Quarterly Progress and Status Report

Transcription

1 Dept. for Speech, Music and Hearing Quarterly Progress and Status Report Effects of lung volume on the glottal voice source and the vertical laryngeal position in male professional opera singers Thomasson, M. journal: TMH-QPSR volume: 45 number: 1 year: 2003 pages:

2

3 TMH-QPSR, KTH, Vol. 45, 2003 Effects of lung volume on the glottal voice source and the vertical laryngeal position in male professional opera singers Monica Thomasson Abstract Lung volume (LV) has been shown to have an effect on the glottal voice source and vertical laryngeal position (VLP) in normal, vocally untrained subjects. If this would apply also to operatic singing, an audible change of vocal timbre would occur as the LV changes during a phrase. This investigation examines LV effects on the voice source and the VLP in 9 professional operatic singers. The subjects sang a sequence of /pae:/-syllables at different loudness levels and pitches using their full vital capacity range. Subglottal pressure, VLP and glottal voice source characteristics were analysed at high and low LV. With regard to VLP and peakto-peak flow, the results were similar to those found for untrained subjects, the former rising and the latter decreasing with decreasing LV. With regard to subglottal pressure, closed quotient, glottal leakage, and H1-H2 level difference the results differed from those found for untrained subjects in that these parameters did not change with decreasing LV. Introduction In normal ventilation there is an intricate interdependence between the laryngeal and the respiratory systems. This is evident e.g. from the movements of the vocal folds during quiet breathing, where activation of intrinsic laryngeal muscles precedes that of the respiratory system (Fink & Demarest, 1978; Brancatisano & al., 1983a, 1983b; Kuna & al., 1988, 1990, 1991). The breathing activities during phonation are superimposed on the vital function of the respiratory system. An interesting question therefore is whether or not a similar interdependence between the respiratory and laryngeal systems exists also during phonation. Lung volume (LV) has been found to affect certain aspects of phonation in normal subjects (Iwarsson & al., 1996; Iwarsson & Sundberg, 1998; Iwarsson & al., 1998; Milstein, 1999). Iwarsson and co-workers, studied the effects of LV on the vertical laryngeal position (VLP) and the glottal voice source in both male and female vocally untrained normal subjects singing a repeated /pae:/-syllable sequence through the full vital capacity (VC) range at 3 different pitches and at 3 different loudness levels (Iwarsson & Sundberg, 1998; Iwarsson & al., 1998). They found that LVs near 90% VC were associated with a lower VLP, a higher subglottal pressure (P s ), a greater peak-to-peak (PtP) flow, a smaller closed quotient (Qclosed), and a larger glottal leakage than low LV (20% VC). Milstein s study of female normal speakers phonating at a speech related F0 corroborated most of these findings (Milstein, 1999). By simultaneously using a flexible endoscope he further observed a more dilated laryngeal configuration at high LV (80% VC) than at mid LV (40% VC) and a greater laryngeal constriction at low LV (20% VC) than at mid LV. There may be several causes of these effects of LV on phonation. The biomechanical tracheal pull is a strong candidate. When the diaphragm descends during deep inspiration, the carina bifurcation may move downward more than 20 mm (Maklin, 1925). According to Zenker and collaborators this implies that the constant background force created by the tracheal pull will vary with LV (Zenker & Glaninger, 1959). Zenker further suggests that the tracheal pull is associated with an abductive component tending to decrease glottal adduction (Zenker, 1964). The effects of the tracheal pull on phonation may be enhanced or counteracted by neurological and/or behavioural factors (e.g Davis & al., 1993; Davis & al., 1996; Vilkmann & al., 1996; Shipp & al., 1985). Statistical data on breathing, gathered from untrained female speakers (Winkworth & al., 1994, 1995,) and from professional operatic singers (Thomasson & Sundberg, 1997; Watson Speech, Music and Hearing, KTH, Stockholm, Sweden TMH-QPSR, KTH, Vol. 45: 1-9,

