You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Dec 9, 2025. It is now read-only.
Repository navigation
This repository was archived by the owner on Dec 9, 2025. It is now read-only.
LTS will add stress 0 to unstressed syllables if a word consists of more than one syllable, for this result, I already removed all the 0 from the LTS result. It would be 475/964 words in different if I don't remove the zero.
There are a few cases when there is no vowel in LTS phonemes: (mostly related to single 'e')
noun :n-aw1-n
n-n
red :r-eh1-d
r-d
touch :t-ah1-ch
t-ch
led :l-eh1-d
l-d
through :th-r-uw1
th-r
yes :y-eh1-s
y-z
great :g-r-ey1-t
g-r-t
say :s-ey1
s
sure :sh-uh1-r
sh-r
be :b-iy1
b
bed :b-eh1-d
b-d
she :sh-iy1
sh
he :hh-iy1
hh
break :b-r-ey1-k
b-r-k
me :m-iy1
m
we :w-iy1
w
a :ey1
ey
above :ah b-ah1-v
ah b-ow1-v
again :ah g-eh1-n
ae1 g-eh1-n
against :ah g-eh1-n-s-t
ae1 g-eh1-n-s-t
ago :ah g-ow1
aa1 g-ow
allow :ah l-aw1
ae1 l-ow
also :ao1-l s-ow
ae1 l-s-ow
always :ao1-l w-ey-z
ae-l w-ey1-z
am :ae1-m
ah-m
among :ah m-ah1-ng
ah m-ao1-ng
an :ae1-n
ah-n
anger :ae1-ng g-er
ae1 ng-er
are :aa1-r
aa-r
area :eh1 r-iy ah
aa r-iy1-ah
arrange :er ey1-n-jh
er-ey1-n-jh
arrive :er ay1-v
er-ay1-v
as :ae1-z
ah-z
atom :ae1 t-ah-m
ae1 t-aa1-m
be :b-iy1
b
beauty :b-y-uw1 t-iy
b-t-iy
bed :b-eh1-d
b-d
before :b-ih f-ao1-r
b-ah f-ao1-r
begin :b-ih g-ih1-n
b-eh1 jh-ih1-n
behind :b-ih hh-ay1-n-d
b-eh1 hh-ay1-n-d
believe :b-ih l-iy1-v
b-ah l-iy1-v
between :b-ih t-w-iy1-n
b-eh1 t-w-iy1-n
body :b-aa1 d-iy
b-ow1 d-iy
break :b-r-ey1-k
b-r-k
busy :b-ih1 z-iy
b-y-ah s-iy
but :b-ah1-t
b-ah-t
by :b-ay1
b-iy
can :k-ae1-n
k-ah-n
carry :k-ae1 r-iy
k-eh1 r-iy
century :s-eh1-n ch-er iy
s-eh1-n t-er-iy
certain :s-er1 t-ah-n
s-er1 t-ey1-n
character :k-eh1 r-ih-k t-er
ch-eh1 r-ah-k t-er
chick :ch-ih1-k
ch-ih-k
chief :ch-iy1-f
ch-ah-f
choose :ch-uw1-z
ch-uw1-s
city :s-ih1 t-iy
s-ay1 t-iy
clear :k-l-ih1-r
k-l-ih1-er
climb :k-l-ay1-m
k-l-ih1-m
close :k-l-ow1-s
k-l-ow1-z
column :k-aa1 l-ah-m
k-aa1 l-ah-m-n
company :k-ah1-m p-ah n-iy
k-ah-m p-ah n-iy
compare :k-ah-m p-eh1-r
k-ah-m p-aa1-r
consonant :k-aa1-n s-ah n-ah-n-t
k-ah-n s-ah n-ah-n-t
continent :k-aa1-n t-ah n-ah-n-t
k-ah-n t-ah n-ah-n-t
control :k-ah-n t-r-ow1-l
k-aa1-n t-r-aa-l
correct :k-er eh1-k-t
k-ao1-er-eh1-k-t
cotton :k-aa1 t-ah-n
k-aa1-ah-n
could :k-uh1-d
k-ow1-l-d
country :k-ah1-n t-r-iy
k-aw1-ah1-n t-r-iy
cow :k-aw1
k-ow
create :k-r-iy ey1-t
k-r-iy-ey1-t
cry :k-r-ay1
k-r-iy
current :k-er1 ah-n-t
k-er1-ah-n-t
