Maria-Mirela Petrea, Dan Cristea * Dealingwith Prosody. A Computer-Assisted Language Learning Approach

The aim of a work session with PROSODICS is to get a series of comparison windows between a master's utterance and consecutive student's utterances that would normally result in a gradual enhancement of pronunciation.

There are three kinds of comparisons computed:

• in terms of duration: a comparison of the lengths of master's and student's utterances. As a result the student is given a score and is possibly instructed to speak more slowly or more quickly;
• in terms of stress: a comparison of where falls the maximum stress in the two utterances;
• a comparison of the final pitch movements of the two utterances.

Fig. 11. - Results of master-student comparison.

The messages appear in the Prosodics window that also permits a visualization of the prosody and loudness contours of the two utterances as well as the text and the alignment markers (see figure 11).

3.10. Prosody annotation (tagging)

Besides its tutoring function PROSODICS could also be used as a laboratory tool for annotating prosody (see [8]) in spoken texts. The elementary annotating units are chosen to rely upon pitch segments. Inside each pitch segment the pitch level, the pitch movements and the corresponding loudness movements are recorded. The text, the silence intervals and time measurements are also put in evidence.

The following tagging conventions were used.

Regarding the pitch level:

W	-	very low	(50 -110 Hz for men, 83 - 140 for women);
L	-	low	(110 - 140 Hz for men, 140 - 170 for women);
M	-	medium	(140 - 180 Hz for men, 170 - 210 for women);
H	-	high	(180 - 220 Hz for men, 210 - 280 for women);
X	-	very high	(220 - 240 Hz for men, 280 - 330 for women);
Z	-	extremely high	(240 - 300 Hz for men, 330 - 400 for women);

Regarding pitch movements:

S	-	level (stationary);
R	-	rising;
/	-	sharp rising;
F	-	falling;
\	-	sharp falling;

Regarding loudness movements:

*	-	local maximum;
#	-	local minimum;
!	-	absolute maximum;
?	-	absolute minimum;
+	-	rising;
-	-	falling.

All the encoding is done on representative segments of pitch (see figure 12), such that for each segment there are recorded the pitch level of the starting point, the pitch movement, the associated loudness movement, and, eventually, the pitch level of the ending point (in case the pitch segment is followed by a pitch interruption).

The above example will receive the code LR+HF*M with the meaning: low level start (L), followed by a pitch raising (R) during which the loudness rises too (+), followed by a high level pitch (H), followed by a pitch falling (F) during which the loudness has a significant local maximum (*). The segment ends in a medium level pitch (M).

160