1. The Science of the Singing Voice
Overview of the course (HC16)
Winter 2008
Pat Keating, Linguistics, UCLA

2. Books
Johann Sundberg, The Science of the Singing Voice. Northern Illinois University Press (1989)
Peter Ladefoged, Elements of Acoustic Phonetics. Second edition. University of Chicago Pres (1996)
Richard Miller, The Structure of Singing: System and Art in Vocal Technique. Wadsworth Publishing (2001)
Richard Miller, National schools of singing: English, French, German, and Italian techniques of singing revisited. Scarecrow Press (2002)
Garyth Nair, Voice – Tradition and Technology: A State-of-the-Art Studio. With CD. Singular (1999)
Ingo Titze, Principles of Voice Production (2nd printing 2000)

3. 1. Intro: Sundberg’s demo
Go to: The ugly voice poster
But we don’t do any synthesis in the course

4. F0 and pitch Vibration, Hz Tuning forks, vocal folds
Relations of Hz to musical notes and intervals (several websites with these) – see next slide
Tone generator in Audacity is another way to relate Hz to notes

5. Frequencies of piano white keys

6. Digital audio SR, QR. compression
File formats – more complicated this year than in 2006!
We need song clips with a single voice (no instruments or other voices)

7. Review questions
1. Which tuning fork has the higher-sounding pitch, 392 Hz or 523 Hz?
2. What part of the body produces the fundamental frequency of the voice?
3. The frequency of the note G2 is 98 Hz. What is the frequency of G3?
4. Is a song on an audio CD or an mp3 player in .wav format?

8. Lab 1: audio clips Ripping CD tracks to .wav (CDex, CLICC)
Ripping audio from commercial DVDs (DVDFab Decrypter to AnyAudioConverter)
Ripping audio from YouTube videos (Freecorder)
Saving .mp3 and various other audio formats as .wav (Audacity, CDex, AudioConverter)
Splitting and saving mono tracks from stereo (Audacity)
File clips kept on our ecampus Discussion Board

9. Examples
From Worst of AI DVD:
From AI on Youtube:

10. 2. From Sundberg How do you experience your own voice?
Why does a recording of your voice sound different to you?
And, why do you sound better in the shower?

11. Pitch Semi-tone = about 6% freq difference
“in tune”: how close to target is close enough (about 20 cents for average listener)
“in tune”: steadiness
Transitions between notes: swooping
But we did some of this in the very first class – add what we did then

12. Example: steadiness, swooping

13. Vibratos Dimensions of vibrato Supposed good classical vibrato
Rate, range, amplitude vibrato
Supposed good classical vibrato
5.5 to 7 Hz rate, + .5 to 2 semitones range
What good a vibrato does, doesn’t do for the singer
Examples next slides

14. Example: D. Fischer-Dieskau

15. Example: Leontyne Price

16. Example: Joan Baez

17. Example: Kelly Clarkson

18. Lab 2 and Assn 1: vibratos Pitchworks, wavesurfer
Measuring F0 from pitchtrack
Calculating vibrato properties

19. Tricks in pitchtracking
Hardest part: keeping track of F0 range and optimizing option settings
Tuning forks and thin voices: don’t use cepstral method, use autocorrelation
Problems tracking trills and other fast F0 changes: need to change step size and/or window length

20. 3. Larynx and phonation
Laryngeal anatomy: physical model, “Vocal Parts” CD, ASA and Painter videotapes, Youtube videos, (DVDs about source and about phonation)
Mechanisms of vocal fold vibration
F0 variation with airflow means pitch and loudness are correlated, which singers need to learn to decouple

21. 4. Spectrum The voice source: F0 and overtones Line spectrum of source
FFT of output in Audacity, wavesurfer
DVD “Human Speech”: a key point of this is that speed of closing of vocal folds determines strength of higher harmonics and thus the brightness of the voice

22. Partials, overtones? Partials = harmonics
Overtones = partials above F0

23. Lab 3 and Assn 2: FFT FFT, LTAS in Pitchworks or wavesurfer
FFT in Audacity: View-Plot spectrum (nice for comparing effect of window length; shows musical note of F0)
Pros, cons of Audacity vs Pworks/wavesurf
Comparing spectra of different voice qualities by strength of H1, number of harmonics, extent of high-freq energy

24. 5. Resonances From Ladefoged on resonance Basic source-filter idea
More of Source-Filter DVD, on filter
Vowel “covering”: lowering the frequencies of front vowel resonances so that brightness is more matched across vowels

25. Singers formant
“Singers formant”: extra energy around 3000 Hz (Sundberg says Hz for basses, for tenors), which allows a solo voice to stand out against an orchestra, or other singers
Sopranos don’t much need a singers formant against an orchestra, because any note above about B4 will stand out by itself. Similarly for amplified singers.

