3: Results

Results of the listening test

For 160 of 172 sounds tested, the identification rate is ≥ 50% according to the intended vowel of the singer. Moreover, an upper limit of F0 of possible vowel identification > 80% was found at 692Hz for /y/, 765Hz for /œ/, 823Hz for /a/, 845 for /o/, 855Hz for /u/, and 859 for /i/ (see Figure 1). Remarkably, sounds of /u/ and /i/ prove to be intelligible on F0 > 800Hz comparable to sounds of /a/, although F1 of /i/ and /u/ are given as the lowest values and F1 of /a/ is given as the highest value in formant statistics [6, 7].

>> Link to Figure 1

Additional documentation

As mentioned in the title page, for each of the sounds investigated, this online presentation shows additional information, numerical values and graphic illustrations:

• Indication of the intended syllable or isolated vowel sound by the singer, and intended long or medium long vowel
  
• Analyzed sound segment and duration (center of vowel sound in terms of vowel nucleus)
• Spectrum and LPC curves (center of vowel sound; frequency range 0 - 5.5 kHz)
  
• Pitch (center of vowel sound)
• Spectrogram and formant patterns (center of vowel sound)
• Entire sound of syllable or isolated vowel sound and its duration
  
• Spectrum (center of vowel sound; extended frequency range 0 - 11 kHz)

Investigated sounds and vowels (overview), confusion matrices

For each of the sounds and vowels investigated, Table 1 indicates the duration of the entire sounds, the duration of the center of the vowel sound (long or medium long vowel), the results of the listening test in terms of confusion matrices, F0 (average, maximum and minimum of the center of the vowel sound), and the archive number of the recording.

>> Link to Table 1

Please note

The links below refer to sound series and provide additional numerical values and graphic illustrations. When selecting a link, the corresponding series are displayed in a separate tab. Please refer to the assistant in the sound archive for further instructions.

In the sound archive, for technical reasons, the vowel qualities are not indicated in IPA standard. The correspondence of IPA and the notation in the sound archive is as follows:
i = i; y = y, œ = oe; a = a; ɔ = openo; u = u.

In the "Mini" and "M" layouts of the sound archive, a short legend is displayed below a spectrum.
Indications given in the first line: Singer (achive number 1300), gender (w for women), age group (A for adults), relation (relation to the longer sequence of recording), sound number in the archive.
Indications given in the second line: Vowel, syllable or isolated vowel sound, F0 (average of the analyzed sound fragment of the vowel nucleus), language (yue = Cantonese), phonation (v for voiced), context (c = (C)V or (V)V:S context, i = vowel sound produced in isolation), vocal effort (nor = normal).

Sound series for single vowels

For the sounds investigated, the following series show numerical values and graphic illustrations for the sound waves, spectra, LPC filter curves, F0, spectrograms, and formant patterns. Formant patterns are calculated using the PRAAT command To LPC (burg). Two patterns are given, the first with the PRAAT parameter "Max. number of formants = 5" (standard for women), the second with the PRAAT parameter "Max. number of formants = 4" (parameter for shorter vocal tract than for woman's standard).

Because of high F0, formant pattern estimation is not methodically substantiated. However, the results of formant analysis are documented here in terms of information and illustration, since for sounds of one vowel but very different F0, the question of the relationship between the LPC filter curves of these sounds is of importance.

>> Link to Series 1, vowel /i/, (see Figure 1, first graphic)

>> Link to Series 2, vowel /y/, (see Figure 1, second graphic)

>> Link to Series 3, vowel /œ/, (see Figure 1, third graphic)

>> Link to Series 4, vowel /a/, (see Figure 1, fourth graphic)

>> Link to Series 5, vowel /ɔ/, (see Figure 1, fifth graphic)

>> Link to Series 6, vowel /u/, (see Figure 1, sixth graphic)

Vowel-specific spectral differences for sounds at F0 of 680–860Hz

Series 7 shows all sounds with an average F0 of the vowel segment in the frequency range of c. 680–860 Hz and an identification rate ≥ 80%. In the series, the sounds are ordered according to the vowel quality intended and the F0 level.

The series illustrates that perceptual vowel discrimination on high F0 corresponds to related vowel-specific spectral differences in terms of differences in the amplitude configuration of the harmonic spectrum.

>> Link to Series 7, sounds on F0 c. 680–860Hz