dad :d-ae1-d
d-ae-d
danger :d-ey1-n jh-er
d-ey1 ng-er
decide :d-ih s-ay1-d
d-ah s-ay1-d
depend :d-ih p-eh1-n-d
d-ah p-eh1-n-d
describe :d-ih s-k-r-ay1-b
d-ah-s k-r-ay1-b
desert :d-eh1 z-er-t
d-ah z-er-t
design :d-ih z-ay1-n
d-ah z-ay1-n
dictionary :d-ih1-k sh-ah n-eh r-iy
d-ih1-k sh-ah n-eh1 r-iy
die :d-ay1
d-iy1
difficult :d-ih1 f-ah k-ah-l-t
d-ih1 f-ah k-ah1-l-t
direct :d-er eh1-k-t
d-er-eh1-k-t
discuss :d-ih s-k-ah1-s
d-ih1-s k-ah1-s
distant :d-ih1 s-t-ah-n-t
d-ih1-s t-ah-n-t
do :d-uw1
d-ow
dog :d-ao1-g
d-aa1-g
dress :d-r-eh1-s
d-r-eh1-s-s
dry :d-r-ay1
d-r-iy
effect :ih f-eh1-k-t
ah f-eh1-k-t
eight :ey1- t
ey1-t
electric :ih l-eh1-k t-r-ih-k
ah l-eh1-k t-r-ih-k
element :eh1 l-ah m-ah-n-t
eh1 l-eh1 m-ah-n-t
engine :eh1-n jh-ah-n
eh1-n jh-iy1-n
enough :ih-n ah1-f
ah n-aw1
equate :ih k-w-ey1-t
iy1 k-w-ah-t
especially :ah s-p-eh1-sh l-iy
eh1-s p-eh1 sh-ah l-iy
even :iy1 v-ih-n
iy1 v-ah-n
evening :iy1-v n-ih-ng
iy1 v-ah n-ih-ng
event :ih v-eh1-n-t
iy1 v-eh1-n-t
every :eh1 v-er iy
eh1 v-er-iy
exercise :eh1-k s-er s-ay-z
eh1-g z-er s-ay1-s
expect :ih-k s-p-eh1-k-t
ih-k-s p-eh1-k-t
experience :ih-k s-p-ih1 r-iy ah-n-s
ih-k-s p-ih1 r-iy-ah-n-s
experiment :ih-k s-p-eh1 r-ah m-ah-n-t
ih-k-s p-eh1 r-ah m-ah-n-t
family :f-ae1 m-ah l-iy
f-ah m-ah l-iy
famous :f-ey1 m-ah-s
f-ah m-ah-s
few :f-y-uw1
f-uw
finger :f-ih1-ng g-er
f-ih1 ng-er
fire :f-ay1 er
f-ay1-er
floor :f-l-ao1-r
f-l-uh1-r
flower :f-l-aw1 er
f-l-aw1-er
four :f-ao1-r
f-aw1-r
from :f-r-ah1-m
f-r-aa1-m
fun :f-ah1-n
f-ah-n
garden :g-aa1-r d-ah-n
g-aa1 r-d-ah-n
gas :g-ae1-s
g-ah-z
general :jh-eh1 n-er ah-l
jh-eh1 n-er-ah-l
get :g-eh1-t
g-ah-t
give :g-ih1-v
g-ay1-v
go :g-ow1
g-ow
great :g-r-ey1-t
g-r-t
half :hh-ae1-f
hh-ao1-f
has :hh-ae1-z
hh-ah-z
have :hh-ae1-v
hh-ey1-v
he :hh-iy1
hh
her :hh-er
hh-er1
his :hh-ih1-z
hh-ih1-s
history :hh-ih1 s-t-er iy
hh-ih1-s t-ao1 r-iy
hour :aw1 er
hh-aw1-r
how :hh-aw1
hh-ow1
hurry :hh-er1 iy
hh-er1-iy
ice :ay1-s
ah-s
idea :ay d-iy1 ah
ay d-iy-ah
imagine :ih m-ae1 jh-ah-n
ih1 m-ae1 jh-iy1-n
in :ih-n
ih1-n
include :ih-n k-l-uw1-d
ih1-n k-l-uw1-d
indicate :ih1-n d-ah k-ey-t
ih1-n d-ah k-ey1-t
industry :ih1-n d-ah-s t-r-iy
ih1-n d-ah1-s t-r-iy
insect :ih1-n s-eh-k-t
ih1-n s-eh1-k-t
instant :ih1-n s-t-ah-n-t
ih1-n-s t-ah-n-t
instrument :ih1-n s-t-r-ah m-ah-n-t
ih1-n-s t-r-ah m-ah-n-t
invent :ih-n v-eh1-n-t
ih1-n v-eh1-n-t
iron :ay1 er-n
ay1-er-aa1-n
island :ay1 l-ah-n-d
ih1 l-ae1-n-d
key :k-iy1
k-iy
know :n-ow1
n-aw1
laugh :l-ae1-f
l-ae1
led :l-eh1-d
l-d
