A little context: I'm a native speaker of American English. I use Google's text-to-speech engine with the British voice, as I find the roboticness (roboticity?) of delivery to be less distracting in a less familiar accent. However, I'm occasionally left scratching my head over particular pronunciations.
Some of these are clearly based on English language formation rules--for example, for some time the proper name "Ian" was said as just a sort of swallowed /n/ sound, likely because of a system rule about words ending in -ian (such as logician or Hessian; presumably there was some other rule governing guardian and the like).
However, two of these have me stumped. "M'" (as in m'lady or m'lord, but NOT m'self) is pronounced something like "MASHna" (/ˈmæʃ-nə/), and "hon" (the short form of "honey", as in "can I get you a warm-up on your coffee, hon?") sounds to me as if it is pronounced "cur" (/kər/). The US engine does not produce these pronunciations.
I've looked these up as abbreviations, both with a general google search and in the OED online, since the TTS engine frequently defaults to words as abbreviations (for example, the common name Ed always becomes editor, and the word "no" at the end of a sentence would be read as "number").
If this were the case, I would think "honourable" would be the obvious default British pronunciation of "hon", or possibly "honorary".
For M', something relating to machines or mass seems possible, but they don't seem like the first association (the OED has six other definitions for M preceding the mathematical symbol for mass, and does not include machine at all). These also don't explain where the second syllable comes from.
So, my questions:
- Is there an English rule, or a British usage, that explains the pronunciation of m' as /ˈmæʃ-nə/ and/or
- Is there an English rule, or a British usage, that explains the pronunciation of hon as /kər/?
I'm willing to accept that this might just be a bug in the engine (which would be out-of-scope for ELL), but most mispronunciations are clearly the product of over-regularization rather than pure error.
Clarification of why I believe this question is on-topic here:
I thought I was clear about this, but I'm NOT interested in the mechanics of how TTS engines work, nor am I assuming that the engine's pronunciations are "correct" for any variety of English. Rather, I am interested in what over-regularizations produced by machines relying on the rules given it by English speakers can tell us about English. (Similarly, linguists have looked at the over-regularizations produced by children learning to speak a language to identify underlying structures of the language.) I am also interested in the differences between American- and British-English usage, as occasionally (incidentally) revealed by listening to a TTS engine predicated on British English input.
Clarification 2: One more piece of information that may help understand why these examples, in particular, are of interest to me: Ordinarily, if the TTS doesn't "know" a word (most "common" words were apparently entered wholesale, rather than relying on "rules"), it will either sound it out using fairly basic rules—which in the case of "hon" I would expect to produce something like hot-ending-with-an-n or con-starting-with-an-h or similar—or else it simply spells out the word, in cases where it doesn't follow normal patterns of English spelling (common with loan words and uncommon names). In these two cases (out of thousands of unique words I've listened to), neither of these strategies was employed and I have been unable to identify a logical reason. For comparison, "'ave" as in "'ave a good day, guv'nor" is pronounced "avenue" by one TTS engine, and "ah-vay" as in "Ave Maria" in another; both of these examples are clearly comprehensible based on English usage, even if a native speaker would not make that mistake.
