TexttoSpeech System for Counting in Filipino

1
Text-to-Speech System for Counting in Filipino

• Jocelynn Olida Wong Cu
• De La Salle University - Manila

2
System Design
3
Table 1.0 Filipino diphones used for counting
from isa to isandaan.
4
Performance Analysis

• Mean Opinion Score (MOS) of 30 listeners
• Listening Effort (LE)
• Syllabication (Sy)
• Stress (St)
• Articulation (Ar)
• Speed (Sp)

5
Specific observations
Figure 1, P in apat, apatnapu, and
apatnaput are not pronounced clearly. This
error can be traced back to incorrect end-point
detection and cutting of this particular sound.
Since cutting up of sounds is done manually, the
diphone /at/ tend to surmount the perception of
the phone /p/ before it.
6
Specific observations
Figure 2, the female pronunciation of s in the
word isa is not clear. Comparing it to the
male version of the same word, it can be observed
that the recorded sample for the female speaker
has significantly lower amplitude resulting to
inaudibility of s in the synthesized word.
7
Specific observations
The male utterance of the word isandaan
consistently scored the lowest among all the
synthesized words used in the experiment. As
illustrated in the figure above, it can be seen
that /nd/ is a cc diphone and such combination
can easily cause incorrect determination of sound
boundary.