You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
42 lines
1.6 KiB
42 lines
1.6 KiB
From text to speech: The MITalk system
|
|
|
|
[:I Initial position
|
|
30 "/] Final position
|
|
|
|
Percent errors
|
|
|
|
AR RN
|
|
|
|
?
|
|
?
|
|
z
|
|
|
|
Stops Fricatives Nasals Affricates Approximants
|
|
Manner class
|
|
|
|
Figure 13-1: Average percent errors across various manner classes
|
|
|
|
across almost all manner categories, except for the nasals in final position which
|
|
showed an error rate of 27.6 percent. It should also be noted that while consonants
|
|
in initial position were identified better than the same ones in final position, the
|
|
relative distribution of the errors across syllable positions is not comparable, as
|
|
shown in Figure 13-2 below.
|
|
|
|
Figure 13-2 provides a detailed breakdown of the errors and the resulting con-
|
|
fusions for consonants in initial and final positions. Each bar in the figure shows
|
|
the total percent errors for a particular phoneme and the rank order of the most fre-
|
|
quent confusions.
|
|
|
|
In examining these data, it should be kept in mind that the error rates which
|
|
make up the data shown in these two panels are quite low to begin with. The total
|
|
percent errors were only 4.6 percent in initial position and 9.3 percent in final posi-
|
|
tion. Inspection of this figure shows that, for the most part, the errors are
|
|
predominantly confusions in place or manner of articulation. Errors in voicing,
|
|
when they occurred, were substantially lower. The fricatives pE and TH show
|
|
very high error rates when considered individually, although both of these
|
|
phonemes occurred with a relatively low frequency in the test when compared with
|
|
other consonants. The presence of the background masking noise may have con-
|
|
tributed to the low performance levels observed with these weak fricatives. As
|
|
|
|
154
|