2.2　Methodology

2.2.1　Test material

All the Ningbo vowels, ten normal-length vowels [i y ʏ e ø ɛ a ɔ o u] that occur in the （C）V position, two short vowels [a o] that occur in the （C）Vʔ position, and two ‘apical’ vowels [ɿ ɥ], were investigated in this study.Meaningful monosyllabic words which contained any one of the vowels were used as the test words （Table 2.1）.All the test words are associated with a high level tone and most of them have a zero initial except for the two apical vowels, which must be preceded by an alveolar consonant, and the dorsal vowel [ø], which must occur with an initial consonant.

Table 2.1：Test words.

2.2.2　Speakers and recording

20 Ningbo native speakers, 10 male and 10 female, provided the speech data.All speakers were born and raised in Ningbo and did not have a history of speech or hearing problems.At the time of recording, the ages of the speakers ranged from 18 to 24, and 14 of them, 9 male and 5 female, were 21 to 22 years old.

The speech data were recorded during my fieldwork in the city of Ningbo in the summer of 2001, using a Sony PCM-R700 digital audio recorder and a Shure SM-58 microphone.Each test word was placed in a carrier sentence [ŋo io doʔ __ paʔ nau thiŋ] ‘I will read __ for you （to） listen’ 我要讀__撥你聽.The 14 test monosyllabic words in the word list were randomized and the speakers were asked to read the word list in a natural way and at a normal speech rate.Five repetitions of the word list were recorded for each speaker.

2.2.3　Data analysis

Spectral data of the test vowels were analyzed using the pitch synchronous linear predictive coding （LPC） method available in the CSL （Computerized Speech Lab） 4400 speech analysis software （Kay Elemetrics）.The speech data were digitized at a sampling rate of 10,000 samples per second, producing an upper frequency cutoff of 5,000 Hz for each vowel spectrum.The LPC spectral analysis was performed at the location of acoustic vowel target.In order to obtain reliable formant values, wide-band spectrograms were also made, serving as a reference for the spectral measurement.In this study, following Lindblom （1963）, the acoustic vowel target is defined as the formant value at a single time point, rather than over an interval of time.In monothongal vowels, the acoustic vowel target typically occurs near the vowel’s temporal midpoint.In practice, the temporal midpoint was chosen for performing the LPC analysis in most cases in this study.However, the formant maxima or minima criterion （Lisker, 1984； Di Benedetto, 1989） was used in certain cases where there was no clearly identifiable steady-state section for the first three formants of the vowel or the steady-state vowel interval differs for each formant.According to this criterion, target vowel values were estimated at the time point where the first formant reaches its maximum value （for the low or mid vowels） or its minimum value （for the high vowels）[1].Figure 2.2 is an example that shows the results of the spectral analysis of the vowel [i].The cursors in the window indicate where LPC spectral analyses were performed for the vowel [i] from Male Speaker 2.

Figure 2.2：Results of the spectral analysis of [i] from Male Speaker 2； speech waveform （Window A）, wideband spectrogram （Window B）, LPC formant trajectories （Window D）, and LPC spectrum （Window C） for the vowel [i].