4 Thomasson M: Effects of lung volume on the glottal voice source... & Hixon, 1985; Watson & al., 1990) show that, on average, singers use more air in their sung breath groups than untrained speakers do in normal speech. Singers accommodate this greater air consumption by starting phrases at higher LVs rather than by ending phrases at LVs substantially below the resting expiratory level (REL); the mode values of singers initiation LV lies in the 70-80% VC bin, while the termination LV is in the 30-40% VC bin for male singers and in the 20-30% VC bin for female singers (Thomasson & Sundberg, 1997). Singer s preference for higher initiation LVs was supported in a study of a singer learning a novel aria (Watson & Hixon, 1996); as the singer got more acquainted with the aria, the initiation LV was raised considerably. These large LV ranges are associated with clear effects on the voice source in untrained subjects. It seems unlikely that such effects are acceptable within the western operatic singing tradition, as audible changes of voice characteristics would occur automatically in normal phrases. This would limit the singers control over both vocally and musically important variables needed for the purpose of musical expressivity. This study replicates the previous studies of vocally untrained voices made by Iwarsson and co-workers (Iwarsson & Sundberg, 1998; Iwarsson & al., 1998), the aim being to experimentally test if LV affects the glottal voice source and the VLP also in trained singers. The results should elucidate an aspect of voice training important to singing pedagogy. Experiment Subjects Nine male opera singers - age range 29 to 45 years - highly trained in the western classical tradition (3 tenors, 3 baritones, and 3 basses) participated in the study. They were all professional in the sense that at the time of the experiment they all earned their livelihood from singing either on national or on international opera stages. Their professional careers extended between 2 and 20 years. The subjects were not informed of the purpose of the study. Task The subjects task was to sing the syllable /pae:/ repeatedly, starting at a maximum LV and continuing until they ran out of air. This syllable sequence was performed at 3 different pitches, a middle comfortable pitch and the pitches 7 semitones above and below this pitch. At each pitch the sequence was sung at 3 different loudness levels (mezzo-forte, piano, and forte). For each of these 9 conditions the sequence was sung twice. Thus, each singer sang a total of 2x9 different sequences. The subjects were standing in an unrestrained position, however asked to avoid body movements. No particular instructions were given regarding breathing behaviour. In the same recording session, the subjects also repeated the same tasks with two other inhalatory strategies, expansion and contraction of the abdominal wall. These data will be analysed in a future investigation. Experimental set-up The experimental set-up is shown in Figure 1. Breathing data was recorded by means of respiratory inductive plethysmography (Respitrace, Ambulatory Monitoring, Inc., Ardsley, NY). The Respitrace system was turned on at least 1.5 hours before the experiment, so as to avoid effects of warming. The elastic transducers of the Respitrace system, henceforth respibands, reflect changes in cross-sectional area. They were placed around the rib cage (RC) and the abdominal wall (AW), with the upper edge at the level of the axilla and the level of the umbilicus, respectively. To avoid slippage, an elastic retainer was worn over the respibands. In order to calibrate the RC and AW signals such that their sum reflects relative LV, the subjects performed a series of iso-volume manoeuvres, i.e., shifting the AW inwards and outwards with occluded airways. This was carried out at a high, a middle and a low LV. For determination of vital capacity the subjects performed maximum inhalations and exhalations. REL was captured in terms of series of relaxed sighs. All these manoeuvres were recorded both before and after the singing tasks. The audio signal was picked up by a microphone attached to a headset, keeping the distance between mouth and microphone constant at 30 cm. VLP was recorded by means of a Glottal Enterprises electroglottograph (EG2) equipped with a pair of double electrodes. Glottal airflow was captured by means of a Rothenberg flow mask (Glottal Enterprises). Flow was calibrated by means of airflow obtained from a pressure tank and measured by means of a Rotameter (VEB Prüfgeräte-Werk, Medingen, type TG06). Ps, low pass filtered at 100 Hz, was measured as the oral pressure 2

