Transposing the human voice

Question

I've been playing around with the pitch changer in Audacity. I noticed that when I change the key on instrumental music, while it may sound off, the timbre is mostly the same. In fact, it only sounds weird for pieces I've heard before since it conflicts with the key I remember; for new pieces, I can transpose it by half an octave and it will sound pretty normal. On the other hand, transposing a song with vocals by even 1 or 2 semitones will make a huge difference. Why is the human voice more sensitive to changes in pitch?

Another thing I've noticed is that in some cases, female voices modulated down a few semitones sound reasonably close to male voices and vice versa, but for other singers no amount of pitch change can make them sound like the other gender. For example, Taylor Swift (song: "You Belong With Me") down 3 semitones sounds somewhat like a tenor. On the other hand, Adele (song: "Someone Like You") down 1 semitone sounds like herself, down 2 semitones sounds in between male and female, and anything 3 or more sounds hideous. Train (song: "Hey, Soul Sister") up 3 semitones could plausibly be female. Elton John (song: "Can You Feel the Love Tonight?") still sounds male up to 3 semitones up, and anything more starts to sound chipmunk-like. What qualities in a singer's voice determine whether they sound like the other gender when transposed or not?

Answer 1

Vowels are formed using formants: the basic characteristic particularly of chest voice is a "pulse train" which has lots of harmonics/overtones. Those harmonics are then amplified or dampened depending on the shape of the mouth. The strongest surviving harmonics are called "formants". Basically, one hears the mouth shape under the "lighting" of the voice box, and it remains recognizable under different "lighting conditions".

A deep chest voice has lots of harmonics and provides a clear picture of the formants. Doubling its frequency doubles all the harmonics as well, painting a picture of a mouth or vocal tract half the original size.

Head voice has less pronounced harmonics, falsetto (and the somewhat equivalent female "flute voice") even less. Vowels become harder to distinguish, both because of the higher fundamental pitch (leading to a more spread out spectrum of harmonics) and because of less pronounced overtones to shape.

With practice and control of the resonances and mouth shapes, one can "gender shift" a lot without actually employing electronics. I can produce a distinctly female sounding singing voice solidly into the baritone range.

If you take a look at male travesty stars, they tend to speak with a flute-like voice in a low range and drop "out of character" into chest voice, usually without much of an actual pitch change, for comedic effect.

On rereading, I was missing sort of a conclusion: "resounding" voices in chest voice are the worst for pitch-shifting and the best for vowel recognition: they have good and distinguishable formants and paint a good picture of the relation of mouth size to speaking pitch. Mouth cavity size changes less between male/female than fundamental pitch. With singing voices, falsetto/head voice is less committed than resonant voices: "chanson-style" female singing is probably the female voice type with strongest chest voice components and will likely pitch-bend rather badly. If you know you are singing for pitch-shifting, you can artificially precompensate with small/big mouth shapes (speaking from a "small mouth" in the front of your mouth by shaping with tongue and vellum, or speaking in a "hot potato" style). The more distinctive your vowels are, the worse for pitch-shifting (not that having your mouth go all over the place when singing is a good idea, anyway, since directional effects amplify the effects of mouth shaping for the non-close listener).