English Language & Usage Stack Exchange is a question and answer site for linguists, etymologists, and serious English language enthusiasts. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Text-to-speech software are applications which try to generate a sound based on a textual input by following linguistic rules of a language (mainly phonetics and phonology). They make a sound for anything you enter, whether it's valid or not.

Because pronunciation of the names of people are really hard to be found over the Internet, I thought suddenly to use these software as a close hit. But I'm not sure if I can trust them or not. Do pronunciations of names follow linguistic rules by a good and acceptable percentage? Could this approach be useful?

share|improve this question

closed as off topic by Mitch, z7sg Ѫ, kiamlaluno, RegDwigнt Jul 28 '11 at 17:22

Questions on English Language & Usage Stack Exchange are expected to relate to English language and usage within the scope defined by the community. Consider editing the question or leaving comments for improvement if you believe the question can be reworded to fit within the scope. Read more about reopening questions here.If this question can be reworded to fit the rules in the help center, please edit the question.

Can you make this question more relevant to English. Right now your question is about evaluating speech production software, which is more of a concern for a software or technology forum. – Mitch Jul 28 '11 at 15:18
@Mitch, Actually I didn't know where to pose this question. While this is related to software, but the result is directly related to English, as I'm seeking a way to become able to pronounce English nouns. :). Now I think if anyone move this question to computer section, they also object to send it back here. – Saeed Neamati Jul 28 '11 at 15:45
That's not what I would call -directly- related to English. Anyway, some are answering, so maybe it's relevant anyway. My short answer: names (first and last) in English follow the pronunciation rules in English, but also have preserved archaic spellings (and changed pronunciations) much more than regular English. -And- most TTS software doesn't have a list special for all these names. So your approach might work but is not particularly trustworthy. – Mitch Jul 28 '11 at 16:51
Also note that as of today, software isn't very good at understanding context. Some words have different pronunciation depending on context (e.g. - the letter Z in my Answer), so that's just another reason not to rely on a software-based approach. – FumbleFingers Jul 28 '11 at 17:17

I don't want to get bogged down in evaluating text-to-speech software as such, but just to give an example which struck close to home for me...

Last year I put together a routine enabling my computer to speak Artist/TrackName when playing music. Initially I just used Microsoft's standard built-in "Mike" voice for the speech (it's an "American" voice). When I switched to an alternative "British" voice, I suddenly found ZZ Top had metamorphosed into the somewhat unexpected Zed Zed Top.

In short, don't put any faith in software. Even though in the above case I could "teach" (i.e. - "configure") the software to speak correctly, that certainly wasn't the default behaviour.

share|improve this answer
Good example @FumbleFingers, I'll consider that. – Saeed Neamati Jul 28 '11 at 15:47
Brits actually pronounce it zizitap, I gather? – msh210 Jan 1 '12 at 18:26
Britain has a wide range of dialectal variants, as I'm sure most countries do. But to be honest, if someone asked me to guess whether a person was British or American based purely on whether the way they pronounce "top" could reasonably be written "tap", I would be tempted to guess they're probably American. When I listen to James Cagney's last line in White Heat, it definitely sounds to me like "Made it Ma! Tap of the world!" – FumbleFingers Jan 1 '12 at 18:37
@FumbleFingers, sorry, mine was IPA. Brits actually pronounce it (in IPA) zizitap rather than zɛdzɛdtap? – msh210 Jan 1 '12 at 20:16
@msh210: Did you not know? The spoken name for the letter "z" in Br. Eng. is "zed", as opposed to American "zee". But obviously the name of the band must be what they call themselves, and they're American. So my "British" text-to-speech software had to be configured to make an exception to its default pronunciation. I don't recognise "a" as an IPA symbol for anything English - the IPA symbol that I know for the vowel in not, top, cough is "ɒ", so the band is called "ziː ziː tɒp" (or "ziː ziː tʰɒp", depending on whose symbol set/pronunciation you use). – FumbleFingers Jan 1 '12 at 22:13

Names tend to be the worst contenders for following general rules. So what speech recognition and tetx-to-speech systems potentially do is have a big database of exceptions. For some applications, e.g. automated telephone directory services, knowing the pronunciation even of fairly unusual surnames is important. (I did some work for a speech technology company where I and other linguists did precisely that: take a huge list of names and create a database of phonetic transcriptions. Needless to say, it's quite painstaking work.)

On the other hand, a general TTS system may either not include such a large database of names and/or not always be able to predict accurately when it needs to use that database. For most general TTS applications, the vast majority of words do not consist of unusual names, so including such a large database of exceptions for the sake of unusual names could actually increase the error rate overall for an average sentence.

Your best bet is: use the best-quality TTS system you can find (NOT the ones built into Windows which are completely dreadful) and type a phrase which obviously hints at the name in question actually being a name (e.g. "Hello, Mr _").Maybe compare with another high-quality system. And still take with a pinch of salt.

share|improve this answer

The answer is that you can't trust them. They do their best to interpret the word they're presented with but still go by a preschool level set of rules when presented with words they don't know.

They'd pronounce "one" as "oh-nay" if they didn't already have the correct pronunciation in their database.

share|improve this answer
I think it's fair to say that all "non-trivial" software speech synthesisers will use the equivalent of "phonetic representations" for at least some words that would otherwise be mispronounced. But virtually by definition these will mainly be common words that most of us wouldn't need to look up anyway, so your overall conclusion is still perfectly valid. – FumbleFingers Jul 28 '11 at 16:06

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