26. Singers formant
Not an additional formant, but a clustering of F3, F4, F5; when they are close together in frequency their strengths are mutually enhanced and they give one broad strong spectral peak
Male singers: enlarge the ventricle (just above the larynx), lower the larynx
It is not known how altos (or sopranos, if they have one) produce their singers formant

27. Miller: singers formant

28. Example: Fischer-Dieskau (last vowel)

29. Speakers formant More like at 3500 Hz than 3000
Property of speaking voices judged to be good
Seen in some singing voices, especially in styles that are more like speaking (e.g. country)

30. Lab 4 and Assn 3: Singers formant
Looking at own voice and at recordings to see if there is a singers formant
trying to increase singers formant in own voice
Emphasized looking at /o/, /u/, where higher formants are expected to be weak so any enhancement will be unambiguous

31. 6. Vowel formants and F0
Average formant frequencies for different English vowels
a strong soprano voice matches F0 (H1) to F1, while a weak voice has no formant near F0
[Good illustration of this on DVD: the good voice and the bad voice samples]
Sundberg says that tuning F0 to F1 can add up to 30dB to the sound level
[other strategies in other ranges: Pavarotti’s tenor tuning of F1 to H2 in chest voice, F2 to H3 or H4 on high notes]

32. When F0 is above F1 F0 > F1 for many soprano notes
F1 cannot match F0, so H1 can’t be boosted by a resonance
vowel qualities are indistinct because F1 is not excited
trained singers tend to adjust the vowel quality so that the F1 moves up, in the direction of F0

33. F1 and F0
F1 is raised by opening the mouth more, or shortening the vocal tract (e.g. smiling)
YouTube videos of Queen of the Night aria singers and their mouth contortions on the high notes

34. Sundberg: F1 tuning when F0>F1

35. The soprano challenge
A few years ago a study of this effect, explicitly testing what Sundberg had said, got a lot of publicity:
They found that a trained soprano singing above about 440 Hz tuned every vowel’s F1 to the F0, where formants were determined by reflection

36. Dani and Shri at USC – MRI study of vocal tract adjustments that cause these formant shifts

37. Assn 4: F1 tuning
Happy Birthday when sung from F4 to F5: not a good match between F0s and F1s
Assignment was to write new lyrics that would give a better match to my vowel formants in this key
Full credit for nonsense, but a prize promised for best meaningful lyrics
Some wild-card vowels allowed where F0 was not near any F1 of mine

38. The winner Yay today yay hurray yay today yay is in
Today (na-me) is a-age
(A-a-age), spring chickin.

39. Lab 5: a total bust Tried to watch video en masse in CLICC
Had planned to make EGG recordings

40. Guest lecture
Gerry Berke from Head & Neck Surgery on their research on neuromuscular control of F0, on vocal pathology, and on care of the voice

41. 7. Consonants
2 chapters each in Miller, Nair, on different aspects of consonants in singing
Miller: oral agility for rapid consonant production

42. 7. Consonants Voiced vs. voiceless consonants
Effects of voiceless consonants on melodic line
Effect of C voicing on vowel F0
Lyricist’s choice of consonants already affects the song, independent of artist’s interpretation

43. Sondheim lyrics example
Bernadette Peters, Not a day goes by

44. Consonant “resonance”
Nair: More vs less sonorous (vowel-like) consonants (“consovowels”) as seen in the narrowband spectrogram
Consonant duration
Using consonant articulation artistically, e.g. for emotion

