previous chapterprevious chapterprevious chapterprevious chapter
next chapternext chapternext chapternext chapter

2: Method

Subject, recording and test material

The speaking and singing of MUI, SHET SZE performing as a female character in three Cantonese Opera pieces, represents the basis of the present investigation (recording of live performance on a special DVD edition, see [9]). The procedure to create the test sample of sounds was as follows: (1) Longer parts of the actress’ singing and speaking on high pitches were labeled. (2) Single syllables or isolated sounds of long and medium-long vowels were extracted. The intention of syllables and vowels were taken from the subtitles and were crosschecked acoustically. (3) This first sample was screened by the authors and sounds with generally poor sound quality were excluded. Because no instances of the vowel /ɛ/ with good vowel quality were found, this vowel was also excluded from the investigation. (4) A final test sample of 172 sounds on F0 ≥ 550Hz of the vowels /i, y, œ, a, ɔ, u/, and a training set of 7 phrases on high pitches and 60 syllables on F0 < 550Hz were created. (For /u/, test sounds had to be included in the training set since no sounds with F0 < 550Hz were found).

Acoustic analysis

For each syllable, the sound nucleus of the long vowel was labeled visually using PRAAT [8]. For the segment selected, average F0 was calculated using the PRAAT standard command. Calculations were acoustically crosschecked relating to the C-major scale. If substantial differences were found, average F0 was estimated acoustically.

Listening test

26 university students of linguistics (19 women, 7 men, no history of hearing problems), aged 18 to 23 years and native speakers of Cantonese, participated in the listening test. Before the test, they were informed about the purpose of the test, and they listened to the training set (single sounds in random order) without reacting, as to become familiarized with the test sounds.

For the listening test, the sounds of the syllables were presented in random order. The subjects listened to the sounds on a PC using a commercial headphone and were asked to write down the perceived long or medium long vowel quality into a prepared list. Long and medium-long vowels were not differentiated because they do not differ in vowel quality.