leg :l-eh1-g
l-ah-g
less :l-eh1-s
l-eh1-s-s
let :l-eh1-t
l-ah-t
lie :l-ay1
l-iy1
liquid :l-ih1 k-w-ah-d
l-ih1 k-w-ih1-d
listen :l-ih1 s-ah-n
l-ih1-s t-ah-n
log :l-ao1-g
l-aa1-g
lost :l-ao1-s-t
l-aa1-s-t
low :l-ow1
l-ow
machine :m-ah sh-iy1-n
m-ih sh-iy1-n
major :m-ey1 jh-er
m-ah jh-er
man :m-ae1-n
m-ah-n
many :m-eh1 n-iy
m-ae1 n-iy
market :m-aa1-r k-ah-t
m-aa1 r-k-ah-t
master :m-ae1 s-t-er
m-ae1-s t-er
material :m-ah t-ih1 r-iy ah-l
m-ah t-ih1 r-iy-ah-l
may :m-ey1
m-ey
me :m-iy1
m
method :m-eh1 th-ah-d
m-eh1 th-aa1-d
million :m-ih1 l-y-ah-n
m-ih1-l y-ah-n
molecule :m-aa1 l-ah k-y-uw-l
m-ah l-ah k-y-uw1-l
month :m-ah1-n-th
m-aa1-n-th
most :m-ow1-s-t
m-aa1-s-t
much :m-ah1-ch
m-ah-ch
multiply :m-ah1-l t-ah p-l-ay
m-ah1 l-t-ah p-l-iy
my :m-ay1
m-iy
nation :n-ey1 sh-ah-n
n-ae1 sh-ah-n
natural :n-ae1 ch-er ah-l
n-ae1 ch-er-ah-l
nature :n-ey1 ch-er
n-ae1 ch-er
necessary :n-eh1 s-ah s-eh r-iy
n-ah s-ah s-eh1 r-iy
new :n-uw1
n-uw
nine :n-ih1-n
n-ay1-n
no :n-ow1
n-ow
noun :n-aw1-n
n-n
now :n-aw1
n-ow
numeral :n-uw1 m-er ah-l
n-uw1 m-er-ah-l
object :aa1-b jh-eh-k-t
ah-b jh-eh1-k-t
occur :ah k-er1
aa1 k-er1
of :ah1-v
ao-f
off :ao1-f
ao-f
offer :ao1 f-er
ah f-er
office :ao1 f-ah-s
ah f-ah-s
often :ao1 f-ah-n
aa1 f-ah-n
oh :ow1
ow-hh
on :aa1-n
ah-n
once :w-ah1-n-s
aa1-n-s
one :w-ah1-n
ow1-n
only :ow1-n l-iy
aa1-n l-iy
operate :aa1 p-er ey-t
aa1 p-er-ey1-t
opposite :aa1 p-ah z-ah-t
aa1 p-ah z-ay1-t
order :ao1-r d-er
ao1 r-d-er
organ :ao1-r g-ah-n
ao1 r-g-ah-n
original :er ih1 jh-ah n-ah-l
er-ih1 jh-ah n-ah-l
our :aw1 er
aw1-er1
own :ow1-n
aw1-n
oxygen :aa1-k s-ah jh-ah-n
aa1-k s-ah g-ah-n
paragraph :p-ae1 r-ah g-r-ae-f
p-eh1 r-ah g-r-ae1-f
parent :p-eh1 r-ah-n-t
p-r-ah-n-t
particular :p-er t-ih1 k-y-ah l-er
p-aa1 r-t-ih1 k-y-ah l-er
party :p-aa1-r t-iy
p-aa1 r-t-iy
pay :p-ey1
p-ey
perhaps :p-er hh-ae1-p-s
p-er1 hh-ae1-p-s
period :p-ih1 r-iy ah-d
p-ih1 r-iy-ah-d
planet :p-l-ae1 n-ah-t
p-l-ae1 n-eh1-t
please :p-l-iy1-z
p-l-iy1-s
plural :p-l-uh1 r-ah-l
p-l-er-ah-l
poem :p-ow1 ah-m
p-ow1-eh1-m
populate :p-aa1 p-y-ah l-ey-t
p-aa1 p-y-ah l-ey1-t
position :p-ah z-ih1 sh-ah-n
p-aa1 z-ih1 sh-ah-n
power :p-aw1 er
p-aw1-er
prepare :p-r-iy p-eh1-r
p-r-iy p-aa1-r
present :p-r-eh1 z-ah-n-t
p-r-eh1 z-eh1-n-t
pretty :p-r-ih1 t-iy
p-r-eh1 t-iy
process :p-r-aa1 s-eh-s
p-r-aa1 s-eh1-s
product :p-r-aa1 d-ah-k-t
p-r-ah d-ah1-k-t
put :p-uh1-t
p-ah1-t
quick :k-w-ih1-k
k-w-ih-k
quiet :k-w-ay1 ah-t
k-w-ay1-ah-t
quotient :k-w-ow1 t-ih ah-n-t
k-w-ow1 t-ah-eh1-n-t