5 TMH-QPSR, KTH, Vol. 45, 2003 Audio U P s VLP Digital data recorder Respitrace RC AW Figure 1. Experimental set-up: The audio signal was picked up by means of a microphone attached to a headset. Airflow (flow) was captured by means of a Rothenberg flow mask. Subglottal pressure (Ps) was captured by means of a plastic tube placed in the flow mask. Vertical laryngeal position (VLP) was recorded by means of electroglottography. Rib cage (RC and abdominal wall (AW) signals were registered by means of Respitrace. All signals were recorded on the multi-channel digital instrumentation data recorder. during the occlusion for the consonant [p] and captured by means of a plastic tube, inner diameter 0.4 cm, that the subject held in the corner of the mouth. Pressure was calibrated by means of a manometer. All signals, including the calibrations, were recorded on a multichannel digital instrumentation data recorder (RD-200T PCM, Teac Corp., Japan). Audio A B U Ps VLP RC AW LV Figure 2. Example of the various signals recorded of a /pae:/-syllables sequence sung at middle pitch in mezzo-forte; audio, airflow (U), subglottal pressure (Ps), vertical laryngeal position (VLP), rib cage (RC), abdominal wall (AW) and lung volume (LV). Of the two repetitions (A and B) the second (B) was selected for analysis. Speech, Music and Hearing, KTH, Stockholm, Sweden TMH-QPSR, KTH, Vol. 45: 1-9,

6 Thomasson M: Effects of lung volume on the glottal voice source... Analysis All signals recorded from the two repetitions of each experimental condition were digitised on separate channels in data files using the Soundswell Signal Workstation program (Hitech Development AB) (Figure 2). As a rule, the second of these repetitions was selected for analysis. Respiratory data LV, AW and RC data are shown as the three lowest curves in Figure 2. In order for the sum of the RC and AW signals to represent relative LV, an amplification factor must be applied to one of these signals. This factor was determined by plotting against each other the amplitudes of the AW and RC signals recorded during the isovolume manoeuvres. The correlation and the best linear fit of these data were then determined. For each subject, the iso-volume manoeuvre that yielded the highest correlation was selected for computing the amplification factor; mostly this correlation was found at middle LV. The factor thus obtained was then applied to the AW signal and thereafter the corrected summed signal was calculated. A custom-made program executed this entire procedure automatically on pre-selected isovolume recordings. LV range was determined within each syllable sequence. This range, henceforth referred to as LVr, was defined as the LV interval measured at phonation start and after phonation end. The second /pae:/-syllable in the sequence was chosen to represent high LV. On average this LV corresponded to 87% of LVr (inter-subject variation range 79-93%). The /pae:/-syllable located at 20% LVr was chosen to represent low LV, thus avoiding extremely low LVs that may be associated with non-typical vocal behaviour (Shipp & al., 1985) and including the LV range typically used by professional operatic singers. Voice data Ps was captured from the oral pressure during the [p]-occlusion preceding the vowel sound chosen for analysis. The glottal airflow signal, low pass filtered at 2000 Hz, was inverse filtered using a custom-made program. A ripple-free closed phase was used as a criterion for tuning the first two formant filters. The flow glottogram measurements were taken from the middle part of the vowel section, where traces of articulatory movements were absent. The flow glottogram characteristics measured in this study are seen in Figure 3: (1) period time (T), (2) closed phase time (Tcl), (3) flow peak amplitude (Û), (4) mean flow during the open phase, (5) and the mean DC flow during the closed phase reflecting glottal leakage. Further, mean AC flow during the flow pulse was calculated by subtracting mean DC flow (glottal leakage) from the mean flow during the open phase. The Qclosed