Dept. for Speech, Music and Hearing Quarterly Progress and Status Report Perception of singing Sundberg, J. journal: STL-QPSR volume: 20 number: 1 year: 1979 pages: 001-048 http://www.speech.kth.se/qpsr STL-QPSR 6/1979 r, I , -1 . I. MUSICAL ACOUSTICS .- . .'' , A. PERCEPTION UF*SINGING . ' %L J. Sundberg <j. , - ,. ... .T ., ., .. , - ' .: , .$:.. , ,i... .-, "' ,, . ....,.<.,:,, '.. .... .................: :. :..... , . , . ' .' '\ Abstract 9s .This article reviews investigations of the singing voice which ,. possess an interest from a perceptional point of view. Acoustic hnction, formant frequencies, phonation, pitch, and phrasing in , singing are discussed. It is found that singing differs from speech in a highly adequate manner, k. g. for the pufpose of increasing the loudness of the voice at a minimum cost as regards vocal ef- fort. The terminology used for different type's of voice timbre . seems to depend on the properties and use of. the voice organ rather than on specific acoustic signal characteris tics. - Pr- STL-QPSR f/i979 ? c 4. partials decrease monotonically with rising frequency. As a rule of thumb a given partial is 12 dB stronger than a partial located one oc- tave higher. However, for high degrees of vocal effort this source spectrum slopes less than 12 d~/octave. With soft phonation it is I steeper than 12 d~/octave. On the other hand, the voice source ;n+rr,. spectrum slope is generally not,,dependeet on which voiced sound is produced. :: br:,iec,:.ceii~9 !: -<; c,*>iriF cat i~::5~:- I VIXJ ko 31i30e .JTT The spectral differences between various voiced sounds arise when the sound from the voice source is transferred through the vocal tract, i. e. from the vocal folds to the lip bpeiliig. The reason for this is that the ability of the vocal tract to transfer sound is highly dependent on; the frequency of the sound being transferred, This ability culminates -,ti-- .lr ------ JAIJ- id LLW~JXS~ -2 at certain frequencies, called the formant frequencies. A& a 16nse- quence of this, those voice source partials which lie closest to the <- formant frequencies are radiated from the lip opening with greater amplitudes than the other, neighboring partials. Hence, the formant frequencies are manifested as peaks in the spectrum of the radiated - - :< sound. - The formant frequencies vary within rather wide limits in response to changing the position of the articulators, i. e. , lips, tongue body, tongue tip, lower jaw, velum, and larynx. We can change the two lowest formant frequencies two octaves or more by changing the posi- t. ' . , :A,' c. CJj, >, tion of the articulators 4 The frequencies of these two formants deter- mine the identity of most vowels. The higher formant frequencies can- 58 not be varied as much. They seem to be more relevant as personal ?r, -? :rii.r~uo3,133 of ls~r5ns 1.1 voice characteristics. Thus, properties of vowel sounds which are of great importance to the vowel identity can be described in a chart showing the frequencies of the two lowest formants, as is done in Fig. .-- 3' .-.. I-A- 1. Note that each vowel is represented by a amall area rather than *a " . STL-QPSR i/f '979 by a poizit in the chart. In other words, these formant frequencies may vary within certain limits without changing the identity of the vowel. This reflects the fact that a given vowel is normally observed to possess higher formant frequencies in a child or in a woman than in a male adult. The .. ..,i reason for such differences is differing vocal tract dimensions, as Grill I r I be shown later;cl-,:J a ill( - -2i In singing, more or less substantial deviations are observed from the vowel ranges shown in Fig. I-A-I. Indeed, a male opera singer may change the formant frequencies so much that they enter the area of dif - ferent vowel. For instance, in the vowel [il * as sung by a male opera .. singer the two lowest formant frequencies may be those of the vowel [ yJ according to Fig. I-A- 1. And in female high-pitched opera singing the formant frequencies may be totally different from those found in normal speech. Still we. tend to identify such vowels correctly. This shows that the frequencies of the two lowest formants do not determine the vowel iden- tity entirely. Next we will see how and why these deviations from normal ,adhau?) as:!,.:.!iq $2 (lr.qz&2 +s Z!$W~)V zL''.j;- .. speech are made in singing, - -.- - - - - -c 7.: s n~ol2;sd: ~i+trr'r,:lr iisw )I. eprij ,91~%i'f-A? 111. ~esonator~aspects ,cc +;; rlh: sfi? , rfq~.r*js jr,rj avt- (faJ $3.S*Gq:s? i . Formant frequencies ;- , 4 I rf?..'CC." 3 .'