I'm trying to understand some basics, hope this is an ok question.

Here's a recording of me producing /i/, /ɛ/, /a/: https://namakajiri.net/misc/i%c9%9ba.webm
nevermind the transition between the first 2 vowels.

I’m a Brazilian Portuguese speaker, but judging from articles I’ve just browsed, the typical formant values in my language aren’t much different than the ones for the closest English vowels.

Here’s the spectrogram:

praat spectrogram edit view

I think it’s easy to see F2 and F3 shifting at more or less plausible values (~2370, 2000, 1550Hz; ~3000, 2800, and a bit diffusely, 2300). But for F1 I have a persistent dark band at ~260Hz that loses a bit of intensity for /ɛ/, /a/ but otherwise doesn't budge at all. Maybe plausible for /i/ but way too low for the other vowels?

There’s a bit of energy above it that becomes a clear formant for the latter two vowels, at ~660Hz, which would be a plausible F1 for /ɛ/. But it doesn’t seem to move for /a/, and the bottom 270Hz band stays there.

Praat formant tracking on default settings more or less agree with this: enter image description here

Is my interpretation of this correct? If so, what is this low 260Hz band below F1?


This is H0, or, the fundamental. Looking at a spectral slice (at various positions) and comparing the computed F0, you can see that this happens at about 230 Hz. The prominence of the fundamental is a factor in individual voice quality, but your productions resemble breathy voiced vowels in languages that have a breathiness contrast (e.g. Dinka). The typical characteristic of breathy vowels is that most of the energy is in the fundamental. When you look at your waveform you will notice that it looks very sinusoidal. Dunno if you can do pirate voice to effect creakiness, but that should make things look very different.

LPC formant computation is not robust for breathy voice, but you can adjust the Praat settings to see if you can make them budge. I discovered that Praat just cannot get credible formant values for the Jones cardinal vowels recordings, because of the problem of energy distribution in the recordings.

  • whoa this is cool – if I try to change from creaky to "normal" in a live tracker one can see the exact moment that the detected F1 gets too weak and is masked. thank you for the solution, I'd never have guessed my voice is breathy – melissa_boiko Jan 8 at 18:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.