was calculated as the Tcl divided by T. Of the glottogram measures, Qclosed and glottal leakage were used in the statistical analysis. DC flow T Tcl Figure 3. Illustration of the flow glottogram characteristics analysed: Period time (T), closed period time (Tcl), peak flow amplitude (Û), and DC flow. Five additional flow glottogram measures were calculated: (1) PtP flow computed as the difference between Û and glottal leakage. (2) H1-H2, the level difference between the first two partials of the voice source measured by means of spectrographic analysis of the flow glottogram. (3) Glottal compliance, computed as the ratio between mean AC flow during the flow pulse to Ps. (4) Permittance computed as the ratio PtP flow over Ps. (5) Estimated glottal area, the ratio PtP-flow over the square root of Ps. The voice data were gathered for all LV, pitch, and loudness conditions. VLP data VLP was analysed in an arbitrary unit. As movement of the larynx during the selected syllable was not uncommon, the average of the middle 0.5s of the vowel section was chosen to represent the VLP for that syllable. These data were normalised with respect to the individual subject s total variability range observed during all phonations. Henceforth, this VLP measure Û 4

7 TMH-QPSR, KTH, Vol. 45, 2003 will be referred to as normalised VLP (VLP N ). Thus, 0% and 100% VLP N correspond to the lowest and the highest VLP values observed in that subject. VLP N data were gathered for all pitch, loudness and LV conditions. Statistical analysis Data were subjected to a 2 x 3 x 3 withinsubjects ANOVA with LV (high, low), pitch (low, mid, high), and loudness (piano, mezzoforte, forte) as within-subject variables and VLP N, Ps, Qclosed, glottal leakage, PtP-flow, H1-H2, compliance, permittance, and estimated glottal area as dependent measures. Results The effects of LV on the different variables are illustrated in Table 1. The statistical analysis revealed a significant result for VLP N, such that high LV was associated with a lower larynx than low LV. VLP N was also affected by loudness (F 2,16 =6.249, p=.010) the forte level being associated with a lower larynx than the mezzo-forte and piano levels. The analysis further revealed a significant interaction effect between pitch and loudness (F 4,32 =4.679, p=.004). For high and low pitch, the larynx tended to descend with increasing loudness, while at middle pitch the larynx was lower in piano and forte than in mezzo-forte. Possibly, larynx position is less critical in middle pitch and loudness conditions. LV did not show any effect on the singers Ps. This means that the singers, unlike vocally untrained subjects, efficiently compensated for the changes in the passive recoil forces in the breathing apparatus associated with LV changes (Iwarsson & al., 1998). As expected, high pitches were produced with higher Ps than low pitches (F 2,16 =53.105, p=.000) and increased vocal loudness was associated with increased Ps (F 2,16 =72.975, p=.000). An interaction effect pitch x loudness was found (F 4,32 =7.276, p=.000), such that high pitch was more affected by loudness than low pitch. The singers did not display any significant change in Qclosed between high and low LV. They did show a significant effect of pitch, such that the Qclosed decreased with increasing pitch, the greatest change occurring between mid and high pitch (F 2,16 =8.223, p=.003). Qclosed was also significantly effected by loudness, such that the quotient was prolonged with increasing loudness, the greatest effect being between piano and mezzo-forte (F 2,16 =28.853, p=.000). There was also an interaction effect between LV and loudness, such that the effect of loudness on Qclosed was greater at low than at high LV (F 2,16 =3.944, p=.041), a clear effect occurring only at middle loudness. LV did not affect the singers glottal leakage. As expected, glottal leakage decreased with increased loudness (F 2,16 =6.796, p=.007). The PtP flow showed a significant effect of LV, the singers producing a higher PtP flow at high than at low LV. As expected, it also increased with increasing loudness (F 2,16 =61.605, p=.000). An interaction effect between LV and pitch indicated that LV had the greatest effect at low pitch (F 2,16 =5.084, p=.020). The H1-H2 level difference was not affected Table 1. Result of analysis of variance (2 x 3 x 3 within-subjects ANOVA) of the effect of lung volume (LV) on analysed voice parameters, and averages across subjects, pitches and loudness levels. Statistically significant results are marked with *. Average Parameter High LV Low LV df F p VLP N (% of individual range) (1,8) * Subglottal pressure (cm H 2 0) (1,8) Closed quotient (1,8) Glottal leakage (l/s) (1,8) Peak-to-Peak flow (l/s) (1,8) * H1-H2 level difference (db) (1,8) Compliance (arbitrary unit) (1,8) * Permittance (arbitrary unit) (1,8) * Est. glottal area (arbitrary unit) (1,8) * Speech, Music and Hearing, KTH, Stockholm, Sweden TMH-QPSR, KTH, Vol. 45: 1-9,