(i -4.4 n* x--jZ . \ . I. - 1 Soprano singer is required to sing at fundamental frequencies as high as f 000 or 1400 Hz. In normal female speech the fundamental £re- 02 ,. i .I;;.i.rrsa: OR ir , * quency rarely exceeds ca 350: Hz. The normal value of the first (and in . _.- some vowels even the second) formant frequency is far below 1000 Hz, as can be seen in Fig. I-A- I. If the soprano were to use the same articula- arf ..>'G tion in singing a high pitched tone as in normal speech, the situation illust- rated in the upper part of Fig. I-A-2 would occur. The lowest partial in the spectrum, i. e. the fundamental, would appear at a frequency far be- . ,, 3 .,, low that of the first formant. In other words, the vocal tract ability * All letters appearing within C J are symbols in the International Phonetic Alphabet. ~a 4 STL-QPSR f/i979 to transfer sound would culminate at a frequency where there is no sound to transfer. It seems that singers tend to avoid this situation. Instead they abandon the formant frequencies of normal speech and move the first f~rmantfrequency to a close neighbothood of the fundamental. The main articulatory gesture used to achieve this tuning of the first formant i~ P change of the jaw opening, which $5 particularly effective for changing the first forrnant frequency (cf, Lindblom, & Swdberg. 1971). This explains why female singers tend to change their jaw opening in a pitch dependent manner rather than in a vowel dependent manner, as in normal speech. The acoustic result of this is illustrated in the lower part of.the same Fig. I-A-2: the amplitude of the fundamental and hence ,>. I. ,the sound power of the vowel increases considerirbly. Note that this gain , 1 in sound pover results from a resonatory phenomenon. It is obtained 1 ', ", 9 I Fig. 1-A-3 shows formant frequencies measured"in a soprano singing 1 various vowels at varying pitches (Sundberg 1975). As can be seen in . the figure, the vowels maintain their formant frequencies cf normal speech up to that pitch where the fundamental coincides with the fir'st formant. Above that fxequency the first formant is raised to a frequen- cy close to the fundamental. If the jaw opening is changed, the main t IX'i $* :.3* ' ' " but effect is observed in the first formant frequency, the higher for- , LT ', , mant frequencies a1 so change to some extent. This is illustrated in Fig. I-A-3: all formant frequencies change when the first formant starts (3 IL 3.~3.lfIk;,bY +<*+ ' 'I to match the fundamental frequency. >,& .,: . !- /,.-: .,yz7. -.:l-.JA:113 5 $ (L 2. Sound intensity and masking , , ' rf~ids yn.$rrlt r- rr - As was mentioned above, the amplitude of the fundamental increases '< when the first formant is tuned to that frequency. *This results in a gain in overall sound pressure level (SPL). The magnitude of the gain can .. .I tlL i,, PARTIALS FREQUENCY FORMANTS W D 3 C- %$ PARTIALS -*FREQUENCY Fig. I-A-2. Schematical illustration of the formant strategy in female singing at high pitches. In the upper case the singer has Fig. I-A-I. Ranges of the two lowest formant fre- a small jaw opening. The first formant appears at a fre- quencies for different vowels repre- quency far below the frequency of the lowest partial of the sented by their symbols in the interna- vowel spectrum. The result is a low amplitude of that tional phonetic alphabet (IPA) . Above partial. In the lower case the jaw opening is widened so the scale of the first formant frequen- that the first formant matches the frequency of the funda- cy is translated into musical notation. mental. The result is a considerable gain in amplitude of that partial. Reprinted from Sundberg (1977). STL-QPSR 1/1979 be seen in Fig.'P-A-4 which shows the increase in SPL associated with the formant frequencies plotted in Fig. I-A-3. We can see that the pitch de- pendent'choice of formant frequencies results in an amplitude gain of almost 30 dB in extre-me cases. This. cor.responds to a thousandfold in- crease of sound power. A perceptually important conclusion is that the female singer will gain in loudness to a corresponding extent. i.0 ,->: The singersJ1need2for exceptionally high degrees of loudness is of 'coujrse a consequence' of the fact that opera singers are generally accom- I.' panied by an orchestra The average SPL of an orchestr&.playing loudly I I ' in a concert hall is on the order of mangitude of 90 to 100 dB. This is , much mForethan we can expect from .a human speaker. The masking ef- fect which the orchestral sound will exert on a-singer' s voice is deter- mined byJthe'distxibution of sound energy along tfrg frequency scale.