After looking for something like this, I have found only IPA as a system that tries to represent human sounds with symbols; but there is an argument that IPA does not actually represents the sounds themselves, but instead it defines them in terms of how they are produced by human organs.

So I wonder: do we really have a standard phonetic alphabet that maps to the actual sounds that humans can produce and distinguish?


There seem to be two questions here, one about alphabet and the other about acoustics. The IPA as standardly conceived is an alphabet (it reduces continuous speech to segments), but the letters are defined in terms of articulatory and aerodynamic properties. You could in principle make the IPA into an acoustic alphabet by replacing the production-oriented definitions with acoustic ones. Right now it is possible to translate IPA letters into phonological features (SPE style), and there is more or less a stranslation from SPE articulatory features to the acoustic features of Jakobson, Fant and Halle, except for some reasonably well-understood differences (features like "mellow", "grave").

However, when you talk about "the actual sound human could produce and distinguish", we do not know what that is. For example, it is well-known that the quality of the vowel [i] as written in IPA actually varies across languages quite a lot. IPA does not purport to represent all of the sounds that people can produce and distinguish (even limiting sounds to language sounds). Instead, it purports to represent all of the sounds which can be "phonemic" in human languages.

There are a bunch of diacritics that can be used to fine-tune the "exact" sound as used in a real language, but there is no unambiguous way to denote the "k" of Kabardian as distinct from the "k" of various languages.

I might mention another way to derive acoustic definitions of IPA letters (a "do it" research project, not a "look it up" project). There are a number of expert recordings of IPA letters, vowels in a steady state and consonants in the environment pair "Ca aˈCa". These could be studied acoustically, for example to get the average formant trajectory of utterances exemplifying each letter. This is essentially researchifying the old-style way of learning IPA, where students would mimic the pronunciations of the instructor, who had (some degree of) expertise in the IPA. Samples from Ladefoged, Esling, House and Wells are available, and there may be other sources (there are Jones cardinal vowel recordings; and I just discovered that the Smalley recordings are on the web but they should be vetted for IPA-correctness – they don't have a consistent vowel in the CV context).

  • Right now it is possible to translate IPA letters into phonological features (SPE style) This is what I also wish that it should be defined as standard like IPA somewhere but cannot found it – Thaina Dec 12 '19 at 6:15
  • 1
    @Thaina That's more or less the case already. The IPA is organized in terms of articulation, but the value of each letter is defined in terms of broad categories (see IPA Principles #2). – Nardog Dec 13 '19 at 4:21

Short answer: no, no such thing exists.

Alternately: does a spectrogram qualify?

Long answer: there are quite a few problems with precise acoustic transcription.

In particular, spoken sound isn't divided into nice distinct units. When an English-speaker says [khæt], the actual acoustic reality is silence, then a tiny bit of turbulence, then a continuous vowel sound, then silence again. The [k] and the [t] don't have any acoustic existence in and of themselves—the only sign of them is how the [æ] gets warped at the beginning and end.

How would an "acoustic phonetic alphabet" deal with this in a useful way? If you record someone saying [æk] versus [æp], or [uk] versus [up], you'll see a different "warping" effect in all four of those, with none of them "warping" in quite the same way. But a human listener will process and interpret all those effects subconsciously without needing to think about it, and tell you that [æk] and [uk] have the same consonant, and [æp] and [up] have the same consonant, and those two consonants aren't the same.

And articulatory studies show that, indeed, there is something the same between [æk] and [uk], and between [æp] and [up], and that something is where the closure occurs in the vocal tract—the place where the airflow is cut off. So that's what the IPA represents, not an acoustic fact, but an articulatory one. Not because it's more accurate or anything like that, but because in this case, the articulatory facts are more useful for linguists than the acoustic ones. When precise acoustic details are needed, phoneticians tend to stick to spectrograms, which can represent all those details more completely and precisely than any alphabet could.

  • Thank you for your clarification, a very useful information. What I wish is that we should have symbol to represent something like similar spectrogram so it could encode and save as character in database – Thaina Dec 12 '19 at 5:28
  • 2
    @Thaina The question is, what would it accomplish? The human brain has a lot of machinery to decode that acoustic information and turn it into a sequence of nice, distinct, individual sounds. So pretty much anything in (non-phonetic) linguistics will involve working with those distinct sounds instead of the original sound waves. – Draconis Dec 12 '19 at 5:30
  • I think we might be able to round it into some nice discreet symbol. It should be able to distinguish that "this sound's consonant roughly between k and ɡ, and feel more like K, so let put it as K" for example. What I would like to accomplish is a universal voice recognition (I am a programmer). And I think that to make it as most efficient as possible for look up. We should be able to define any sound as phonetic. And to represent the sound from digital source, acoustic phonetic might be better than IPA – Thaina Dec 12 '19 at 6:22
  • 1
    @Thaina At that point, isn't that just the IPA? – Draconis Dec 12 '19 at 6:31
  • 1
    @Thaina You might be surprised! [kæ] and [ku] have a lot less in common, acoustically, than one would expect. The main similarities are in the articulation, not in the acoustics. This is why I don't know of any speech recognition or TTS system that works with individual phonemes. The smallest unit commonly used is the diphone. – Draconis Dec 13 '19 at 4:58

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.