8 Thomasson M: Effects of lung volume on the glottal voice source... by LV in the singer subjects. The effect of pitch revealed that while H1-H2 was similar at low and mid pitch, it increased at high pitch (F 2,16 =5.915, p=.012). As expected, it decreased with increasing loudness (F 2,16 =26.843, p=.000). An interaction effect between pitch and loudness revealed that the effect of loudness was greater at high than at middle and low pitch (F 4,32 =3.082, p=.030). Glottal compliance and glottal permittance showed similar results. For both, high LV was associated with higher values than low LV. An increase in pitch was associated with a decrease in both, (F 2,16 = , p=.000; and F 2,16 =26.052, p=.000, respectively) and an increase in loudness was associated with a decrease (F 2,16 =35.708, p=.000; and F 2,16 =20.421, p=.000, respectively). Also an interaction pitch x loudness was observed in both, the effects of loudness being greatest at low pitch (F 4,32 =8.483, p=.000 and F 4,32 =2.763, p=.044). Estimated glottal area, i.e., PtP flow amplitude over the square-root of P s, revealed a significant effect of LV, such that high LV was associated with higher values than low LV. An increase in pitch was associated with a decrease of the estimated glottal area (F 2,16 =8.680, p=.003), while an increase in loudness was associated with an increase (F 2,16 =6.841, p=.007). An interaction LV x pitch was observed, the effect of pitch being greater on high LV (F 2,16 =5.390, p=.016). Discussion In the experiment, the subjects were unrestrained, since it could be assumed that an invasive condition might disturb their typical singing behaviour. As a consequence, however, the Respitrace signals sometimes changed during the experiment as evidenced by the moderate differences between LV calibrations. We circumvented this complication by defining the LV range for each syllable sequence and selecting the second syllable in each sequence as representing the high LV condition. Furthermore we selected the syllable produced at 20% of this LV range as representing the low LV condition. The relationship between these relative LVs and the absolute LV seemed of limited relevance to our study. Nevertheless, it seems reasonable to assume that these highly experienced subjects reached their extreme LVs during the breath groups. The VLP data were normalised in a similar way, although for obvious reasons, the reference range was not the specific breath group, but the subject s individual variation across the entire experiment. For this reason, the VLP values are not necessarily comparable between subjects. On the other hand, the statistical analysis concerned only within subject effects. The finding that the larynx had a significantly lower position at high LV than at low LV is in agreement with other investigations of normal subjects (Zenker, 1964; Hoit & al., 1993; Iwarsson & al., 1998; Milstein, 1999,) and of singers (Sundberg & al., 1989; Shipp & al., 1985; Pabst & Sundberg, 1993). VLP is affected by tongue position, jaw movement, tracheal pull, and the strap muscles, i.e., the infra- and suprahyoid muscle groups (e.g. Vilkmann & al., 1996). The first two of these factors do not seem relevant to the present findings. No change of tongue position was needed during the sung sequence, as the vowel remained the same. Jaw opening would cause a local lowering of the larynx in the middle of the vowel segment, and this was not observed. The tracheal pull and/or strap muscle seem more relevant. Milstein studied normal female subjects VLP at high, middle, and low LV (80%, 40% and 20% VC, respectively). He found a greater difference between high and middle LV than between middle