Formant strategies of professional female singers at high fundamental frequencies Andrea Deme Research Institute for Linguistics, Hungarian Academy of Sciences, Hungary Eötvös Loránd University, Hungary deme.andrea@nytud.mta.hu Abstract When the soprano raises the fundamental frequency above the first formant of a vowel, a remarkable loss of acoustic energy and linguistic information occurs along with an abrupt change in the voice timbre. To avoid these effects, sopranos are assumed to tune their first formant to the raised funda- mental frequency. The support for this claim is mostly based on formant data provided by indirect measurement methods and articulatory data, since direct acoustic data becomes more difficult (or even impossible) to obtain as the fundamen- tal frequency gets higher. In the present study a new combina- tion of measurement methods is introduced. The aim was to extract formant data of three sopranos in the entire set of the Hungarian vowel inventory in a wide pitch range. The results provide evidence for the technique of tuning the first formant to the raised high fundamental frequency in a substantial amount of data. Keywords: formant tuning in soprano, inverse filtering, EGG, external excitation 1. Introduction In high-pitched singing the fundamental frequency (f0 or h1) often exceeds the typical frequency region of the first formant (F1) of vowels. In these cases unaltered articulation of vowels would result in a remarkable loss of acoustic energy (i.e. sound level) and the loss of an acoustic cue (i.e. F1) which is regarded to be important for defining and identifying vowel qualitiy. Singers of the Western operatic style who are re- quired to sing loudly (without any amplification) supposedly tend to compensate for these losses by changing the articulato- ry configuration of vowels while producing them at higher pitches. This way they can enhance the efficiency of resonance utilization. Several investigations have been conducted to provide data on the techniques singers use at high pitch re- gions to avoid the negative acoustic consequences of raising f0 above the F1 of the vowels in speech (F1 speech ). With respect to articulatory maneuvers Sundberg (1975), and Sundberg and Skoog (1995) concluded that pitch raising is accompanied by gradually increasing jaw opening in sopranos when f0 approaches the region of F1. Based on the traditional interpretation of the interrelation between articulatory move- ments and acoustics, the authors suggest that the articulatory data they obtained reflect the acoustic event of F1 “tuning” in an indirect way: they propose that F1 is tuned to f0 when f0 reaches the region of F1 speech during pitch raising. Sundberg and Skoog (1995), however, also noted that the modification of jaw opening starts only at a higher f0 in the case of the close-mid vowels /e o/, whereas (in some singers) it is not clearly present in the close vowels /i u/. The authors claim that in the case of close vowels singers tend to tune their F1 as well as in all other cases, but this tuning is probably achieved by decreasing the degree of the tongue constriction (and not by modifying the jaw opening) for reasons of articulatory conven- ience. As opposed to Sundberg (1975), and Sundberg and Skoog (1995), a study by Bresch and Narayanan (2010) found that the dependency of F1 on f0 is only validated in the case of the close vowels /i u/. They also conclude that this dependency is probably more singer-dependent than it was suggested before. The authors base their claims on articulatory and formant data obtained from MRI-based vocal tract area func- tion calculations. Direct resonance data, on the other hand, are much more difficult to obtain for high-pitched singing. Due to wide har- monic spacing, the raised f0 causes undersampling of the vocal tract (VT) transfer function in the output sound. Consequently, analyzing the spectrum of the output signal is an inefficient way to determine formant frequencies (see e.g. Deme 2012). Therefore, the investigation of the acoustic characteristics of high-pitched sung vowels requires novel methodology to extract formant data in order to by-pass the problem of low resolution of the VT transfer function. The studies in this field concentrate mostly on the open vowel /a/ at particular pitch ranges. Sundberg (1975) used an external vibrator applied at the larynx for measuring five vowels at four fundamental frequencies below 700 Hz. Hertegård and Gauffin (1993) investigated the production of /a/ at 250, 390 and 750 Hz by inverse filtering the flow signal recorded with a Rothenberg mask. Joliveau et al. (2004) applied external excitation at the mouth for measuring four vowels in a pitch range below 1100 Hz. Garnier et al. (2010) used the same technique for investi- gating the vowel /a/ above 440 Hz. With regard to the cases when f0 > F1 speech the above mentioned studies agreed on the tendency of F1: f0 tuning. However, Joliveau et al. (2004) also noted the lack of this tendency in vowels that use lip rounding above approximately 900 Hz. Some of these studies have also reported that the methods used might have also biased the data to some extent or required some modification of the singing technique intended to be observed: Hertegård & Gauffin (1993) noted that their subject complained that the mask ham- pered her in opening the mouth, while Joliveau et al. (2004) and Garnier et al. (2010) instructed their singer to sing very softly and without vibrato even at high pitches which might have also affected the control of formant tuning strategies. The aim of the present study was to extend on the previous research by investigating the entire vowel set of a particular language in a wide pitch range (covering most of a soprano’s range), thus to discover the formant value changes accompa- nying pitch raising. This aim was achieved by the use of a new combination of some of the previously introduced methods, in a way that the problem of measuring formant frequencies at high f0s is resolved while the influence of the measurement method on the data obtained is minimalized. It was hypothe- sized that F1: f0 tuning occurs in each vowel in the cases of f0 > F1 speech regardless of the degree of closeness or lip rounding. However, it was also suggested that F1: f0 tuning appears in 10 TH ISSP · COLOGNE,5–8MAY 2014 90