Dept. for Speech, Music and Hearing Quarterly Progress and Status Report Relationship between changes in voice pitch and loudness Gramming, P. and Sundberg, J. and Ternstrom,¨ S. and Leanderson, R. and Perkins, W. H. journal: STL-QPSR volume: 28 number: 1 year: 1987 pages: 039-055 http://www.speech.kth.se/qpsr STL-QPSR 1/1987 A. RELATIONSHIP BETWEEN CHANGES IN VOICE PITCH AND LOL7llPJESS Patricia Gramming*, Johan Sundberg, Sten Ternstrom, Fblf Leanderson**, and William H. Perkins*** Abstract The change in mean fundamental frequency accompanying changes in loudness of phonation during reading is analyzed in nine professional singers, nine nonsingers and in ten male and ten female patients suffer- ing from vocal fatigue and/or functional dysfunction. The subjects read discursive texts with 19-filtered noise in earphones, and some also at voluntarily varied vocal loudness. Also, the healthy voice subjects phonated as softly and as loudly as possible at various fundamental frequencies throughout their pitch ranges, and the resulting mean phone- tograms are compared. The mean fundamental frequency was found to increase by between 0.2 and 0.6 semitones per dB equivalent sound level. No great differences were found between these subject groups, although the singers were found to vary their mean fundamental frequency more than the nonsingers. It is possible to explain the voice pitch changes as the passive results of the changes of subglottal pressure applied in order to vary sound level of phonation. Introduction Recordings of minimum and maximum phonatory sound level as func- tions of fundamental frequency are called phonetograms. These are fre- quently used in some voice clinics as a help to describe the voice function. The axes of a phonetogram are related to two important phonatory dimensions, namely voice fundamental frequency and loudness of phona- tion. These parameters reflect, in turn, basic aspects of the voice source that would be revealing to voice function: the pitch of phonation is determined by the vibration frequency of the vocal folds while the phonatory sound level reflects the maximum amplitude of the dif f eren- tiatecl transglottal airflow (Gauffin & Sundberg, 1980; Fant, 1979; Fant, Liljencrants & Tin, 1985). For these reasons, the phonetogram seems promising also in clinical work, especially as the recording time needed is comparatively short. * Dept. of Phoniatrics, ENT Clinic, Max> Ceneral EBspital. ** Dept . of Phoniatrics , Karolinska S juld~uset, Stockholm. ***Univ. of Southern California, Ins Anqeles, CAI IJSA. STL-QPSR 1/1987 It is a notorious observation that speakers who raise their loud- ness of phonation also raise their mean voice fundamental frequency. This suggests the existence of habitual pathways in the phonetogram. Such pathways may help understanding the phonetogram from a practical point of view, and are thus interesting to describe. It seems worth- while to find out how subjects vary their mean fundamental frequency and mean sound pressure level (SPL) when asked to speak at different degrees of vocal loudness. Speakers tend to increase loudness of speech with the ambient ~oise level (Lane & Tranel, 1971). With regard to vocal abuse, clinicians discourage speaking under noisy conditions; this implies that speaking loudly may be detrimental to vocal health. Coricert and opera singers, and stage actors, often use their voices at loudness and pitch levels that would render untrained voices hoarse; yet they do not normally suffer vocal damage. It is interesting to see whether the type of voice use, that speaking in noise induces, leads to a combination of mean SPL and mean fundamental frequency different from that chosen when loudness of phonation is voluntarily raised. The aim of the present investigation was to examine the vocal behavior of different groups of speakers with respect to mean fundamen- tal frequency and mean sound level and to relate these data to the subjects' phonetogram. The investigation combines data gathered for two different pur-- poses. One purpose was to explore singers' and nonsingers' phonatory reactions to various noise and auditory feedback conditions. The other was to describe an aspect of status of voice function in dysphonic patients. Experiment The effects of voluntarily changed phonatory loudness on level of phonation and voice fundamental frequency was analyzed in two groups of subjects: nine male singers and nine male nonsingers, all with no history of voice disorders. One singer had only vocal training whereas the other eight had performed professionally in opera and concert. Also, the effects of reading in noise was studied in these subjects; the subjects heard a loud noise in earphones and were asked to read a text loud enough to get heard through this noise. In addition, the effects of reading in noise was studied in ten male and ten female patients suffering from dysphonic phonastenia, i.e., a functional voice