and low LV. He regarded the passive tracheal pull as a more likely agent than strap muscle action, since there seems to be no logical reason why the strap muscles would consistently pull the larynx downwards with large inspirations, and release this pull gradually as LVs gets smaller. (Milstein, 1999, p.141). As male singers in the classical western tradition habitually sing with a VLP lower than in resting (Shipp & al., 1985), it seems reasonable to assume activity in the strap muscles lowering the larynx. A change in tracheal pull cannot be avoided when moving from a high to a low LV, but this effect can be compensated for by a gradual increase of strap muscle action. In our singers, VLP rose with decreasing LV, thus implying that they did not fully compensate for the decrease of the tracheal pull. It is interesting that we found an effect of LV on VLP in our professional singers that was similar to that found for non-singer subjects. On the other hand, our VLP data cannot be quantitatively compared with those studies and the relationship between the VLP data and resting position is not known. It seems likely, 6

9 TMH-QPSR, KTH, Vol. 45, 2003 however, that our singers VLP was lower than those of nonsinger subjects (Shipp & al., 1985). Also it is possible that the change in VLP is smaller in singers than in nonsingers (Neuschaefer-Rube & al., 1996), since a changed VLP alters the vocal tract and hence will change the formant frequencies and alter the voice timbre. The mean change in VLP between high and low LV amounted to about 11% of the individual singer s total VLP variation during the entire experiment. Assuming that this individual range of VLP variation equalled the largest possible change in VLP that the EGG device can measure, i.e., 4 cm, a worst case scenario would mean that the mean change of vocal tract length would be around 0.4 cm. For a vocal tract length of 17.5 cm, this would imply an average formant frequency increase of about 2.5%, which is barely audible. In other words, it is likely that the changes in VLP with LV did not cause audible shifts in formant frequencies. The Ps was not affected by a change in LV in our singers. In vocally untrained subjects, on the other hand, Iwarsson and collaborators found a higher Ps at high LV than at low LV. Milstein, in his investigation of female normal subjects, observed a decrease of Ps between middle and low LV. In both studies, it was concluded that this was caused by failure to compensate for the increasingly negative passive recoil pressures in the breathing apparatus when reaching low LVs. The discrepancy between singers and untrained subjects is not surprising, since Ps is the main physiological control factor for vocal loudness, a variable that singers need to control accurately for musical purposes. In other words, the independence of Ps on LV would be an important goal of vocal pedagogy. An interaction effect was observed; the difference in Ps between piano, mezzo-forte, and forte being greater at high than at low pitch. As shown by Titze (1992) loudness variation is proportional to the normalised excess pressure rather than to Ps itself. This measure of Ps is related to the phonation threshold pressure, the lowest pressure that produces vocal fold vibration, which increases with fundamental frequency (F0). Thus, an increase of Ps by, say, 5 cm H 2 O produces a greater increase of sound pressure level (SPL) at low than at high pitch. This would be the cause of this interaction effect. Unlike the vocally untrained subjects, the professional singers did not show an effect of LV on Qclosed (Iwarsson & al., 1998). The Qclosed normally increases with an increase in Ps. As the singers Ps remained basically constant regardless of LV, the lack of effect of LV on Qclosed is not surprising. In untrained subjects, it was found that while Ps was higher at high LV than at low, the Qclosed was smaller. This clearly indicated the presence of an abductive force component that was greater at high than at low LV in untrained subjects. Qclosed was clearly higher in our singers than was found in the study of the untrained voices; means across conditions for singers were 0.43 and 0.44 at high LV and low LV and for the untrained voices 0.31 and The fact that the singers used clearly higher Ps than the untrained subjects may have