I'm having a strange experience trying to reproduce a token of ɛ / æ from a native speaker of Thai. When I play my attempt back it sounds like a reasonable copy, but looking at it in Praat, the formants are quite a bit off. With some experimentation I can produce a vowel with similar formants, but it doesn't sound like the original token. This would seem to indicate that I'm latching onto something other than the formants in the original, that that quality is present in my initial attempt to copy the sound but not my later attempt to reproduce the formants, and that it's throwing my perception of the formants off... but what could it be?

Is there any research into the perception of foreign language sounds that might cast light on how the perception of vowels can be distorted by factors like (say) nasalisation or voice quality?

1 Answer 1


There is a lot of research on how some acoustic factor can influence perception of a sound: the challenge would be narrowing the choices down to something more specific. The situation that you face involves some specific limitations / complications such as the fact that you are perceiving a foreign language, and you are perceiving your productions of a foreign language. This paper may be a useful hook into research on perception issues in foreign language acquisition

The other challenge would be correctly characterizing the difference between the target vowel and what you (or any subject) is producing. For example, there was a study by Blodgett, Rytting and possibly others on Americans attempting to produce Vietnamese tones, where Americans or at lease the subjects sampled are bad enough that Vietnamese speakers have problems understanding productions. The reason turns out to be that Vietnamese tones are not just pitch profiles, they involve phonatory properties which American speakers were not getting. A sub-problem regarding the difference between target and speaker production is that the formant information from Praat may be wrong (the settings may be inappropriate).

  • Thanks. I'm reading the formants off the spectrogram so assuming F1 is going to be stronger than the 5th and 6th harmonics there shouldn't be a problem there. I think it must be something to do with voice quality because if I synthesize a vowel based on the formants from the original token and copy that, the formants match. The original is quite nasalized and there's hardly anything above F2 (1950), so maybe that's what's throwing me off. OTOH it could be that if you turn something into a non-linguistic sound by synthesizing it, there's less interference from L1 when you come to imitate it.
    – rchivers
    Commented Jun 27, 2020 at 21:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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