radio :r-ey1 d-iy ow
r-ey1 d-iy-ow
range :r-ey1-n-jh
r-ae1-n-jh
read :r-eh1-d
r-iy1-d
ready :r-eh1 d-iy
r-iy1 d-iy
reason :r-iy1 z-ah-n
r-iy1 s-ah-n
red :r-eh1-d
r-d
remember :r-ih m-eh1-m b-er
r-ah m-eh1-m b-er
repeat :r-ih p-iy1-t
r-ah p-iy1-t
reply :r-ih p-l-ay1
r-iy p-l-iy
represent :r-eh p-r-ah z-eh1-n-t
r-eh1 p-r-ah z-eh1-n-t
require :r-iy k-w-ay1 er
r-eh1 k-w-ay1-r
result :r-ih z-ah1-l-t
r-ah z-ah1-l-t
row :r-ow1
r-ow
said :s-eh1-d
s-ey1-d
say :s-ey1
s
school :s-k-uw1-l
sh-k-uw1-l
science :s-ay1 ah-n-s
s-ay1-ah-n-s
sea :s-iy1
s-iy-ah
second :s-eh1 k-ah-n-d
s-eh1 k-aa1-n-d
separate :s-eh1 p-er ey-t
s-eh1 p-er-ah-t
several :s-eh1 v-r-ah-l
s-eh1 v-er-ah-l
she :sh-iy1
sh
shoe :sh-uw1
sh-ow1
should :sh-uh1-d
sh-ow1-l-d
shoulder :sh-ow1-l d-er
sh-ow1 l-d-er
show :sh-ow1
sh-aw1
sign :s-ay1-n
s-iy1-n
silver :s-ih1-l v-er
s-ih1 l-v-er
sister :s-ih1 s-t-er
s-ih1-s t-er
sky :s-k-ay1
s-k-iy
snow :s-n-ow1
s-n-aw1
so :s-ow1
s-ow
soft :s-aa1-f-t
s-ao1-f-t
soldier :s-ow1-l jh-er
s-ow1 l-d-iy-er
some :s-ah1-m
s-ah-m
son :s-ah1-n
s-ah-n
square :s-k-w-eh1-r
s-k-w-aa1-r
study :s-t-ah1 d-iy
s-t-uw1 d-iy
substance :s-ah1-b s-t-ah-n-s
s-ah1-b-s t-ah-n-s
suffix :s-ah1 f-ih-k-s
s-ah f-ih-k-s
sugar :sh-uh1 g-er
s-uh1 g-er
suggest :s-ah-g jh-eh1-s-t
s-ah-g jh-ah-s-t
sun :s-ah1-n
s-ah-n
supply :s-ah p-l-ay1
s-ah p-l-iy
sure :sh-uh1-r
sh-r
surface :s-er1 f-ah-s
s-er1 f-ey1-s
system :s-ih1 s-t-ah-m
s-ih1-s t-ah-m
ten :t-eh1-n
t-ah-n
than :dh-ae1-n
th-ae1-n
that :dh-ae1-t
th-ae1-t
their :dh-eh1-r
th-eh1-ih1-r
them :dh-eh1-m
th-eh1-m
then :dh-eh1-n
th-ah-n
there :dh-eh1-r
th-ih1-r
these :dh-iy1-z
th-iy1-z
they :dh-ey1
dh-iy
thick :th-ih1-k
th-ih-k
thing :th-ih1-ng
th-ih-ng
those :dh-ow1-z
th-ow1-z
though :dh-ow1
th-ao1
thousand :th-aw1 z-ah-n-d
th-aw1 s-ah-n-d
through :th-r-uw1
th-r
throw :th-r-ow1
th-r-ow
thus :dh-ah1-s
dh-ah-s
tie :t-ay1
t-iy1
tire :t-ay1 er
t-ay1-er
to :t-uw1
t-ow
touch :t-ah1-ch
t-ch
toward :t-ah w-ao1-r-d
t-ow1-er-d
triangle :t-r-ay1 ae-ng g-ah-l
t-r-iy-ae1-ng g-ah-l
try :t-r-ay1
t-r-iy
two :t-uw1
t-w-ow
unit :y-uw1 n-ah-t
y-uw1 n-ih1-t
us :ah1-s
ah-s
use :y-uw1-s
y-uw1-z
value :v-ae1-l y-uw
v-ae1 l-uw1
very :v-eh1 r-iy
v-r-iy
village :v-ih1 l-ah-jh
v-iy l-ih-jh
vowel :v-aw1 ah-l
v-aa1-ah-l
we :w-iy1
w
were :w-er
w-ih1-r
what :w-ah1-t
w-ae1-t
when :w-eh1-n
w-ah-n
which :w-ih1-ch
w-ih-ch
who :hh-uw1
hh-ow
whose :hh-uw1-z
hh-ow1-z
why :w-ay1
w-iy
wire :w-ay1 er
w-ay1-r
with :w-ih1-dh
w-ih1-th
would :w-uh1-d
w-ow1-l-d
yes :y-eh1-s
y-z