contributed to their higher Qclosed. The glottal leakage was not affected by LV in these professional singers. In vocally untrained subjects, the glottal leakage was found to be greater on high LV (0.17 l/s) than low LV (0.12 l/s) a result corroborated also by Milstein (1999). Milstein used a flexible endoscope simultaneously in his study of normal female voice, and the visual inspection of the larynx gave that the size of a posterior glottal gap decreased or disappeared completely with decreasing LV. It is not likely that a posterior glottal gap would be present at all in professional operatic singers, and this could account for the difference between the singers and the vocally untrained subjects. The suggestion is supported by the fact that the singers mean glottal leakage at both high and low LV is similar to that of the untrained at low LV. The mean PtP flow across subjects and conditions decreased from 0.57 l/s at high LV to 0.50 l/s at low LV. This decrease can be compared with that observed for untrained subjects, from 0.39 l/s to 0.30 l/s. The singers greater flow rates may reflect their preference for flow phonation. The lower PtP at low LV may induce a marginal decrease of sound level. The H1-H2 level difference was not affected by LV in our singer subjects. For untrained voices, a trend to a lower H1-H2 at low LV was found that however failed to reach significance. Our main finding was that LV affects certain aspects of phonation also in highly trained singers, such that high LV was associated with lower VLP and higher PtP flow than low LV. However, these changes were so small that their acoustic effects would hardly be audible. Furthermore, while Ps, Qclosed, glottal leakage Speech, Music and Hearing, KTH, Stockholm, Sweden TMH-QPSR, KTH, Vol. 45: 1-9,

10 Thomasson M: Effects of lung volume on the glottal voice source... and H1-H2 changed with LV in untrained voices they did not do so in our singers. This is not surprising, since these parameters have a direct influence on the resulting voice quality. It seems unlikely that singers would allow LV to affect audible aspects of phonation, as this would limit their range of phonatory variation likely to be needed for the purpose of musical expressivity. The fact that Ps, Qclosed and glottal leakage were LV dependent in vocally untrained subjects, as opposed to professional singers, further emphasises the effects of training. Main emphasis would be placed on eliminating audible effects of LV. The results also indicate that it is important to train singing throughout the whole VC range, as vocal parameters are likely to change with a change in LV in beginners, as the LV range used in singing is clearly larger than during speech. We analysed the effect of varied LV on three different measurements that could be assumed to reflect glottal adduction, compliance, permittance, and estimated glottal area. LV was found to have a significant effect on all of these measures. All of these measures reflect the ratio between a measure related to the PtP flow and a measure related to Ps. LV showed an effect on the peak-to-peak flow but none on Ps. Even though the estimated glottal area yielded the highest level of significance, our results do not provide any reasons for regarding any of these measures as more revealing than the others. Conclusions Singers and untrained voices seem to differ with regard to the effect of LV on voice source characteristics. Thus, while in untrained voices Qclosed, glottal leakage and estimated glottal area all changed significantly between high and low LV, these parameters remained essentially constant over a large LV range in classically trained professional singers. In addition, while Ps tends to decrease with decreasing LV in untrained voices, it seems basically independent of LV in the singers. On the other hand, PtP flow amplitude and VLP seems to change in a similar way for both untrained voices and singers, although the effects seem to be smaller in the singers. The findings support the assumption that singers learn to compensate for perceptually salient effects of decreasing LV. Acknowledgements Many thanks to our singer subjects for their kind participation in this investigation, to Joakim Westerlund for providing the statistical analysis, to Svante Granqvist for creating new software and