disorder. All the patients were on the waiting list for treatment at the Phoniatric Department, Malmij General Hospital. In the case of the healthy voices, the recordings were carried out in an anechoic room and the sound was picked up by a Sennlleiser (MD 211) microphone at a constant distance in front of the subject's mouth. For STL-QPSR 1/1987 the patients, a sound treated room (4*3*2.3 m) was used and the micro- phone was an AKG CMSE CElO electret. Voice fundamental frequency was determined from the signal recorded by an accelerometer, fastened to the subject's neck just below the thyroid cartilage; for the patients this microphone was complemented with an electroglottograph. The subject wore a calibrated headset (TDH 49P) with capsules tightly sealing off the ears from external sound. The sound level of the auditory feedback, obviously critical to the effect of the noise, was adjusted so that a vowel yielding a 90 dB SPL at 1 m distance generated a 105 dB SPL in the headset. This approximates the level at which the subjects would hear their own voices without the headset in an anechoic room. In the case of the patients Sennheiser (MD414) headset was used, which does not seal off the ears. In the headset noise was presented. In the case of the healthy voices, a white noise was used filtered at 2.2 kHz by an LP filter with a roll-off of 18 dB/octave, see Fig. 1. In the case of the patients, unfiltered white noise was used which was presented at 70 dB SPL. The sound picked up by the microphone was recorded on one track of a calibrated tape recorder, and on the second track a signal was record- ed for fundamental frequency measurement. MASKING NOISE 0.1 0.2 0.5 1 2 5 10 FREQUENCY (kHz1 Fig. 1. Ilong-term-average spectrum from a third-octave filterbank of the LP-filtered masking noise. All subjects read a discursive text material with and without the masking noise. In the noise-free condition, the voice patients were asked to read at a normal loudness only, while the healthy voices read two times (Trial I and 11) for at least 30 sec under each of three loudness conditions : ( 1 ) norma 1 conversational loudness, (2) a lotdness level appropriate for an audience, and (3) the loudest level pssible. Phonetograms were made for the healthy voices only, before reading the text. To provide data for phonetograms, the reconunencld procedure is STL-QPSR 1/1987 to have the subject phonate at certain, given pitches as loudly and as softly as possible (Schutte & Seidner, 1983). In this case, however, we used a slightly different procedure. The singers sang triads, while the nonsingers made pitch glides from highest to lowest pitch or vice versa. Analysis The signal for fundamental frequency analysis was fed to a funda- mental frequency extractor connected to a digital analyzer, displaying the fundamental frequency distributions in terms of a histogram. Also, the mean and the standard deviation of fundamental frequency were shown. Using these means, the fundamental frequency mean and standard deviation were determined for each sample. The program also displays fundamental frequency and sound level as function of time. Such graphs were used for analyzing SPL and funda- mental frequency in the pitch glides and triads that the subjects per- formed during the recording sessions. The resulting data were used for plotting phonetograms. As different voices have differing mean fundamental frequencies, the relevant measure is the change in each subject's mean fundamental frequency induced by the experimental conditions. Therefore, this change was determined for each sample, and expressed in semitones (henceforth st). In the case of the healthy voices, a computer program for long- term-average spectrum analysis was used for determining the equivalent sound level, For the voice patients L was determined by a B&K Leq* eq 2218 sound level meter. Results The results from the two trials of the healthy voices did not show any significant differences (a variance analysis is published elsewhere, ( Perkins, Ternstrom, Sundberg & Gramming, 1987). Fig. 2 shows the average for the various conditions in the healthy voices. The singers produced higher Leq than the nonsingers. It can be noted that the data point for the unmasked conditions adhere to a straight line. This implies that, on the average, the subjects inter- preted "loud" as situated on a logarithmic sound pressure scale midway between "normal" and "loudest". The increment was 9 and 7 dB for the singers and nonsingers, respectively. Under conditions of masking noise, the nonsingers raised their L while the singers reduced their eq somewhat as compared with the loudest reading. Leq Evidently, the mean fundamental frequency varies considerably be- tween subjects, reflecting morphological and habitual differences. Therefore, Fig. 3 shows, on a 1ogariWic scale, the fundamental fre- STL-QPSR 1/1987 - 43 - 50 I I I I NORMAL LOUD LOUDEST MASKED EXPERIMENTAL CONDITION Fig.