your :y-ao1-r
y-aw1-r
The dictionary is more acurate, I don't think we should replace the entry with the result from LTS.
The chance of getting different result is roughly 1/3 for the whole dictionary(28,000);
Easy to fix:
1.Remove 0 from LTS result
2.if there is only one syllable, add stress to the vowel phoneme
What LTS get's wrong:
vowel
stress
syllables
minor consonant case
Some statics of the wrong pronunciation in phoneme: (dict -> LTS,cases > 200, in whole dictionary)
ih -> ah:2561
ah -> aa:363
aa -> ah:342
ah -> ae:326
ih -> eh:305
eh -> ah:293
ah -> ih:292
ae -> ah:283
iy -> ah:246
ah -> eh:209
LTS got many phonemes to ah
ih -> ah:2561
aa -> ah:342
eh -> ah:293
ae -> ah:283
iy -> ah:246
ey -> ah:187
ay -> ah:108
Example Analysis for the case ih -> ah
Case1: ends with ed[ex:abated t-ih-d -> t-ah-d];
Case2: ends with es [acknowledges];
Case3:starts with be [betray];
Case4:starts with re [resemble];
Case5:starts with pre [presume];
Case6:starts with se [seduce];
Case7:contains nist [pianist,administration];
Case8:ends with ness [friendliness]
Case9:ends with ility [imposibility]
...
First try to do some analysis of the result, if this direction seems fine to you, I will keep update more later.
Also, I'm quite surprised to see so many basic prefix or suffix in the cases above...
One thing we should programmatically fix for sure if we can is the LTS problem with missing vowels in single-syllable words
I think we might want to be more specific with the situation we want to improve with the lts result.
Do we want to improve the lts result for words that are contained in the full lexicon but not in lexicon1000? Or do we want to improve the lts result for words in lexicon 1000?
When LTS rules are trained, they are not aimed for pronouncing single-syllable words. Words less than four letters are suggested to be removed from the training set. So it makes sense why LTS rules get so many single-syllable words wrong
For LTS rules, the 'ed' in "yelled" and "red" are treated in the same way, that's why the vowel is missing. Same for "yes" and "she".
One way is to create special word set that includes pronunciation of these short common words, because they are less regular.
These words are already in the full lexicon and the 1k lexicon. So the only case when LTS would be used for them is when LTS works with no Lexicon attached. If we do want to get them right for this scenario, then the pronounciation would be stored as part of the LTS code.