assisting in the analysis whenever needed. Last but not least, I am deeply indebted to Johan Sundberg, my supervisor, for all his advise and assistance throughout the entire process of this investigation. References Brancatisano T, Collet PW & Engel LA (1983a). Respiratory movements of the vocal cords. J Appl Physiol, 54/5: Brancatisano T, Dodd D & Engel LA (1983b). Factors influencing glottic dimensions during forced breathing. J Appl Physiol, 55/6: Davis PJ, Bartlett D & Luschei ES (1993). Coordination of the respiratory and laryngeal systems in breathing and vocalization. In: Titze IR, ed. Vocal Fold Physiology/Frontiers in basic science. San Diego: Singular Publishing Group, Inc., Davies PJ, Zhang SP, Winkworth A & Bandler R (1996). Neural control of vocalization: Respiratory and emotional influences. J Voice, 10/1: Fink BR & Demarest RJ (1978). Laryngeal Biomechanics. Cambridge, MA: Harvard Univ Press. Hoit JD, Solomon NP & Hixon TJ (1993). Effect of lung volume on voice onset time (VOT). J Speech Hear Res, 36: Iwarsson J & Sundberg J (1998). Effects of lung volume on vertical larynx position during phonation. J Voice, 12/2: Iwarsson J, Thomasson M & Sundberg J (1996). Lung volume and phonation: a methodological study, Log Phon Vocol, 21: Iwarsson J, Thomasson M & Sundberg J (1998). Effects of lung volume on the glottal voice source, J Voice, 12/4: Kuna ST, Insalaco G & Woodson GE (1988). Thyroarytenoid muscle activity during wakefulness and sleep in normal adults. J Appl Physiol, 65/3: Kuna ST, Insalaco G & Villeponteaux RD (1991). Arytenoid muscle activity in normal adult humans during wakefulness and sleep. J Appl Physiol, 70/4: Kuna ST, Smickley JS & Insalaco G (1990). Posterior cricoarytenoid muscle activity during wakefulness and sleep in normal adults. J Appl Physiol, 68/4: Macklin C (1925). X-ray studies on bronchial movements. Am J Anatomy, 35: Milstein C (1999). Laryngeal function associated with changes in lung volume during voice and speech production in normal speaking women. Dissertation. University of Arizona. 8

11 TMH-QPSR, KTH, Vol. 45, 2003 Neuschaefer-Rube C, Wein B, Angerstein W & Klajman S (1996). MRI examination of laryngeal height during vowel singing, Folia Phoniatr Logop, 48/4: Pabst F & Sundberg J (1993). Tracking multichannel electroglottograph measurement of larynx height in singers. Log Phon Vocol, 18: Shipp T, Morissey P & Haglund S (1985). Laryngeal muscle adjustment for sustained phonation at lung volume extremes In: Askenfelt A, Felicetti S, Jansson E, Sundberg J, eds. Proceedings of the Stockholm Music Acoustics Conference 1983 (SMAC 83). Stockholm: Royal Swedish Academy of Music, 46/1: Sundberg J, Leandersson R & von Euler C (1989). Activity relationship between diaphragm and cricothyroid muscles. J Voice, 3/3: Titze IR (1992). Phonation threshold pressure: A missing link in glottal aerodynamics J Acoust Soc Am, 91: Thomasson M & Sundberg J (1997). Lung volume levels in professional classical singing. Log Phon Vocol, 22: Vilkman E, Sonninen A, Hurme P & Körkkö P (1996). External laryngeal frame function in voice production revisited: A review. J Voice, 10: Watson PJ & Hixon TJ (1985). Respiratory kinematics in classical (opera) singers J Speech Hear Res, 28: Watson PJ & Hixon TJ (1996). Respiratory behavior during the learning of a novel aria by a highly trained classical singer In: Davis PJ & Fletcher NH, eds. Vocal Fold Physiology/Controlling Complexity and Chaos. San Diego: Singular Publishing Group, Inc, Watson PJ, Hixon TJ, Stathopoulos ET & Sullivan DR (1990). Respiratory kinematics in female classical singers. J Voice, 4/2: Winkworth AL, Davis PJ, Adams RD & Ellis E (1995). Breathing patterns during spontaneous speech. J Speech Hear Res, 38: Winkworth AL, Davis PJ, Ellis E & Adams RD (1994). Variability and consistency in speech breathing during reading: Lung volumes, speech intensity, and linguistic factors. J Speech Hear Res, 37: Zenker W (1964). Questions regarding the function of external laryngeal muscles. In: Brewer D, ed. Research potentials in voice physiology. Syracuse, NY: State university of NY: Zenker W & Glaninger J (1959). Die Stärke des Trachealzuges beim lebenden Menschen und seine Bedeutung für die Kehlkopfmechanik. Ztschr Biol, 111/2: Speech, Music and Hearing, KTH, Stockholm, Sweden TMH-QPSR, KTH, Vol. 45: 1-9,