45. Example: lyrics + articulation
Bernadette Peters again, 2 clips

46. Example: lyrics + articulation
Melinda Doolittle vs. Gregorian chant

47. Lab 6 and Assn 5: consonants
Listening to, looking at, and making consonants in different ways

48. 8. Vocal warm-ups
Titze explains warm-up exercises in terms of bringing all systems up gradually
Acoustic loading for respiratory warm-up
increase the acoustic loading on the vocal folds with humming, trills, singing into a straw - lets the vocal folds vibrate with more abduction, and with overall lower Ps for an easy start
increase F0 so that Ps must increase
Fun with straws

49. 9. EGG
Ch. 13 in Nair (1999) = “The Use of the Electroglottograph in the Voice Studio” by D. Miller and H. K. Schutte
“one of the primary aims of training the classical singing voice will be to establish the habit of complete and abrupt closure, at least in mezzo forte and forte”
Seeing this in the EGG waveform

50. Falsetto vs chest voice on [i]: little contact in falsetto

51. Lab 7 and Assn 6: EGG
We made individual EGG recordings of students’ voices
Assn 6 on EGG analysis

52. Lab 7: webpages
The course requires a term project, which is presented as a webpage visible to the whole class
This year the webpages were by default on Googlepages (linked from, but not on, the ecampus site)
In-class instruction on using Googlepages by our ITC

53. 10. Aerodynamics
Normal breathing: about .5 liters 12 times/minute, with active inspiration and passive expiration.
Muscles of expansion: external intercostals, diaphragm
As in speech, in singing expiration is actively controlled, first by holding it back, then by increasing it
Muscles of contraction: internal intercostals, abs

54. Breathing in singing
Trained singers take much longer breaths, and more total air in a breath. More of the air in the lungs is exhaled by professional singers.
Trained singers have lower airflow rates in singing than do untrained singers, but the same airflow rates in speech.
Trained singers thus have more efficient phonation: they use less air to get strong vocal fold vibrations.

55. Sundberg: airflow vs. pitch
sound level (S), subglottal pressure (P) and oral airflow (A) from a professional singer’s ascending scale, showing that pressure increases a lot as pitch increases, even when airflow is fairly constant and sound level increases only somewhat

56. Air pressure in singing
Classically trained singers have lower subglottal pressures than do untrained singers, and these pressures are lower in speech as well as singing.
In singing, subglottal pressure is higher for louder phonation and for higher pitches: A doubling of subglottal pressure gives about a doubling in loudness, and subglottal pressure also about doubles when F0 doubles.

57. Sundberg: Ps vs. pitch
the clear relation of loudness, pressure and pitch in these quicker triads

58. The flow glottogram
Ug, from inverse filtering of Uo signal

59. 2 key aspects of the flow glottogram
the maximum amplitude of the flow is directly proportional to H1, the amplitude (in the source, not in the output) of the fundamental component
·    and this affects the perceived “strength” of the voice, though not necessarily its overall loudness, which instead depends on the strongest partial
the maximum closing rate is proportional to the amplitudes of the overtones

60. Breathy phonation
the glottis is somewhat abducted without complete closure
so some air flows through continuously, and the maximum flow is quite high
high airflow = a strong H1 in the source
High airflow also = high-frequency noise
Slower closing rate = lower-energy higher partials, which are then covered by noise

61. Pressed phonation The glottis is more adducted than normal
So stronger lung pressure is needed to get vibration
But the small and brief glottal opening means that little air flows through
Lower Ug means a weaker H1
Closing is usually more abrupt, so higher partials are stronger

62. Sundberg’s flow phonation
The sweet spot: the most abducted glottis that will still give complete closure
Most abducted, to give highest flow and thus strongest H1
H1 in flow phonation can be 15 dB or more greater than in pressed phonation
Complete closure, to reduce glottal noise and to strengthen higher partials

63. Loudness control with
a. phonation: the right amount of vocal fold adduction (Sundberg’s flow phonation)
b. the vocal tract: formant tuning, singers formant
c. lung pressure: higher pressure and higher airflow through the glottis. The power of the glottal source increases by 6 dB for every doubling of the lung pressure

64. Lab 8: aero Pressure and flow recording by each student
Did they show the relation of Ps, Uo, and F0 (with relatively fixed loudness) as in the Sundberg example figures?

65. Exam week: project presentations
During the scheduled exam period, students gave 5 minute overviews of their projects to the class, displaying their webpages, which were not due until the end of that day