However, we don't have this case according to the size specification we have currently:
rita.js (core-only, no lexicon) -> tiny
rita-small.js (1k-word lexicon + lts) -> medium
rita-full.js (35k-word-lexicon + lts) -> large
You talked about merging RiLexicon to main rita.js, so I'm not sure whether we will face this case for 'tiny' in the future(core, RiLexicon,LTS, no lexicon)
If the LTS and part-of-speech are correct/non-ambiguous, then such words could be removed from the lexicon (though I suppose there could be a performance hit for doing the LTS vs. the fast lexicon lookup)
Two aims: 1) to get the outputs matching; 2) to remove unnecessary entries from lexicons (entries whose pronunciation data matches the LTS engine, and whose POS can be correctly guessed, primarily 'nn')
Just did a quick check in the current full lexicon,
10171 out of 28095 entries (36.2%) got the same pronunciation result from LTS and from the lexicon
There are 2811 words whose POS tag only contains 'nn' and the pronunciation data from LTS engine matches with what we have in the lexicon. These should be the safe words to delete in terms of Pronunciation and POS tags.
Activity
cqx931 commentedon Apr 21, 2017
http://chenqianxun.com/resources/dictLtsDifference.txt
Total: 305 words out of 964 words have results from lexicon and LTS
Issues:
dhowe commentedon Apr 21, 2017
Can we generalize any rules here? Differences that occur every time?
One thing we should programmatically fix for sure if we can is the LTS problem with missing vowels in single-syllable words
Also, we need to remove 0s from all LTS results (in the code)
cqx931 commentedon Apr 21, 2017
The dictionary is more acurate, I don't think we should replace the entry with the result from LTS.
The chance of getting different result is roughly 1/3 for the whole dictionary(28,000);
Easy to fix:
1.Remove 0 from LTS result
2.if there is only one syllable, add stress to the vowel phoneme
What LTS get's wrong:
vowel
stress
syllables
minor consonant case
Some statics of the wrong pronunciation in phoneme: (dict -> LTS,cases > 200, in whole dictionary)
LTS got many phonemes to ah
Example Analysis for the case
ih -> ahCase1: ends with ed[ex:abated t-ih-d -> t-ah-d];
Case2: ends with es [acknowledges];
Case3:starts with be [betray];
Case4:starts with re [resemble];
Case5:starts with pre [presume];
Case6:starts with se [seduce];
Case7:contains nist [pianist,administration];
Case8:ends with ness [friendliness]
Case9:ends with ility [imposibility]
...
First try to do some analysis of the result, if this direction seems fine to you, I will keep update more later.
Also, I'm quite surprised to see so many basic prefix or suffix in the cases above...
dhowe commentedon Apr 21, 2017
Lets do these to start, and push the rest to 1.7
cqx931 commentedon Apr 24, 2017
1.6 part is finished, modified the milestone to Version 1.7
cqx931 commentedon May 16, 2017
I think we might want to be more specific with the situation we want to improve with the lts result.
Do we want to improve the lts result for words that are contained in the full lexicon but not in lexicon1000? Or do we want to improve the lts result for words in lexicon 1000?
When LTS rules are trained, they are not aimed for pronouncing single-syllable words. Words less than four letters are suggested to be removed from the training set. So it makes sense why LTS rules get so many single-syllable words wrong
For LTS rules, the 'ed' in "yelled" and "red" are treated in the same way, that's why the vowel is missing. Same for "yes" and "she".
One way is to create special word set that includes pronunciation of these short common words, because they are less regular.
dhowe commentedon May 16, 2017
you mean add them to the lexicon?
cqx931 commentedon May 17, 2017
These words are already in the full lexicon and the 1k lexicon. So the only case when LTS would be used for them is when LTS works with no Lexicon attached. If we do want to get them right for this scenario, then the pronounciation would be stored as part of the LTS code.
However, we don't have this case according to the size specification we have currently:
You talked about merging RiLexicon to main rita.js, so I'm not sure whether we will face this case for 'tiny' in the future(core, RiLexicon,LTS, no lexicon)
dhowe commentedon May 17, 2017
If the LTS and part-of-speech are correct/non-ambiguous, then such words could be removed from the lexicon (though I suppose there could be a performance hit for doing the LTS vs. the fast lexicon lookup)
cqx931 commentedon May 17, 2017
So the aim of the comparison is to remove pronunciation entries from the lexicon? For both Full lexicon and the 1k lexicon?
dhowe commentedon May 18, 2017
Two aims: 1) to get the outputs matching; 2) to remove unnecessary entries from lexicons (entries whose pronunciation data matches the LTS engine, and whose POS can be correctly guessed, primarily 'nn')
cqx931 commentedon May 18, 2017
Just did a quick check in the current full lexicon,