Here I am, @FumbleFingers.

:D

I don't actually synthesise speech using my own code (I'm not that good!). In my music player plugin, I just call "system-installed" voices. Usually they read the actual text (of artist/track name, etc.), but I do have support for overriding, say "ZZ Top" as spoken by a British voice (which has to be explicitly told to say "Zee Zee" instead of "Zed Zed").

room parent site changed to linguistics.stackexchange.com

but I could "pre-process" any IPA string, which in principle some users might find easier if I could map each possible IPA symbol (or combination) onto something that looks like a "word" in English.

@FumbleFingers By the way, you said it could be "Oh-key. Hi, Jay-go", but it couldn't be "Hi" in English, since he wrote "hi:". That is "hee".

I'm certainly not defending my interpretation! But my point is the software rendition isn't exactly "accurate" compared to actual speech, in any accent.

OK, here I am, too. The text was

ˌoʊkeɪˈhiːɹjəˌgoʊ

OK, so how do I boldface and italicize here?

Never mind.

@jlawler The same as in answers.

bold

There's no B or I key.

OK.

One asterisk for italics and two for bold.

So it's the same as in comments, not answers.

No also answers. We use Markdown everywhere.

Anyway, ˌoʊkeɪˈhiːɹjəˌgoʊ

symbol by symbol

Ok I'm reading. Please go on. :)

ˌ First the little secondary stress marker to mark the following vowel

o Next the mid back tense rounded vowel [o]

Ok.

ʊ Next the high back rounded diphthong offglide, allophonic (automatic) in American English. Phonemically [oʊ] would be simply /o/ because there's no non-diphthongized [o] for it to contrast with.

k Next the voiceless velar stop /k/.

You can keep writing, I'm here. Don't wait for my ok. :)

I'm not waiting for anything except careful typing. I don't need constant feedback, though thanks.

Ok, I had the feeling you were waiting for me.

To resume.

e Next the mid front tense unrounded vowel [e].

ɪˈ Next the high front unrounded diphthong offglide, also allophonic in American English.

In American English all tense vowels are redundantly diphthongized, leading to typical American accent problems in languages like French, Spanish, and German

*oʊkeɪ is now recognizable as the word OK or okay. Onward.

About "In American English all tense vowels are redundantly diphthongized"... Since you say AmE behaves like this, is it in contrast with BrE?

ˈ Next the primary stress marker (above the line for primary, below for secondary) to mark the next syllable as the main one.

There are differences between British and American English, yes. I don't know what you mean by "in contrast".

To continue.

As in "AmE does it, BrE doesn't".

I don't know enough about British English to say.

I'm a syntactician and semanticist primarily, not a sociolinguist.

And certainly not a phonetician.

Oh ok. :)

To resume.

h Next the voiceless vocal onset /h/, technically simply a voiceless version of the following vowel (in this case it's a voiceless [i] which would be represented by a little open circle under the "i"). This means that the phoneme /h/ has as many allophones as there are distinct vowels, which varies from dialect to dialect.

i Next the stressed high front tense unrounded vowel [i], as in bee.

ː Next a colon marking (redundant) length of the preceding vowel. Stressed vowels in American English are long before voiced and short before voiceless segments, and most segments (all resonants and all vowels, plus half the obstruents) are voiced. So it's common and automatic and predictable and ignored; i.e, allophonic.

ɹ Next a retroflex "rhotic" offglide, which is how I'd say here; I'm from northern Illinois, and that's the way we talk in, for example, Chicago. Note that there is only one syllable signalled here; many people would say here with one and a half or two syllables, as [ˈhijɹ] or [ˈhijɨɹ].

[ˈhi:ɹ] can now be recognized as here, btw. Think of this as 78K played at 33.

j Next a yod, Germanic (but not English) spelling for /y/.

ə Next a shwa, the most common vowel in English, phonemically a centralized reduced vowel with dozens of micro-allophones depending on speech rates and social factors. Shwa doesn't contrast with any other central vowels, taking up one-third of the perceptual vowel space on the phonemic chart.

OK. Break. I have to go do stuff.

Eheh ok, you're almost done though!

Hey @Manishearth

@Alenanno hey :)

I've been away a while occupied with my own "stuff I have to do", so I didn't realise this "element-by-element" breakdown was going on. It's all useful, and there are several points there I couldn't easily establish from the simple IPA charts I normally reference, but...

I was hoping for a detailed explanation as to why I can't find free software to read IPA reasonably intelligibly. Just because the market for such software isn't incredibly lucrative shouldn't mean no-one would actually produce it. I'll never get a penny for the dozens if not hundreds of hours I spent making my music player "talk". If a thing is reasonably useful, you'd expect someone to do it for nothing, in a whole world full of people with relevant technical skills.

The problem is not the money I think but the target-audience... How many people are going to actually use it? It's full of people who don't even know of the existence of IPA... :/

@Alenanno Yes, but I did my "talking player" project for the benefit of a single (blind) person. I only realised later that I might want to use it for myself. Having thought about it, I now realise it's seriously "non-trivial" for me to write a routine translating IPA symbols into "words" so an existing reader can speak them (there are no words for non-vowels, for example).

But since IPA designates component sounds (as opposed to whole words) surely there should be many people who can customise speech engines at a lower level to achieve this. I'm baffled that so few people have tried (or had much success, unless there's some reason why it's unexpectedly difficult).