition,
you
might
get either
(1)
or
(2),
but
both would be interpreted to
mean
"ENTER."
Articulation
People have a tendency to drop some sounds out
of
speech. For
example,
"A" and
"8"
sound
alike if the
"T" sound
is dropped
from
"8".
Be careful to
articulate clearly, or choose
words where
such
confusion
is not
possible.
Words That Sound
Alike
Many
speech sounds are hard to distinguish.
For example,
M and N,
L and R, or E, B, D,
and P. Avoid
words that
only differ
by these
sounds, for
example, MINE and NINE.
Use of String
Arrays
Coding is simplified
if
you set up a
string
array
with
your
vocabulary
words assigned the
same index
as
they
are given in
the
speech tables.
This is especially
true
when
you use
multiple
training
samples.
27
To Next Page
Loading...