IPA Pronunciation
May 6, 2014 1:59 AM Subscribe
Is there an online tool that will take an IPA pronunciation and speak it to me?
I've never heard of such a tool and I don't expect that one exists yet. There are some huge problems in the way of creating one.
If you're looking for information on how to pronounce real words, then you should start looking at language-specific resources: dictionaries with audio files, forvo, google translate, etc...
posted by Kutsuwamushi at 3:41 AM on May 6, 2014
If you're looking for information on how to pronounce real words, then you should start looking at language-specific resources: dictionaries with audio files, forvo, google translate, etc...
posted by Kutsuwamushi at 3:41 AM on May 6, 2014
IPA transcriptions are usually phonemic rather than phonetic. So this could work for an individual dictionary in a single language, but for an IPA reading app to work across multiple sites with transcriptions of words and phrases in multiple languages would require sites to have something like custom style sheets for phonemic IPA.
A naive IPA reader with no knowledge of how to realize phonemes in a particular language would produce really heavily accented pronunciations. For instance, treating wiktonary's transcription of the second American English pronunciation of "portmanteau" as phonetic, a naive reader would come out with something sounding kind of like "borrrt ma'n DOH oo" with completely flat intonation. (The sounds would be in there, but it wouldn't sound much like English.)
posted by nangar at 6:26 AM on May 6, 2014
A naive IPA reader with no knowledge of how to realize phonemes in a particular language would produce really heavily accented pronunciations. For instance, treating wiktonary's transcription of the second American English pronunciation of "portmanteau" as phonetic, a naive reader would come out with something sounding kind of like "borrrt ma'n DOH oo" with completely flat intonation. (The sounds would be in there, but it wouldn't sound much like English.)
posted by nangar at 6:26 AM on May 6, 2014
I've used AT&T Labs' Natural Voices Demo for creating small audio samples. It's not designed for this sort of use so you don't get a lot of control over things, but you can fudge it a bit here and there by choosing different voices or clevering up some workarounds, depending on what you're after.
posted by iamkimiam at 9:43 AM on May 6, 2014
posted by iamkimiam at 9:43 AM on May 6, 2014
Best answer: IPA transcriptions are usually phonemic rather than phonetic.
This is not true. The IPA is the International Phonetic Alphabet. It is constructed in such a way as to account for every sound made in any language, with diacritics to dictate all variations and details. It is often phonemic, as phonemes are usually distinguished by sound variation. But this is not always the case, as with allophones. Consider the tap vs. the voiceless alveolar plosive (/t/) - phonemically these are usually identical, but always phonetically distinguished.
I'm not sure there exists a dictionary of all languages that has IPA pronunciation recordings. However, this tool has all of the various phones on recording. Probably the easiest solution however would be to just learn the IPA.
posted by Lutoslawski at 12:05 PM on May 6, 2014 [2 favorites]
This is not true. The IPA is the International Phonetic Alphabet. It is constructed in such a way as to account for every sound made in any language, with diacritics to dictate all variations and details. It is often phonemic, as phonemes are usually distinguished by sound variation. But this is not always the case, as with allophones. Consider the tap vs. the voiceless alveolar plosive (/t/) - phonemically these are usually identical, but always phonetically distinguished.
I'm not sure there exists a dictionary of all languages that has IPA pronunciation recordings. However, this tool has all of the various phones on recording. Probably the easiest solution however would be to just learn the IPA.
posted by Lutoslawski at 12:05 PM on May 6, 2014 [2 favorites]
Best answer: IPA transcriptions can be either phonetic or phonemic. Dictionaries usually use phonemic transcription because the phonetic details are predictable from the context as long as you speak that language. Some include a few phonetic details, but I've never seen one with extremely detailed phonetic transcription.
But even if you had that, the most detailed IPA transcription still can't capture every detail of pronunciation. It was designed only to capture potentially contrastive features - that is, features that in some language somewhere make the difference between phonemes. It's simply not true that you can use the IPA to transcribe "all variations and details." There are systematic and unsystematic variations in pronunciation that can't be transcribed with the IPA.
You can get close enough that you would be understood though, even if you sounded a little funny.
The real issue with creating a synthesizer that takes an IPA string and pronounces it with you is that segments are an abstraction (and some would claim a fiction, but I wouldn't go that far). The speech stream is continuous, not segmented. There are incredibly complicated contextual effects - coarticulation, etc - that make it such that we actually can't identify acoustic invariants for most speech sounds. We still don't have a complete model of these effects.
Source: Phonetics/Phonology grad student
posted by Kutsuwamushi at 11:18 AM on May 10, 2014 [1 favorite]
But even if you had that, the most detailed IPA transcription still can't capture every detail of pronunciation. It was designed only to capture potentially contrastive features - that is, features that in some language somewhere make the difference between phonemes. It's simply not true that you can use the IPA to transcribe "all variations and details." There are systematic and unsystematic variations in pronunciation that can't be transcribed with the IPA.
You can get close enough that you would be understood though, even if you sounded a little funny.
The real issue with creating a synthesizer that takes an IPA string and pronounces it with you is that segments are an abstraction (and some would claim a fiction, but I wouldn't go that far). The speech stream is continuous, not segmented. There are incredibly complicated contextual effects - coarticulation, etc - that make it such that we actually can't identify acoustic invariants for most speech sounds. We still don't have a complete model of these effects.
Source: Phonetics/Phonology grad student
posted by Kutsuwamushi at 11:18 AM on May 10, 2014 [1 favorite]
« Older UK emergency travel document - any experiences? | Screen-printed t-shirts, a la Print Liberation? Newer »
This thread is closed to new comments.
posted by lollusc at 2:15 AM on May 6, 2014 [3 favorites]