3.1.1.The subjects
Ten volunteer subjects (seven men and three women) took part in the experiment.
Approval for the use of human subjects was given by the ethics committee of the University of New South Wales, where all experiments were conducted. One subject (S6) was a professional singer employed by a national opera company, another (S9) was an amateur soprano. The remaining subjects were untrained in speaking or singing.
3.1.2.Measuring acoustic resonances of the vocal tract
Acoustic impedance measurements were made following the microphone three-calibration method of Dickens et al. (2007). A cylindrical aluminum measurement head was constructed with internal diameter 26.2 mm and microphones spaced at 20, 60, and 260 mm from the downstream measurement end. A broadband signal synthesised from sine waves between 10 Hz and 4.2 kHz (Smith, 1995) was injected via an enclosed 2” dome midrange loudspeaker (Jaycar CM-2092, Australia) at the upstream end. The signals from the three microphones (B&K 4944A, Brüel & Kjær, Nærum Denmark) were connected via a conditioning preamplifier (Nexus 2690, Brüel & Kjær, Nærum, Denmark) to a Firewire audio interface (MOTU 828, Cambridge, MA) and sampled at 44.1 kHz with 16 bit
5 This work was performed under the supervision of John Smith and Joe Wolfe at UNSW. A study based on data from a single subject was published as ‘Low frequency response of the vocal tract: acoustic and mechanical resonances and their losses,’ Proceedings of the Australian Acoustical Society Conference, Fremantle, Australia, November 2012 (Hanna et al., 2012a). Further preliminary portions of this work were presented in ‘Resonances and bandwidths in the vocal tract and why they are important for speech comprehension,’ Australian Institute of Physics Congress, Sydney, Australia, (Hanna et al., 2012c); ‘Acoustic measurements of vocal tract resonances,’ International Conference on Voice Physiology and Biomechanics
resolution. The upper frequency limit of the measurements is determined by the smallest spacing between any two of the three microphones, which here causes a singularity at 4.3 kHz. The lower frequency limit of around 10 Hz was determined by the frequency response of the loudspeaker and the microphones.
The impedance head was calibrated using three non-resonant acoustic loads: a quasi-infinite impedance (a large rigid mass), a purely resistive impedance (an acoustically quasi-infinite pipe), and a large flange (Dickens et al., 2007). Measurements from the microphone array allow determination of the acoustic impedance Z at a reference plane, located in these experiments at the end of the cylindrical impedance head just inside each subject’s lips.
With this measurement technique, the geometry on the loudspeaker side of the microphones does not affect the measurement. Here, for the measurements on subjects, a narrow tube (5 cm long and 7.8 mm diameter) filled with acoustic fibre was connected in parallel with the loudspeaker, which permitted the DC airflow necessary for phonation. A schematic of the apparatus is shown in Figure 3-1.
Figure 3-1 Schematic diagram (not to scale) showing how the vocal tract impedance is measured during phonation.
3.1.3.Experimental protocol
The subjects were asked to find a comfortable position for the impedance head in their mouth and to ensure an airtight seal with their lips around the outer diameter of 31.8 mm.
For phonation, the subjects were asked to sustain the vowel in the word ‘heard’, which corresponds approximately to the vowel [әː] in the International Phonetic Alphabet, for the duration of the broadband signal. Subjects were given time to practice this unusual configuration, and some subjects chose to pinch their nose to avoid nasalization. At least
three measurements were made for each subject during each of the two gestures: one during a comfortable low pitched modal (mechanism 1) phonation comparable with their speaking voice, with the velum closed, and the other one while miming the vowel with the glottis and velum closed.
Since the carrier word ends in “rd”, movement of the tongue would be expected to occur during its pronunciation. This effect was controlled in two ways: firstly the subjects were asked to focus on producing the vowel sound and not the entire word. Secondly, the initial and final cycles of the injected broadband signal were discarded prior to analysis to remove any transient parts of the measurement, including those due to the loudspeaker.
During each gesture, 8 or 16 cycles of the broadband signal (plus an initial and final cycle that were discarded) were injected through the measurement head into the subject’s mouth.
The measured impedance for each cycle was displayed in quasi real-time so that changes in the geometry of the vocal tract during measurements, particularly a clear change in the impedance spectrum corresponding to the opening of the velum, could be detected and the measurement repeated if necessary. Later, the approximate location of the maxima and minima of the impedance spectra were identified with a 6th order Savitzky–Golay smoothing filter (Savitzky and Golay, 1964). Parabolas were fitted to both sides of each of the extrema and their crossover point used to determine the frequency f and the impedance magnitude |Z|. B was determined as the difference between the frequencies at half maximum power.
3.1.4.Adjustments for low frequency and during phonation
Measurements with the glottis closed were initially made using a single broadband spectrum from 14 to 4200 Hz. However, the low frequency components in the initial broadband signal excite mechanical resonances in the tissues surrounding the vocal tract.
These resonances sometimes disturbed the subjects and caused them to alter their tract geometry during a measurement, or to produce an uncontrolled vibrato; in both cases this interfered with the measured impedance spectrum. For this reason the frequency range of interest was divided into three different broadband frequency ranges.
To determine the acoustic resonances, both with glottis closed and during phonation, a broadband signal of 200-4200 or 300-4200 Hz with a resolution of 2.69 Hz (44.1 kHz/214)
When studying the mechanical resonances, two narrow, low frequency windows were used:
10-50 Hz with a frequency resolution of 0.34 Hz (44.1 kHz/217), and 14-300 or 14-400 Hz with a resolution of 0.67 Hz (44.1 kHz/216). During these measurements four of the subjects also had a small magnet attached to their cheek and/or neck with adhesive to monitor the movement by inducing a voltage in a coil of wire placed on the axis a fixed distance away.