Our initial body of utterances was collected
with a program that periodically called staff
members and asked them to say 5
names selected at random from 64 full
Japanese names (surname followed by first name).
Using this program, 684 utterances
were recorded from 47 native Japanese speakers
(3/4 of which were male) and tagged with the
utterance transcription.
The utterances were represented as Bark-scale
power spectra of 20 ms speech frames, Hamming
windowed at 5 ms shifts. The
utterances were time synchronously phoneme
labeled using their transcriptions in an
automated process. The results were
manually checked and adjusted to correct
any missegmentations.
From this data we generated our initial models as
described above and used them to bring the automated
attendant system
online. The system, open to about 100 users,
ran as described in section 3,
and after some months we had collected over 350
additional utterances. The newly collected
utterances were briefly checked and a few
mislabeled ones were deleted.
Even with the new utterances, this is not
a large data set (especially considering that
the task is multi-speaker, and recorded
over telephone lines), but we nonetheless
performed the following experiments to assess
the effects of incremental retraining.
The 350 new utterances were added in 4 stages
(preserving their temporal sequence) to the
initial set of 684 (e.g. 684+87,
684+175, ...). At each stage one third of
all the utterances were selected at random and
held out for testing. The remaining
two thirds became the training data,
from which a new set of models was made using
the three step procedure outlined
above.
At each stage we made 2 tests. The first checked basic
recognition accuracy when new models were generated from the
expanded training data and the new testing data was
incorporated into the test set. The second used the new testing data
but no new training data in order to check how well
the original models generalized to unseen data.
These two tests were
conducted on both the models produced by embedded
k-means clustering (step 2 above) and on the
models after minimum
error training (step 3 above).
Results for these tests are shown in Figure 2.
the principles and the organization o
where. This effort is truly interdiscip
ogy, chemistry and physics to comp
large part of Artificial Life is devote
s we know it - that is life on earth -
earch for principles of living system
rticlular subtrate. Thus, Artificial Lif
exploring artificial alternatives to a
described as attempting to understand
vel rules; for example, how the simp
lead to high-level structure, or the
etween ants and their environment
ior. Understanding this relatinoship
provide novel solutions to complex
invention, stock-market prediction, and
living systems out of non-living part
l the areas of Artificial Life. At prese
two largely independent endeavors:
al building blocks of nature (carbon
sing the same principles but a differ
computer. The former explores the
ng to construct self-replicating mole
populations of self-replicating enti
eristics of different chemistries in su
us, both the biochemical and the co
Life" is used to describe research int
some of the essential properties of
such systems that meet this criterio
nical—and these can be used to per
he principles and the organization o
where. This effort is truly interdiscip
ogy, chemistry and physics to comp
large part of Artificial Life is devote
s we know it - that is, life on earth -
earch for principles of living system
rticular substrate. Thus, Artificial Lif
exploring artificial alternatives to a
described as attempting to underst
vel rules; for example, how the simp
lead to high-level structure, or the
etween ants and their environment
ior. Understanding this relatinoship
provide novel solutions to complex
ention, stock-market prediction, and
living systems out of non-living part
l the areas of Artificial Life. At prese
two largely independent endeavors:
al building blocks of nature (carbon
sing the same principles but a differ
computer. The former explores the
ng to construct self-replicating mole
populations of self-replicating enti
eristics of different chemistries in su
us, both the biochemical and the co
hed light on the compelling question
Not only about the construction and
her artificial or natural; an impressive
rds the construction of adaptive aut
m the classical robotics approach, in
its environment and learns from thi